Biology-aligned cervical cancer screening: target-centric biomarkers and next-generation diagnostic platforms

Abstract

Cervical cancer remains a leading cause of cancer-related mortality among women worldwide, with mortality disproportionately concentrated in low-and middle-income countries (LMICs), where screening infrastructure is limited. Although persistent high-risk human papillomavirus (hrHPV) infection drives nearly all cervical cancers through a prolonged preinvasive window, making the disease both preventable and detectable, current screening modalities face biological and operational limitations that constrain their global impact. Cytology-based methods suffer from moderate sensitivity and subjective interpretations. HPV DNA testing is highly sensitive, it lacks specificity for transforming infections and shifts the diagnostic burden toward triage and colposcopy. Meanwhile, visual inspection methods are accessible, they offer limited reproducibility. In this review, we adopt a target-centric diagnostic framework that organizes cervical cancer screening not by detection platform but by the biological class of the marker being interrogated. We first examined the molecular pathogenesis of cervical carcinogenesis, including HPV genotype-specific biology, viral integration dynamics, oncogene-driven transformation, and epigenetic consolidation, to establish the biological rationale for each biomarker category. We then systematically evaluated conventional screening modalities and their limitations before reviewing emerging diagnostic technologies across four target domains: HPV-derived markers (DNA, mRNA, capsid proteins), host cell-cycle regulators (p16^INK4a, Ki-67), non-protein biomarkers (biothiols, volatile metabolites), and marker-free platforms (Raman/FT-IR spectroscopy, AI-assisted cytopathology). Particular emphasis is placed on point-of-care (POC) deployability, noninvasive sampling strategies, and the integration of multilayered risk stratification into scalable screening pathways. By mapping diagnostic innovation onto the biological continuum of cervical carcinogenesis, this review provides a conceptual foundation for developing next-generation screening approaches that align molecular precision with global accessibility, thereby supporting progress toward the World Health Organization goal of cervical cancer elimination.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40364-026-00964-6.

Introduction

Cervical cancer is a malignant neoplasm that arises from the epithelial cells of the cervix, the lower portion of the uterus that connects to the vagina. The disease develops through a multistep process in which persistent infection with oncogenic high-risk human papillomavirus (hrHPV) induces precancerous lesions that can progress to invasive carcinoma over several years to decades [1]. Owing to this prolonged preclinical phase, cervical cancer is considered one of the most preventable and detectable malignancies through vaccination and screening [2]. Nevertheless, it remains a major global health burden, with an estimated 662,000 new cases and 349,000 deaths reported worldwide in 2022 [3, 4]. Survival is strongly stage-dependent, exceeding 90% when detected at the localized stage but dropping below 20% in metastatic disease, underscoring that early and accurate diagnosis represents the most critical opportunity to reduce mortality.

Despite the availability of preventive tools, the burden of cervical cancer is unevenly distributed across regions. Mortality rates are disproportionately higher in LMICs, largely reflecting differences in screening access, diagnostic infrastructure, and continuity of care rather than biological variation alone [4, 5]. In response, the World Health Organization (WHO) launched the “90–70–90” elimination initiative, within which effective and accessible diagnostic strategies are central to reducing global outcome disparities [6].

Conventional screening strategies, including cytology, HPV DNA testing, and colposcopy, have substantially reduced cervical cancer incidence and mortality in many high-resource settings; however, important limitations remain [7–9]. Current approaches often rely on detecting the presence of viruses or morphological abnormalities rather than the biological processes driving malignant transformation and may be constrained by infrastructure dependence, multistep clinical workflows, invasive sampling, and variable diagnostic sensitivity. These limitations contribute to persistent gaps in the early detection of this disease, particularly in resource-limited settings. Recent advances in molecular biology, biomarker discovery, and diagnostic technologies have begun to reshape the landscape of cervical cancer detection [10, 11]. Emerging efforts aim to overcome these limitations by enabling point-of-care (POC) testing, simplified and less invasive sampling, improved diagnostic accuracy through AI-assisted analysis, and earlier identification of biologically meaningful transformations. Parallel progress in the discovery of HPV-related and non-HPV host-derived biomarkers has expanded opportunities to refine molecular risk stratification and improve diagnostic performance beyond conventional screening paradigms.

Although several recent reviews have summarized cervical cancer diagnostic technologies, many have primarily focused on specific platforms or HPV detection alone [12–15]. In contrast, the present review adopts a target-centric diagnostic framework, organizing diagnostic strategies according to the biological classes of markers being interrogated, including HPV-derived, host-derived, and non-protein signals, rather than by diagnostic platform. This perspective highlights how the biological nature of diagnostic targets shapes technology design, clinical interpretation, and implementation, providing a more pathology-aligned and clinically interpretable view of the diagnostic landscape of cervical cancer.

Accordingly, this review examines cervical cancer diagnostics across the full prevention continuum, integrating epidemiological patterns, historical screening evolution, molecular pathogenesis, and emerging diagnostic technologies. Next-generation approaches are evaluated for their potential to enable decentralized testing, timely triage, and earlier detection. Figure 1 summarizes this conceptual framework, depicting the shift from morphology-based screening toward biology-aligned, biomarker-driven diagnostics.

Fig. 1.

Conceptual framework linking cervical carcinogenesis with next-generation diagnostic strategies. Conventional screening methods are limited by their low sensitivity, invasive sampling, and delayed detection. Emerging biomarker-based diagnostics, including viral, host genomic, epigenetic, and circulating signatures, aim to enable earlier and more specific detection while supporting non-invasive sampling and implementation in low-resource settings

Global burden and epidemiological disparities in cervical cancer

Cervical cancer exhibits some of the most pronounced geographic disparities among major cancers [16]. Mortality remains disproportionately high in regions where screening coverage, timely diagnosis, and access to treatment are limited, indicating that differences in population outcomes are driven primarily by the effectiveness of early detection and healthcare delivery, rather than intrinsic biological variation. These patterns underscore that, despite the biological preventability of cervical cancer, real-world outcomes are strongly shaped by health system capacity and access to diagnostic care.

As illustrated in Fig. 2, cervical cancer mortality exhibits striking geographic inequality, with the highest burden concentrated in regions with limited screening coverage and diagnostic infrastructure [3]. Absolute mortality also reflects the population scale, whereas the inverse relationship between the human development index (HDI) and mortality highlights the central role of healthcare system capacity in determining population outcomes. Together, these patterns demonstrate that the cervical cancer burden reflects disparities in screening access, diagnostic timeliness, and continuity of care. The “screening divide” extends beyond test availability to encompass triage capacity, confirmatory diagnosis, and access to treatment [17, 18]. Evidence from organized screening systems confirms that mortality reductions are greatest when screening is reliably linked to diagnostic confirmation and treatment [19].

Fig. 2.

Global burden and inequality in cervical cancer mortality [3]. (A) Global map of age-standardized mortality rates (ASR; per 100,000 females) in 2022, displayed using a low-to-high color gradient. Overlaid cyan circles indicate the absolute number of cervical cancer deaths by country, with circle size proportional to the total deaths. (B) Inverse association between the human development index and cervical cancer mortality. Boxplots show the distribution of country-level ASR across the HDI categories, with individual points representing countries. Dark red points denote group means, and the connecting line illustrates the declining mortality trend with an increasing HDI. Statistical significance was assessed using the Wilcoxon rank-sum test

Socioeconomic gradients in cervical cancer burden remain significant. In lower-resource settings (LRS), cervical cancer continues to be a leading cause of cancer-related mortality among women, whereas in higher-resource settings, it accounts for a comparatively smaller proportion of cancer deaths [16]. These disparities reflect differences in screening access, diagnostic infrastructure, and continuity of care rather than inherent biological factors—a pattern that persists even within high-income countries [17]. Cytology-based screening requires multiple visits, centralized infrastructure, and sustained quality assurance; when these are limited, delays and incomplete follow-up reduce program effectiveness. Scalable strategies supporting decentralized testing and streamlined decision-making are therefore needed to reduce loss to follow-up and strengthen the screening-to-treatment linkage [18].

By improving the accessibility and timeliness of detection, such strategies have the potential to narrow survival gaps and enhance population-level disease control. Looking ahead, demographic trends suggest that the absolute burden of cervical cancer will continue to grow without improvements in early detection and care delivery. These projections reinforce the need for diagnostic systems that are not only accurate but also accessible, scalable, and capable of supporting timely clinical intervention, underscoring the central role of improved diagnostics in advancing global cervical cancer control.

History and evolution of screening methods

The modern era of cervical cancer screening began with the introduction of cytological examination. In 1928, George Papanicolaou demonstrated that malignant and premalignant cervical cells could be identified through the microscopic evaluation of exfoliated cells from a vaginal smear [20]. This discovery, later validated in large clinical cohorts by Papanicolaou and Traut in 1941, established the conceptual foundation for population-based screening by enabling the detection of cervical precancerous lesions long before the development of invasive disease. The widespread implementation of cytology-based screening programs in the United States and other high-income countries has led to an approximately 70% reduction in cervical cancer mortality, firmly establishing the Pap test as one of the most successful cancer screening interventions in medical history [21].

Figure 3 illustrates the conceptual evolution of cervical cancer screening across four major eras: morphology-based detection, viral etiology discovery, molecular screening, and the emerging precision diagnostic era. In the morphological era (1940–1980s), screening relied primarily on the microscopic identification of abnormal cellular architecture. Although highly impactful at the population level, this approach is inherently dependent on morphological interpretation and laboratory infrastructure.

Fig. 3.

From morphology to molecular precision: transformation of cervical cancer diagnosis. This schematic timeline depicts the paradigm shift from cytological screening to HPV etiology, molecular biomarker testing, and contemporary precision diagnostics. Advances in viral oncogene biology, tumor suppressor signaling, and proliferation markers have laid the foundation for modern multi-omics, AI-enabled, and minimally invasive detection platforms that aim to improve early detection and risk stratification

A major paradigm shift occurred in the 1980s when persistent hrHPV infection was definitively identified as the causal agent of cervical cancer. In particular, HPV-16 and HPV-18 account for nearly 70% of cases worldwide, redefining cervical carcinogenesis as a virally driven, molecularly traceable disease process [22]. This discovery marked the transition into the viral etiology era and provided a biological foundation for molecular testing. In the 1990s, liquid-based cytology (LBC) platforms, such as ThinPrep and SurePath, were introduced to address the pre-analytic limitations of conventional smears, including inadequate cell preservation and sampling variability. These platforms improve specimen quality and laboratory workflow while maintaining the fundamental morphology-based principle of detection. Although LBC enhances standardization and sensitivity, screening at this stage remains primarily dependent on cytologic interpretation. The subsequent molecular screening era (2000–2010s) was characterized by the integration of HPV DNA testing into screening protocols [23, 24]. HPV testing improved the sensitivity for detecting high-risk infections and enabled more refined risk stratification approaches, including co-testing strategies and, eventually, primary HPV screening. The approval of the first prophylactic HPV vaccine in 2006 marked a major shift in primary prevention [2]. Subsequent guideline updates in 2012 formally incorporated HPV–cytology co-testing into routine screening algorithms for women aged > 30 years [6, 24]. Entering the 2020s, cervical cancer screening is transitioning into a precision diagnostic era. As illustrated in Fig. 3, this evolution reflects a progression from morphology-based detection to biologically anchored and multi-omics approaches. Contemporary strategies increasingly incorporate primary HPV testing, self-sampling, AI-assisted cytology, and emerging biomarkers, including host genomic and epigenetic alterations, methylation signatures, and circulating tumor DNA [25–27]. This shift represents a movement beyond the detection of infection alone toward the identification of biologically meaningful transformations.

Etiology and pathogenesis: from infection to malignancy

The epidemiological burden of cervical cancer is due to its distinctive and unusually tractable biological history. Persistent hrHPV infection is a necessary etiological driver in the vast majority of cases, although a small HPV-independent subset has been described [28]. Large-scale genomic analyses, including landmark studies by The Cancer Genome Atlas (TCGA) Research Network, have further clarified a key nuance: hrHPV infection is necessary but not sufficient for invasive cancer development [29]. Progression reflects virus–host interactions that reshape the host’s molecular landscape. Viral integration, genomic instability, and immune microenvironment remodeling are recurrent features along the path from infection to malignant transformation. These biological checkpoints define the mechanistic stages through which infection progresses to malignancy.

High-risk HPV: genotype classification and the adenocarcinoma screening gap

HPVs comprise a diverse group of double-stranded DNA viruses with compact genomes of approximately 8 kb in length. The genome is organized into early, late, and long control regions, which function as a regulatory hub for viral replication and transcription, notably through E1- and E2-mediated mechanisms of action [30]. As shown in Fig. 4A, the circular genome structure highlights the early genes (E1-E7), late capsid genes (L1-L2), and long control region (LCR), which collectively coordinate viral persistence and replication.

Fig. 4.

High-risk HPV biology and histological patterns. (A) Circular HPV genome (~ 8 kb) showing early (E1–E7), late (L1–L2), and long control regions (LCR). (B) Risk and phylogenetic classification of HPV genotypes. High-risk types cluster mainly within the alpha-9 (e.g., HPV-16) and alpha-7 (e.g., HPV-18) clades. (C) HPV-16-associated squamous cell carcinoma (SCC) arising in the cervical transformation zone, typically detected by cytology. (D) HPV-18-associated adenocarcinoma arising in the endocervical canal with lower cytologic detectability

HPV genotypes are broadly categorized into low- and high-risk groups according to their oncogenic potential [31]. Low-risk HPV types, such as HPV-6 and HPV-11, are primarily associated with benign epithelial lesions, including anogenital warts, and are not typically implicated in malignant transformation. In contrast, approximately 14 genotypes are classified as hrHPV based on consistent epidemiological and molecular evidence linking them to cervical carcinoma. This classification aligns with the International Agency for Research on Cancer (IARC) carcinogenicity framework, in which HPV-16 and HPV-18 are designated as group 1 carcinogens. The most clinically significant high-risk genotypes are clustered within alpha-9, including HPV-16, HPV-31, and HPV-33, and alpha-7, including HPV-18 and HPV-45 (Fig. 4B) [31]. Notably, the alpha-7 and alpha-9 types account for the majority of invasive cervical cancers. HPV-16 and HPV-18 are the most prominent types, responsible for approximately 70% of cervical cancers worldwide [32]. Persistent hrHPV infection is necessary for the development of most cervical cancers; however, the downstream trajectory from infection to invasive disease is shaped by genotype-specific viral biology, host factors, and the anatomical and cytological contexts in which precursor lesions arise.

In organized screening settings, declines in squamous cell carcinoma (SCC) incidence and mortality have generally been more pronounced than those in adenocarcinoma. Consequently, glandular cancers have tended to remain relatively stable over time, with a proportional increase reported in some populations [33]. Importantly, this proportional shift can occur even when the absolute adenocarcinoma rates are stable or declining more slowly than SCC, consistent with differential detectability and precursor capture rather than a uniform rise in glandular disease. This pattern is commonly explained by the inherent limitations of cytology. Pap testing reliably detects squamous abnormalities in the transformation zone, whereas glandular lesions higher in the endocervical canal are more difficult to sample and may show less distinct morphological changes, reducing sensitivity. HPV-16-driven carcinogenesis typically occurs in the transformation zone and is strongly associated with SCC, a histological subtype that is more readily detected by cytological screening (Fig. 4C) [34]. In contrast, HPV-18-associated carcinogenesis more frequently involves the glandular epithelium of the endocervical canal and is linked to adenocarcinoma, a subtype for which the sensitivity of Pap tests is lower (Fig. 4D) [35].

Genotype-specific biology intersects with these screening limitations in ways that influence lesion detectability, precursor presentation, and clinical behavior. Consistent with these anatomical patterns, HPV-16 is most frequently detected in high-grade squamous intraepithelial lesions, whereas HPV-18 shows relative enrichment in glandular and neuroendocrine histology [32]. In addition to phylogenetic classification, clade membership is associated with distinct biological behaviors. By contrast, HPV-18 and types such as HPV-45 are grouped within alpha-7, which has been reported to be more frequently observed in invasive cervical cancer and, in selected cohorts, associated with less favorable relapse-free survival relative to alpha-9 [36]. Consistent with clade-associated differences in host interaction and persistence, alpha-7 types are often reported to exhibit higher integration frequencies than alpha-9, although estimates vary depending on cohort composition and sequencing depth. In the TCGA cervical cancer analysis, HPV DNA integration was reported at a very high frequency in HPV-18–positive tumors and in a substantial fraction of HPV-16–positive tumors, and integration was associated with structural rearrangements and increased expression of nearby host genes [29].

These observations illustrate how genotype-specific viral biology, anatomical site of lesion development, and integration behavior converge to shape the histologic phenotype and progression dynamics. Therefore, the adenocarcinoma screening gap reflects not only sampling limitations but also underlying differences in viral–host interactions across HPV clades.

Mechanisms of viral persistence and cellular transformation: the integration dilemma

HPV infection is common; in contrast, progression to invasive cancer is comparatively rare. Population-based studies indicate that most infections clear spontaneously within 1–2 years of infection [37]. Only a minority of infections persist long-term, and persistence is consistently associated with substantially higher risks of high-grade intraepithelial neoplasia and cancer. Viral persistence is therefore a central determinant of malignant risk because it extends the timeframe for host genomic and epigenomic changes (Fig. 5) [30, 37].

Fig. 5.

HPV persistence and the integration dilemma in cervical carcinogenesis. Most HPV infections clear spontaneously, whereas a minority persist and progress to malignancy. Persistent infections sustain E6/E7 oncogene expression through episomal mechanisms or viral DNA integration. Integration may impair productive replication but confer a clonal growth advantage (“integration dilemma”). Sustained E6/E7 activity drives p53 suppression, pRb inactivation, genomic instability, and clonal selection, ultimately promoting the cellular transformation

Within this extended window, sustained or dysregulated expression of E6/E7 viral oncogenes can arise through multiple routes. Integration-associated disruption of viral regulatory control is a well-recognized pathway [38]. However, episomal mechanisms may also contribute to this process. For example, methylation at viral E2-binding sites within the LCR has been reported to attenuate E2-mediated repression and may support persistent oncogene expression, even when viral genomes remain episomal [39]. These alternative routes explain why oncogenic transcriptional states are not strictly dependent on the physical integration of the virus.

Viral DNA integration into the host genome is a recurrent event in many HPV-associated cancers and illustrates an “integration dilemma” [38, 40]. Integration often compromises productive viral replication; however, it can confer a selective growth advantage to the host cell and become clonally retained (Fig. 5). Genome-wide studies have reported that HPV–host junctions often show microhomology and are enriched in transcriptionally active, open chromatin regions [40]. Recurrent integration hotspots have also been described near super-enhancer–enriched loci, which is compatible with the involvement of DNA damage–repair processes in at least a subset of tumors. In cervical-derived cell line models, tandemly integrated HPV-16 has been shown to form viral–cellular super-enhancer–like elements enriched for H3K27ac and BRD4 [41]. These structures are associated with elevated E6/E7 expression, providing a mechanistic example of how integration could be clonally favored via enhanced oncogene output.

Recent multi-omics studies have further delineated functionally distinct forms of integration, namely, productive versus silent integration. Productive integration preferentially occurs in chromatin-accessible regions and produces viral-host fusion transcripts that enhance E6/E7 oncogene expression, thereby promoting clonal expansion [41]. In contrast, silent integration may lack oncogenic activity due to disruption by epigenetic mechanisms, such as DNA methylation. This conceptual framework transcends the simplistic binary classification of integration and helps explain the molecular heterogeneity of progression [42].

In addition to insertional disruption, integration can reshape the local and higher-order regulatory architecture. Integrative genomic studies have described several non-mutually exclusive oncogenic consequences [40]. These include functional loss of E2 with consequent derepression of E6/E7, disruption of host genes at or near the insertion site, and focal amplification with overexpression of nearby host genes. Sustained E6/E7 activity remains the central axis of transformation: E6 undermines p53-dependent apoptosis and DNA damage responses, whereas E7 disables pRb-mediated cell cycle control [43]. These primary alterations are frequently accompanied by broader transcriptional reprogramming toward proliferative and less differentiated cellular states, consolidating the transformation of the phenotype.

The pre-invasive window: progression kinetics from CIN to invasive carcinoma

A defining and clinically actionable feature of cervical carcinogenesis is its slow and stepwise course [44]. In many cases, the transition from persistent hrHPV infection to invasive cancer unfolds over a decade or longer, passing through histological precursor stages collectively termed cervical intraepithelial neoplasia (CIN) [45]. For clinical management, these lesions are commonly grouped into low-grade squamous intraepithelial lesions (LSIL/CIN1) and high-grade squamous intraepithelial lesions (HSIL/CIN2–3), reflecting meaningful differences in their short-term natural history and the likelihood of true precancer [45]. This extended preinvasive window provides a temporal substrate within which viral–host interactions accumulate and clonal selection occurs.

HPV integration is strongly correlated with histological progression. As detailed in Section “Mechanisms of viral persistence and cellular transformation: the integration dilemma”, integration acts as a molecular driver of malignant transformation. Although reported frequencies vary across studies depending on the methodology, a consistent pattern has been observed: integration events are infrequent in CIN1, increase in CIN2/3, and are prevalent in invasive carcinomas. Notably, genotype-specific tendencies exist; for example, HPV-18-positive lesions demonstrate higher integration frequencies than those associated with HPV-16. This gradient supports the view that integration is accompanied by increasing biological severity rather than representing a binary event. The biological heterogeneity of CIN underlies its variable clinical courses. LSIL generally represents a transient, productive HPV infection that regresses spontaneously within one to two years [45]. HSIL, in contrast, reflects a transforming state characterized by sustained oncogene expression, increasing genomic instability, and progressive epigenetic consolidation [43]. CIN2 remains particularly challenging because it encompasses a biologically heterogeneous spectrum; while some lesions regress, others exhibit molecular features consistent with clonal expansion and disease progression. Importantly, progression across the CIN continuum is not strictly linear. Regression, persistence, and advancement represent competing evolutionary trajectories shaped by viral persistence, integration architecture, host genomic context, and immune surveillance [38]. Therefore, the preinvasive window reflects a dynamic equilibrium rather than a deterministic sequence. Over time, selective pressures favor clones in which oncogenic signaling is stabilized through integration-associated remodeling and epigenetic reinforcement, culminating in invasive carcinoma.

Taken together, the CIN framework illustrates how temporal duration, integration frequency, genomic instability, and epigenetic consolidation converge during the transition from persistent infection to malignancy. Thus, the prolonged preinvasive phase is best understood as a period of layered molecular evolution, during which viral and host determinants cooperatively shape the trajectory toward either regression or irreversible transformation.

Molecular targets for screening and risk stratification

The biological trajectory outlined above—from persistent infection through viral integration to epigenetic consolidation—defines a series of measurable molecular states along the path to malignancy. The central challenge for screening is to translate this continuum into quantifiable signals that separate progressive disease from transient infection. We therefore next examine the molecular target classes that operationalize this biology for risk stratification. The clinical problem is not simply HPV detection but determining which HPV-positive individuals are likely to develop clinically meaningful precancer, typically defined as CIN2 +. The biomarker classes discussed below reflect successive biological layers of this transformation. These biomarker layers become clinically actionable through triage. Triage is the second step in the evaluation performed after a positive primary screening [46]. Its purpose is to decide the next action with greater precision: who should proceed to colposcopy or treatment and who can be followed safely with repeat testing. This step has become central to HPV-based screening programs because HPV positivity is common and frequently transient. Most infections resolve without any progression. Without effective triage, the universal referral of HPV-positive individuals for colposcopy would lead to excessive procedures, unnecessary biopsies, and potential overtreatment. Such inefficiencies have broader implications for health systems, particularly in settings where colposcopy capacity is limited or follow-up pathways are fragmented or lacking. Therefore, modern guidelines embed triage within risk-based management. They combine viral, cytologic, and biomarker signals to estimate the short-term risk of CIN3 + and invasive cancer [24, 47]. These estimated risks determine whether individuals return to routine screening, undergo intensified surveillance, or are referred for immediate diagnostic evaluation of the disease. Large-scale programmatic datasets have demonstrated that the 5-year risk of CIN3 + following a negative HPV test or negative co-test is approximately 0.1%, supporting extended screening intervals when high-quality primary testing and triage systems are implemented [47].

From a diagnostic perspective, viral assays capture oncogenic exposure or activity, host cell-cycle markers reflect downstream transformation, epigenetic and integration-related markers indicate durable molecular remodeling, and microenvironmental or metabolic signals may provide additional context for progression risk.

Viral genomic and transcriptomic markers

HPV DNA testing offers high sensitivity for CIN2 + and CIN3 + but is limited by specificity, because many HPV-positive infections are transient and do not progress to clinically significant disease. Molecular stratification of HPV-positive results is therefore central to effective triage. Partial genotyping represents the first layer of biological refinement. High-risk genotypes are not equivalent in terms of oncogenic potential. HPV-16 and HPV-18 account for a disproportionate share of invasive cancers and are prioritized in risk-based management algorithms, even when cytology results are negative [48]. Large risk-based datasets have demonstrated that HPV-18 positivity with negative cytology can carry an immediate cancer risk of approximately 0.2% and a 5-year cancer risk approaching 0.5–0.6%. Importantly, this elevated cancer risk may occur even when the estimated CIN3 + risk is not maximal. This genotype-specific discordance highlights a critical principle: cancer risk and precancer risk are related but not identical constructs, and management strategies must consider both.

Real-world cohort data also clarify why the combination of genotyping and cytology improves discrimination. In a large screening cohort, women with normal cytology but HPV-16 positivity had a CIN3 + risk of approximately 20% over approximately two years. For HPV-18, the risk was approximately 10%, and for other high-risk types, it was approximately 5% [49]. When HPV-16 infection persisted for 12 months in women with normal cytology, the CIN3 + risk increased to approximately 30% [50]. These gradients show that genotyping adds actionable separation, even among cytology-negative HPV-positive women. They justify treating HPV-positivity not as a single undifferentiated state but as a set of tiers defined by genotype and persistence. Building on genotyping, transcript-based assays, such as E6/E7 mRNA, further refine risk by capturing oncogene activity rather than viral presence alone.

E6/E7 mRNA assays add an activity-based triage layer by distinguishing transforming infections from transient viral detection. Across settings, these assays often maintain high sensitivity while improving specificity compared with HPV DNA testing alone [51, 52]. However, enhanced biological specificity does not automatically reduce referral burden, underscoring the need to balance sensitivity with downstream procedural capacity. Together, HPV DNA positivity, genotype stratification, and transcript-based assays form a layered framework that converts HPV detection from a binary infection marker into a biologically graded predictor of progression risk.

Host‑response and cell‑cycle regulators

Triage can also target host cell consequences of viral oncogene activity, which may be detectable even when morphology is subtle. p16/Ki-67 dual staining captures HPV-driven cell-cycle dysregulation by identifying the simultaneous expression of p16 and the proliferation marker Ki-67 [53]. Comparative evaluations have shown that adding a cell cycle marker improves accuracy relative to cytology alone [54, 55]. In one study of high-risk HPV-positive women, Pap cytology at an ASC-US threshold showed CIN3 + sensitivity in the low‑90% range and specificity of approximately 50%. In the same cohort, p16/Ki‑67 dual‑stain triage maintained similar sensitivity in the low‑90% range, while increasing specificity by more than 10% points into the low‑60% range [55]. When dual staining was applied to HPV‑positive women with low‑grade or normal cytology, the sensitivity for CIN3 + remained in the mid‑90% range, with specificity in the high‑50% range [56]. This profile fits the triage objective: it improves colposcopy efficiency without substantially increasing the number of missed high‑grade diseases. Longitudinal data also support risk separation after negative dual staining. In a cohort of HPV-positive women with normal cytology, dual staining detected more than 70% of underlying CIN3 + at baseline. The 5‑year cumulative incidence of CIN3 + after a negative dual stain was approximately 3–4%, compared with roughly 7% in women managed without dual‑stain triage. This divergence is precisely what triage aims to create: it defines a group that can be followed rather than immediately referred. These findings help explain why many programs consider cytology combined with a cell cycle marker to be more robust than cytology alone.

Epigenetic alteration and integration signatures

Epigenetic biomarkers provide a more stable molecular readout of transformation than viral detection. Although transient HPV infections may clear without consequence, host DNA methylation changes tend to accumulate with increasing lesion severity and reflect progressive regulatory remodeling. This stability makes methylation particularly attractive for large-scale screening programs, as assays can be standardized, automated, and are less dependent on subjective cytologic interpretation [57].

Clinical data support methylation as an effective triage tool with tangible programmatic benefits. In a large study of HPV-positive women with ASC-US or LSIL cytology, a FAM19A4/miR124‑2 methylation test showed CIN3 sensitivity around the high‑70% range and specificity in the high‑70% range. The sensitivity for cervical cancer was in the mid‑90% range, and the negative predictive value (NPV) for CIN3 approached 97% [58]. Importantly, in the low-grade cytology group, methylation-based triage reduced colposcopy referrals by approximately two-thirds. This illustrates that the value of triage lies not only in diagnostic performance but also in its impact on care pathways and resource utilization. Combining methylation with cytology further refines risk discrimination. In a Dutch screening cohort, the integration of FAM19A4/miR124-2 methylation with cytology achieved CIN3 + sensitivity in the mid-80% range, with specificity in the high-60% range [57]. Compared with either modality alone, the combined approach achieved a more balanced trade-off between detection and referral burdens. This illustrates how molecular and morphological dimensions can be layered to optimize triage efficiency. Viral integration signatures provide a complementary diagnostic axis, but their value for triage depends on capturing functional consequences rather than detecting breakpoints alone. At present, integration-aware assays are more suitable for confirmatory or research applications than for routine population-level triage.

In practice, methylation and integration markers do not need to compete. Methylation reflects the epigenetic stabilization of dysregulated programs. Integration reflects structural rewiring and disrupts transcriptional regulation. A combined approach could plausibly improve the sensitivity and specificity for clinically meaningful CIN2 + beyond what either axis alone can achieve, particularly in programs aiming to reduce unnecessary procedures without missing high-risk disease.

Despite these promising findings, several limitations constrain the current clinical applicability of methylation-based triage. First, the evidence base remains concentrated on FAM19A4/miR124-2 as a panel, with limited prospective comparative data against other candidate methylation markers or established triage tools such as p16/Ki-67 dual staining in the same populations [59]. Second, existing validation cohorts are predominantly European, raising questions about the generalizability of methylation cutoffs and assay performance to populations with differing HPV genotype distributions and disease prevalence [59, 60]. Third, there is currently no regulatory-grade, standardized methylation assay with a universally validated threshold, meaning that assay-to-assay variability and laboratory-dependent interpretation remain unresolved barriers to wide-scale implementation [61]. Until these gaps are addressed through prospective, multicenter, population-diverse validation studies, methylation-based triage is best positioned as a complement to, rather than a replacement for, established triage modalities.

Microenvironmental and metabolic signatures

Transformation and progression are not confined to genomic changes in epithelial cells. They also involve immune surveillance, stromal interactions, and metabolic adaptation [62]. Although these dimensions are not yet part of standard triage algorithms in primary screening, they are increasingly measurable and may become relevant as multianalyte testing platforms evolve [63].

Immune evasion pathways are an example of this. PD‑L1 expression has been extensively studied in cervical cancer, and pooled clinical evidence has linked higher PD‑L1 expression to worse survival [64, 65]. This supports the idea that immune escape is not only therapeutically relevant, but also a measurable correlate of aggressive tumor biology. For triage, the most plausible near‑term roles are not in first‑line screening, but in refining the risk among high‑grade lesions that fail immune control and in supporting response monitoring as immunomodulatory strategies expand.

Metabolic signatures are an emerging axis. Hypoxia and glycolytic reprogramming are repeatedly associated with aggressive behavior in cervical neoplasia and cancer, and the overexpression of hypoxia markers, such as HIF‑1α, has been linked to adverse outcomes [66, 67]. Currently, these markers are more mature for prognostication and treatment response assessment than for primary screening and triage. Their role in triage remains exploratory, but they reinforce the broader framework that the highest-risk subset may be better captured by integrated readouts that include viral activity, host dysregulation, and contextual biology [63]. That said, the clinical translation of these markers remains constrained by several methodological limitations. Existing metabolomic studies in cervical cancer are predominantly small and cross-sectional in design, with limited histopathological correlation and no prospective validation in independent screening cohorts [68]. Inter-platform reproducibility across different sample types — serum, urine, and tissue — remains inconsistent, and metabolomic profiles are susceptible to confounding from dietary patterns, hormonal status, and comorbid metabolic conditions, reducing diagnostic specificity [69]. As a result, no metabolic signature has been validated to a standard sufficient for clinical triage application, and their inclusion in screening frameworks will require large-scale, confounder-controlled prospective studies.

Overall, the logic of triage is the consistency between biology and workflow. Viral testing can identify oncogenic exposure. Genotyping and viral transcription refine transforming activity. Cytology and cell cycle markers capture functional dysregulation. Methylation and structural signatures reflect durable remodeling. Microenvironmental and metabolic features may help define the highest-risk subset among persistent or ambiguous cases. The strongest screening pathways are those that integrate these layers to match the intensity of follow-up to quantified risk, preserving cancer prevention while reducing avoidable procedures and improving care completion. Collectively, these biomarker classes represent distinct but complementary biological layers of progression risk assessment. As illustrated in Fig. 6, molecular stratification extends from viral detection to host cell dysregulation, epigenetic remodeling, and contextual microenvironmental signals.

Fig. 6.

Layered molecular framework for risk stratification in HPV-based cervical cancer screening. Persistent hrHPV infection initiates stepwise progression from cellular abnormalities to invasive cancer. Molecular triage integrates viral markers (HPV DNA, genotyping, and E6/E7 mRNA), host cell cycle indicators (p16/Ki-67), epigenetic and integration signatures, and emerging microenvironmental or metabolic features. Together, these layers refine progression risk and inform risk-based management decisions, enabling tailored follow-up intensity while minimizing unnecessary procedures

Importantly, these molecular layers are not independent signals but interrelated biological processes that co-evolve during cervical carcinogenesis. Combining viral transcriptomic activity, host epigenomic remodeling, and microenvironmental signals offers a more comprehensive picture of the transforming state than any single biomarker class alone. For instance, concurrent assessment of E6/E7 mRNA and host promoter methylation — reflecting both active oncogenic transcription and durable epigenetic consolidation — yields stronger predictive value for CIN3 + risk than either marker independently, while immune or microenvironmental readouts can further delineate high-risk subsets within equivocal triage categories [70].

From a clinical perspective, multi-omics integration is most practically operationalized through a tiered framework rather than simultaneous universal testing [71]. Primary HPV DNA testing establishes oncogenic exposure, genotyping and mRNA assays define transforming activity, methylation profiling confirms epigenetic stabilization, and microenvironmental or metabolic markers resolve ambiguous cases. Key translational challenges remain in standardizing multi-analyte interpretation and developing validated risk algorithms; however, advances in multiplexed and lab-on-chip platforms support the technical feasibility of this approach. Such a tiered strategy offers a biologically coherent path to maximize diagnostic precision while minimizing procedural and infrastructure burdens across diverse screening settings.

Conventional screening methods

The molecular target classes described above represent what an ideal screening program would measure to capture transforming disease. In current practice, however, population screening still rests on a small set of established modalities that interrogate only part of this biological continuum. Before considering how emerging platforms address the remaining gaps, we first examine these conventional methods and the biological and operational constraints that define the requirements next-generation diagnostics must meet. Conventional cervical cancer screening platforms form the backbone of current prevention programs and have enabled substantial reductions in incidence and mortality rates. Cytology, HPV molecular testing, and colposcopy address the complementary layers of the screening continuum, which span primary detection, risk stratification, and histological confirmation (Fig. 7). However, their population-level impact depends not only on analytical performance but also on effective implementation within care pathways that ensure follow-up and treatment. Despite their proven value, each modality has inherent biological and operational limitations. Morphology-based approaches are subject to interpretative variability and have a moderate sensitivity. Although HPV testing is highly sensitive, it increases the downstream triage demand. Colposcopy is resource intensive and operator dependent. Although visual inspection methods are accessible, they have limited specificity and reproducibility. Collectively, these tools function as an integrated yet imperfect system in which gains in one component may shift bottlenecks elsewhere. Accordingly, understanding both the strengths and limitations of conventional screening platforms is essential for interpreting the current program performance and defining the requirements that next-generation diagnostics must fulfill.

Fig. 7.

Current cervical cancer screening approaches. Following cervical sample collection, screening strategies include cytology-based methods, HPV molecular testing, and clinical evaluation, such as VIA and colposcopy. These complementary approaches enable population screening and lesion assessment; however, they are limited by moderate sensitivity, technical complexity, and operator dependence

Cytology (Pap smear and liquid-based cytology)

For more than half a century, cytology-based screening using the Pap test has been the core tool for the secondary prevention of cervical cancer. Cytology identifies the morphological features of squamous intraepithelial lesions and invasive carcinoma by evaluating exfoliated epithelial cells collected from the transformation zone using standardized reporting frameworks, such as the Bethesda System [72]. When embedded in organized screening programs with systematic recall and follow-up, cytology is associated with sustained reductions in cervical cancer incidence and mortality. Despite its historic success, cytology is intrinsically constrained by biological and operational factors. In conventional Pap smears, the collected cells are directly transferred to a glass slide and must be promptly fixed to preserve nuclear details and minimize air-drying artifacts. This process is vulnerable to uneven smearing, low cellularity, and obscuring elements, such as blood, mucus, inflammation, or lubricant, all of which can compromise specimen adequacy and interpretability. These specimen-level sources of noise, combined with the subjective nature of morphologic interpretation, particularly in borderline categories, inevitably introduce inter-observer variability, a limitation documented in large multicenter studies, such as the A To Z Laryngeal and Thyroid Screening (ALTS) [73]. Accordingly, the sensitivity of cytology for CIN2 + is moderate and variable, and its overall performance remains highly dependent on sustained quality assurance systems and trained workforce capacity.

To address these pre-analytic and operational limitations while preserving the morphology-driven nature of the diagnosis, LBC has been introduced. Instead of direct smearing, the sampling device is rinsed into a preservative vial, and the laboratory processes the specimen to generate a thin, more uniform cellular layer with reduced background debris. This workflow decouples sample collection from slide preparation, improves preparation consistency, and facilitates partial automation or computer-assisted reading in organized screening programs in the future [74]. From a clinical performance standpoint, LBC has generally shown sensitivity and specificity comparable to those of conventional cytology in detecting clinically relevant precancer endpoints. In the NTCC randomized trial, LBC did not significantly increase the sensitivity for CIN2 + compared with conventional cytology; however, it reduced the proportion of unsatisfactory smears and increased the detection of low-grade abnormalities, illustrating a trade-off between improved specimen adequacy and potentially increased downstream diagnostic work-up [75]. Similarly, a large population-based Dutch randomized trial did not demonstrate a clearly superior relative sensitivity or positive predictive value for precancer detection with LBC, supporting the interpretation that the principal advantage of LBC lies in workflow standardization and specimen quality rather than a fundamental gain in intrinsic diagnostic accuracy [76]. Slide-reading strategies also influence cytologic yield. Imager-assisted and computer-supported workflows have been explored to improve throughput and reduce oversight, and program-level studies have indicated that reading protocols can modulate detection outcomes under routine referral threshold [77].

In high-income settings, cytology-based screening, particularly when implemented within organized programs, has contributed to substantial and sustained reductions in cervical cancer incidence and mortality. Nevertheless, from a biological and operational perspective, cytology remains a pattern recognition-based screening tool with only moderate and variable sensitivity, unavoidable inter-observer variability, and a reliance on relatively short screening intervals to maintain population-level protection. Although LBC successfully mitigates several pre-analytic and technical weaknesses of conventional smears, it does not alter the fundamental reliance of cytology on subjective morphologic interpretation. These intrinsic constraints explain why cytology, despite its historical importance, cannot serve as a long-interval, risk-anchored foundation for screening in the molecular era and has increasingly been repositioned toward a triage or adjunctive role in HPV-based screening frameworks.

HPV DNA/mRNA

HPV DNA/mRNA testing provides the molecular anchor of contemporary cervical cancer screening. Most clinically implemented HPV DNA assays detect carcinogenic HPV using nucleic acid or signal amplification chemistries, enabling high analytical sensitivity in exfoliated cervical samples. Several such systems have received regulatory approval and are widely deployed in organized screening programs, including Onclarity (Becton, Dickinson, and Company) and Cobas HPV (Roche Molecular Systems). In addition to DNA detection, assays targeting E6/E7 mRNA provide information on viral oncogenic activity, offering improved biological specificity for identifying transforming infections.

Primary HPV DNA testing improves sensitivity for high-grade precancers but shifts the operational bottleneck downstream, as more screen-positive individuals require triage and colposcopy. Triage efficiency is therefore a dominant determinant of program performance. The Dutch transition to primary hrHPV screening illustrated this tension: increased colposcopy referrals for ≤CIN1 prompted evaluation of alternative triage strategies to reduce low-yield procedures while maintaining safety [78]. In settings with limited colposcopy capacity, high primary-test sensitivity must therefore be paired with effective triage to preserve program impact [79].

Visual inspection with acetic acid (VIA)

Visual inspection with acetic acid (VIA) is a low-cost, accessible screening method that is widely implemented in LRS, where laboratory infrastructure is limited. Following the application of dilute acetic acid (3–5%), areas of increased nuclear density appear as acetowhite lesions, allowing immediate visual identification of suspicious regions and enabling same-visit screen-and-treat strategies. The principal advantages of VIA are its accessibility, minimal infrastructure requirements, and rapid clinical decision-making. It can be performed by trained non-physician providers and does not require laboratory processing. However, VIA is highly operator-dependent and exhibits variable sensitivity and specificity across different settings. False-positive rates may be substantial, leading to potential overtreatment, and subtle or endocervical lesions may be missed. Reproducibility and quality assurance remain challenging. Consequently, VIA improves screening access but provides limited biological resolution and diagnostic precision compared with molecular approaches.

Colposcopy

Cytology and HPV testing function as screening and risk stratification tools; however, neither can establish a definitive diagnosis or directly determine treatment. Because clinical management ultimately requires histologic confirmation of disease severity, abnormal screening or triage results must be translated into a tissue-level diagnosis. Colposcopy with directed biopsy serves as a critical diagnostic bridge, converting screening-derived risk into histologically confirmed disease categories that anchor management decisions. The examination is performed with speculum placement and colposcopic magnification of the cervix, with a focused assessment of the transformation zone. Lesion borders and vascular features (punctuation, mosaicism, and atypical vessels) were used to guide biopsy site selection. In some protocols, Lugol’s iodine is added to distinguish glycogenated normal squamous epithelium (brown staining) from abnormal or immature epithelium (pale or unstained), further refining the target selection [80].

A key limitation of colposcopy-guided biopsy is its imperfect sensitivity when only a single worst-appearing site is sampled, because high-grade disease may be multifocal, and the most severe histology does not necessarily coincide with the visually most prominent lesion. Accordingly, evidence supports the use of multiple lesion-directed biopsies when more than one acetowhite area is present. Large datasets, including analyses derived from ALTS-era practice and subsequent multicenter evaluations, show that the sensitivity for clinically significant precancer increases when two or more directed biopsies are obtained compared with a single biopsy, at the cost of increased procedural burden [55, 81].

Endocervical sampling, such as endocervical curettage (ECC), can further increase the diagnostic yield in selected scenarios, such as incomplete visualization of the squamocolumnar junction, high-risk referral results with limited visible disease, or suspicion of glandular lesions. Comparative studies evaluating directed biopsy, random biopsy, and ECC have demonstrated that ECC and additional sampling can identify a subset of ≥CIN2 lesions not captured by directed biopsy alone, supporting context-dependent sampling strategies [82]. Consistent with this evidence, The American Society for Colposcopy and Cervical Pathology (ASCCP) colposcopy standards emphasize the documentation of examination adequacy, targeted sampling of the most abnormal areas, and selection of sampling strategies that balance incremental detection against procedural burden [83].

Taken together, colposcopy, HPV testing, and cytology form a complementary, workflow-integrated diagnostic continuum rather than competing alternatives. Each modality occupies a distinct role: HPV testing provides a molecularly grounded population-level risk anchor with high sensitivity but limited specificity for transforming infection; cytology refines risk among screen-positive individuals through morphological interpretation but is constrained by inter-observer variability and moderate sensitivity; and colposcopy translates risk estimates into tissue-confirmed diagnoses but is resource-intensive, operator-dependent, and not universally accessible (Table 1). The strength of each modality therefore corresponds to the limitation of another: HPV testing increases triage demand, cytology provides contextual interpretation but remains subjective, and colposcopy offers diagnostic confirmation at the cost of procedural burden. These complementary strengths and interdependent limitations underscore that further gains in cervical cancer control will require not only optimization of existing tools but also next-generation strategies that address systemic workflow gaps.

Table 1.

Comparative performance and implementation requirements of major cervical cancer screening modalities

Parameter	Cytology (Pap smear)	VIA	HPV DNA test	Colposcopy
Sensitivity	Moderate. operator-dependent	Moderate	High for CIN2+	High for lesion detection
Selectivity	Moderate to high	Low–moderate	Moderate	High (when biopsy confirmed)
Infrastructure requirement	Basic laboratory and \microscopy; specimen transport required	Minimal equipment, no laboratory	Molecular testing platform; cold chain and laboratory workflow	Dedicated clinical setup optical equipment, biopsy, and pathology support
Workforce requirement	Trained cytotechnologists and pathologists	Basic trained healthcare worker	Limited specialized personnel; less operator dependent	Highly trained specialist required
Cost per test	Low to moderate	Very low	Moderate to high	High
Feasibility in LRS	Limited by workforce and quality assurance	High	Increasingly feasible with simplified platforms and self-sampling	Limited; constrained by infrastructure and specialist available

These operational constraints are most acutely felt in LMICs, where cost, infrastructure limitations, and population-specific factors create compounding barriers to screening delivery. Standard molecular platforms demand substantial capital investment and cold-chain logistics that strain health budgets, while even simplified point-of-care formats remain costly without external subsidies [84]. Furthermore, infrastructure gaps extend beyond laboratories to workforce shortages, unreliable power, and fragile tracking systems that fail to ensure follow-up after a positive screen. Population-specific factors introduce further biological complexity: regional variations in hrHPV genotype distribution (such as a higher prevalence of non-16/18 types in sub-Saharan Africa and Asia) can lead to underestimation of disease risk, while highly prevalent HIV coinfection alters the specificity of triage biomarkers validated in immunocompetent cohorts [85]. Cultural acceptability of sampling modalities and systemic trust also dictate real-world uptake beyond mere analytical metrics.

Collectively, these observations indicate that bottlenecks in contemporary screening programs arise less from the absence of biologically effective tests than from operational fragility across the care continuum. Improvements in primary test sensitivity alone are insufficient unless the triage strategy, downstream capacity, and quality assurance are aligned and sustained.

Emerging diagnostic platforms for biomarker detection

Building directly on the workflow gaps identified above, this section turns to emerging diagnostic platforms designed to close them. Although current clinical practices, namely cytology and HPV DNA testing, are well established, they face significant implementation barriers in resource-limited environments. WHO recommends HPV molecular testing as a primary screening tool, but standard PCR-based platforms can remain difficult to deploy in decentralized settings because of equipment cost, workflow complexity, and cold-chain requirements [86, 87]. This section evaluates emerging diagnostic platforms across a broad spectrum of biological targets, ranging from conventional HPV DNA to non-protein metabolic markers, and organizes them by target class rather than technology type. For each platform, we assess four key dimensions: analytical performance, operational deployability, sampling strategy, and stage of clinical validation. This comparative framework enables a more structured evaluation of which platforms are closer to real-world deployment and which require further development before integration into scalable screening pathways.

HPV-targeted detection technologies

HPV detection remains the most etiologically specific anchor in cervical cancer screening. The clinical objective, however, extends beyond identifying viral presence to generating actionable diagnostic information across operationally diverse settings. Conventional nucleic acid–based assays offer excellent sensitivity but depend on centralized infrastructure, multistep workflows, and skilled personnel—barriers that limit deployment in resource-constrained environments. Because most HPV infections are transient, emerging detection technologies increasingly aim to refine specificity by targeting viral proteins, oncogene expression, or host-associated biomarkers alongside viral genomes. Colorimetric sensors have attracted considerable attention in LRS because of their ability to generate visually interpretable signals without the need for sophisticated instrumentation. Numerous studies have explored colorimetric approaches for HPV detection, frequently leveraging gold nanoparticles (AuNPs) as signal transducers owing to their strong optical properties and ease of use. Colorimetric sensors are often implemented as paper-based analytical devices (PADs), offering advantages such as portability, low cost, and ease of operation, making them well-suited for POC testing in LRS [88]. However, a major limitation of colorimetric detection platforms is their inherently lower sensitivity compared to fluorescence-based assays or next-generation sequencing (NGS)-based methods. To address this limitation, AuNP-based colorimetric platforms have evolved beyond simple aggregation-based readouts by incorporating flow-assisted formats and signal amplification strategies to improve their analytical performance.

In this context, lateral flow strips (LFS) have emerged as a particularly practical format, as demonstrated by their widespread clinical adoption, most notably in early pregnancy test kits [89]. By combining the visual simplicity of AuNP-labeled LFS with upstream nucleic acid amplification, Chen et al. developed a biosensor that integrates PCR with LFS to enable the simultaneous detection of high-risk HPV-16 and HPV-18 [90]. It is important to note that although this is a paper-based sensor, it does not rely on AuNP aggregation to generate a signal. Instead, it uses AuNPs as visual signaling reporters captured on designated test lines via a bridge effect created by dual-functionalized dsDNA amplicons. Central to this platform is a one-step direct lysis protocol that enables rapid nucleic acid extraction from cervical exfoliated cells at room temperature in less than 5 min, significantly reducing the complexity compared to traditional magnetic bead-based methods [91]. Amplicons were generated using species-specific primer sets labeled with FITC/biotin for HPV-16 and FITC/digoxin for HPV-18. During the lateral flow process, a universal AuNP-FITC antibody conjugate recognizes the FITC tag on the amplicon, whereas the second tag is captured by its respective receptor, streptavidin on the T_2 line for HPV-16 or a DIG antibody on the T_1 line for HPV-18, resulting in visible red color signals on the corresponding test lines (Fig. 8A). The platform demonstrated a high degree of analytical performance, detecting as few as 700 copies of HPV-16 and 18 DNA simultaneously within a total turnaround time of 75 min. This system exhibited extraordinary specificity, successfully discriminating between the two target genotypes and 18 other HPV types without any cross-interference. Clinical validation using 20 clinical samples showed complete agreement with the gold-standard real-time PCR, indicating that the platform can reliably identify co-infections and discriminate between severe and mild infection levels based on signal intensity. While the physical constraints of a single strip currently limit simultaneous detection to approximately five or six genotypes, researchers suggest that combining multi-LFS units or employing multi-round screening could address the need for broader HPV genotyping. Ultimately, this amplification-assisted LFS provides a robust alternative to expensive laboratory hardware, with the potential for future instrument-free diagnosis through the integration of isothermal amplification techniques such as recombinase polymerase amplification (RPA).

Fig. 8.

Gold nanoparticle (AuNP)-based colorimetric strategy for HPV biomarker detection. (A) Lateral-flow strip sensor: A rapid diagnostic workflow involving DNA extraction (5 min), multiplex PCR (mPCR, 1 h), and visual detection (3 min) using a multilayered test strip with specific capture probes (streptavidin, Dig-Ab, etc.). Adapted with permission. Adapted with permission [90]. Copyright 2022, Springer Nature (B) Sequence-dependent AuNP aggregation: Sensing mechanism using acpcPNA probes. In the presence of complementary HPV DNA, AuNPs remained dispersed (red), whereas non-complementary DNA induced aggregation (purple), achieving a detection limit of 1 nM. Adapted with permission. Adapted with permission [92]. Copyright 2021, RSC (C) Aptamer-based HPV L1 protein detection: A colorimetric assay in which aptamers stabilize AuNPs at pH 3. The specific binding of the HPV-16 L1 protein triggers aggregation, demonstrating high selectivity over other types, such as HPV-18 or HPV-6. Adapted with permission. Adapted with permission [95]. Copyright 2021, RSC (D) Aggregation-enhanced signal amplification: Target (HPV-16/18 L1)-induced aggregation of nanoparticle–antibody conjugates amplify the signal output via cluster formation, enhancing detection sensitivity

While some approaches improve sensitivity by changing the detection format or by incorporating upstream amplification, others focus on refining molecular recognition and nanoparticle chemistry underlying aggregation-based colorimetric sensing. A representative example of this approach is the paper-based colorimetric sensor developed by the Naorungroj group, which employs several complementary strategies to improve both the analytical sensitivity and sensor stability [92]. First, the authors addressed the limited stability of conventional citrate-capped AuNPs by adopting dextrin-stabilized AuNPs, which remained well-dispersed under biologically relevant ionic conditions [93]. Second, to enhance target recognition, a conventional DNA probe was replaced with a neutral peptide nucleic acid (PNA) probe [94]. A conformationally constrained pyrrolidinyl PNA with an α,β-peptide backbone derived from D-prolyl-2-aminocyclopentane carboxylic acid (acpcPNA) was used, providing stronger and more specific hybridization to the target DNA than conventional PNA. Finally, systematic optimization of the salt additives enabled the identification of conditions that maximized sensitivity by modulating AuNP aggregation behavior. The sensing mechanism relies on electrostatic interactions between negatively charged dextrin-capped AuNPs and positively charged acpcPNA probes. While incubating negatively charged target DNA with negatively charged AuNPs does not induce aggregation, the introduction of acpcPNA, bearing positively charged lysine residues at the C-terminus, promotes AuNP aggregation through charge-driven interactions. The hybridization of acpcPNA with complementary target DNA reduces the amount of free PNA available to induce aggregation, resulting in a target-dependent color change (Fig. 8B). However, at low target DNA concentrations, residual unbound acpcPNA can still trigger AuNP aggregation, potentially leading to ambiguity between the true target signals and background responses. Consequently, accurate quantification requires careful control of the probe concentration and precise construction of the calibration curve. This study successfully demonstrated a paper-based colorimetric detection system with smartphone-assisted RGB analysis, enabling visual and semi-quantitative readouts under low-resource conditions. The platform achieved sensitive detection across a DNA concentration range of 1–1000 nM and exhibited single-base-mismatch discrimination. Nevertheless, limitations became apparent in real-sample analysis. Although PCR-amplified DNA from HPV-positive (SiHa and CaSki) and HPV-negative (C33A) cell lines was tested, a clear signal was obtained only for the SiHa cell line, whereas CaSki samples produced weak or indistinguishable responses. This discrepancy was attributed to differences in the total DNA concentration between the samples, indicating that a minimum DNA concentration of approximately 1 µg/mL is required for reliable detection in practical settings.

Shifting the diagnostic focus from nucleic acids to structural biomarkers, a third approach employs a colorimetric aptasensor to detect the HPV-16 L1 capsid protein, a key biomarker representing 80% of the total capsid proteins [95]. This platform utilizes an RNA aptamer that initially adsorbs onto the surface of 20 nm AuNPs, providing a protective layer that ensures dispersion even under high-salt conditions. When the target L1 protein is introduced, the aptamer detaches from the AuNPs and binds with the protein owing to its strong affinity (K_D = 0.05 pM), causing the nanoparticles to aggregate and shift in color from wine red to purple (Fig. 8C). Under optimized conditions (60 mM NaCl at pH 3.0), the sensor achieved a visual detection limit of 9.6 ng/mL and a linear response up to 201.6 ng/mL. Despite its rapid 30-min incubation time and successful validation in clinical samples, this platform has certain limitations. The recovery rates in clinical samples ranged from 74.8% to 85.4%, indicating potential matrix interference from complex biological environments. Furthermore, although the sensor is highly selective for HPV-16 L1, its current design is restricted to a single high-risk genotype. For comprehensive clinical utility, this protein-based screening should be combined with DNA typing to reduce the rate of missed diagnoses and improve the overall accuracy.

Recently, antibody-based recognition of HPV capsid proteins using AuNPs has been reported as an alternative to aptamer-based colorimetric platforms (Fig. 8D) [96]. In this platform, ~ 20 nm AuNPs were synthesized using an optimized Turkevich method— a classical citrate reduction of Au(III) to metallic Au(0) in aqueous solution—guided by response surface methodology, yielding narrowly distributed particles (21.6 ± 0.1 nm, dispersity ~ 0.08) with high colloidal stability. The nanoparticles were sequentially stabilized with Tween 80 and hexadecyltrimethylammonium bromide and subsequently functionalized via the physical adsorption of monoclonal antibodies specific to HPV-16-L1 or HPV-18-L1. Upon exposure to the target antigen, antibody-mediated crosslinking induced AuNP aggregation, resulting in a pronounced surface plasmon resonance redshift from approximately 524 nm to 590 nm and a visible color change from red to violet. This aggregation process is accompanied by a marked increase in hydrodynamic diameter from ~ 27 nm to ~ 470 nm, directly linking antigen binding to the optical readout. Analytically, the assay exhibited a concentration-dependent response with good linearity for recombinant proteins, spanning 0.4–2.0 µg/mL for HPV-16-L1 (R² = 0.999) and 0.2–1.2 µg/mL for HPV-18-L1 (R² = 0.985). The limit of detection (LOD) for HPV-16-L1 was 0.18 µg/mL, with acceptable intra- and inter-assay precision (relative standard deviation 8.8–11.2%) and statistically validated accuracy across multiple concentration levels. Notably, the functionalized AuNPs retained their physicochemical properties and sensing capability for at least 90 days when stored at 4 °C, underscoring the robustness and practicality of the system for deployment outside centralized laboratory settings. When applied to 173 clinical cervical samples and benchmarked against PCR, the colorimetric assay demonstrated high specificity (77.8% for HPV-16 and 87.3% for HPV-18) and excellent negative predictive values exceeding 96% for both variants, indicating its strong performance as a rule-out screening tool. However, the clinical sensitivity was comparatively modest (54.5% for HPV-16 and 42.9% for HPV-18), and the positive predictive values were low, largely reflecting the limited number of PCR-positive cases in the studied cohort. A substantial fraction of false-positive results was associated with the presence of other HPV genotypes or abnormal cytological findings, which the authors attributed to structural conservation among HPV L1 proteins and partial antibody cross-reactivity. While this feature may allow the assay to capture broader HPV-related pathological changes beyond strict genotype specificity, it also highlights an inherent limitation of antigen-based colorimetric detection compared to highly sensitive nucleic acid amplification techniques. Overall, this work illustrates the trade-off between analytical sensitivity and operational simplicity, positioning AuNP-based colorimetric antigen detection as a promising, low-cost, and equipment-free approach for HPV screening in LRS where rapid exclusion of disease is prioritized over definitive genotyping.

In recent years, CRISPR–Cas–based nucleic acid detection strategies have gained significant attention as highly sensitive and programmable alternatives for HPV diagnostics, particularly in settings where rapid turnaround and multiplex capabilities are required. Leveraging the collateral (trans) nuclease activity of Cas12a, Yin et al. developed a CRISPR/Cas12a-based diagnostic platform coupled with multienzyme isothermal rapid amplification (MIRA) for the detection and genotyping of hrHPV in clinical specimens [97]. In this system, HPV L1 gene targets are first amplified isothermally within 15 min, after which target-specific crRNA–Cas12a complexes trigger nonspecific cleavage of a fluorophore–quencher–labeled ssDNA reporter, generating either a quantitative fluorescence signal or a visually interpretable readout under blue or UV light (Fig. 9A). By designing type-specific crRNAs, the authors achieved simultaneous detection and genotyping of all 14 hrHPV types, including HPV-16 and HPV-18, within a single-tube multiplex format. The assay demonstrated an analytical sensitivity as low as 2 copies/µL, with no detectable cross-reactivity against low-risk HPV types or other common microbial genomes, and delivered results within approximately 30 min. Importantly, large-scale validation using 258 clinical cervical samples showed 100% clinical sensitivity and 100% specificity when benchmarked against commercial qPCR assays, underscoring the robustness of CRISPR-based detection for clinical HPV screening.

Fig. 9.

(A) Fluorescence detection workflow: Following DNA extraction, target sequences are amplified via multi-enzyme isothermal rapid amplification (MIRA) at 37–42 °C for 15 min. The CRISPR/Cas12a complex then identifies the target and triggers the collateral cleavage of reporter molecules. Adapted with permission [97]. Copyright 2024, ASM (B) Electrochemical sensing mechanism: An alternative signal transduction method in which methylene blue-labeled DNA probes are immobilized on an electrode. Upon activation of the Cas12a enzyme by the HPV target, the probes are cleaved, leading to a significant decrease in the electrochemical current. Adapted with permission [98]. Copyright 2021, ACS (C) CMOS-based lab-on-chip (pH-LAMP): A semiconductor-based approach in which DNA is amplified via loop-mediated isothermal amplification (LAMP). The reaction causes pH changes, which are captured by an ion-sensitive field-effect transistor (ISFET) on a microchip. The resulting voltage changes were processed and visualized using a smartphone interface, enabling real-time monitoring without bulky optical equipment. Adapted with permission [99]. Copyright 2022, Springer Nature (D) CRISPR-enhanced digital testing (CreDiT): a highly integrated one-pot assay system. Following rapid nucleic acid (NA) extraction (16 min), target replication via RPA and Cas12/gRNA reactions occurred simultaneously. The integrated CreDiT device features a 12-sample holder/heater and compact fluorescent optics (LED and photodiode), allowing for rapid, automated, and high-throughput digital detection in field settings. Adapted with permission [100]. Copyright 2024, Springer Nature

In addition to fluorescence-based CRISPR diagnostics, Zamani et al. reported a distinct CRISPR/Cas12a strategy that integrates loop-mediated isothermal amplification (LAMP) with an electrochemical readout, which is specifically tailored for low-resource and point-of-care applications [98]. In this approach, LAMP-amplified HPV DNA activates Cas12a, which subsequently cleaves methylene blue–labeled oligonucleotides immobilized on ultra-low-cost gold leaf electrodes, resulting in a measurable decrease in the electrochemical signal (Fig. 9B). By targeting the E7 genes of HPV-16 and HPV-18, the platform achieved a lower LOD of approximately 1.2 × 10⁴ total copies, corresponding to clinically relevant viral loads associated with cervical cancer. Although the analytical sensitivity was lower than that of fluorescence-based CRISPR assays, the system compensated for this through extreme affordability (∼$0.50 per electrode, ∼$2.30 per test) and compatibility with portable potentiostats, making it particularly attractive for decentralized screening. Moreover, CRISPR-mediated sequence recognition effectively suppressed false positives commonly associated with LAMP, thereby enhancing assay specificity when applied to cervical swab samples. Together, these CRISPR–Cas–based strategies highlight a paradigm shift in HPV detection, wherein programmable nucleic acid recognition enables rapid, highly specific, and amplification-compatible diagnostics. While fluorescence-based CRISPR assays excel in analytical sensitivity and multiplex genotyping, electrochemical CRISPR platforms emphasize cost-effectiveness and field deployability, collectively illustrating how CRISPR technology can be flexibly adapted to diverse clinical and resource settings.

Despite the remarkable sensitivity of CRISPR–Cas-based nucleic acid detection, translating these assays into deployable POC formats remains challenging because of workflow complexity, signal instability, and instrumentation requirements. Building on technological progress toward high-precision and automated diagnostics, recent research has focused on integrating isothermal amplification with advanced signal processing and microfluidic sensors. In this context, Wormald et al. reported a handheld lab-on-chip (LoC) platform that combines LAMP with semiconductor-based sensing for cervical cancer detection [99]. Rather than relying solely on HPV DNA, this system multiplexes HPV-16/18 DNA detection with the tumor-specific biomarker human telomerase reverse transcriptase (hTERT) mRNA. hTERT, the catalytic subunit of telomerase, is markedly overexpressed in cervical lesions and is detectable in more than 90% of cervical cancers, while showing little to no expression in normal tissues, which underpins its utility as a tumor-specific biomarker [101, 102]. By integrating viral and host-derived molecular markers, this platform effectively addresses the limited clinical specificity often associated with HPV-only screening.

The LoC device is built on CMOS-based ion-sensitive field-effect transistor (ISFET) sensors that monitor pH changes generated during nucleic acid amplification in real time, eliminating the need for optical components (Fig. 9C). Using synthetic standards, the assay achieved single-copy sensitivity for mRNA targets, with an LOD of approximately 10² copies per reaction for hTERT, while maintaining amplification times below 25 min. In a pilot evaluation using cervical biopsy samples, hTERT mRNA was detected in all malignant tissues and in none of the benign controls, whereas HPV DNA was present in both malignant and some benign samples, highlighting the value of a dual-marker strategy for triage at the POC. By integrating amplification, detection, and electronic readout within a portable format, this study illustrates how LoC architectures can translate nucleic-acid-based diagnostics into clinically relevant, field-deployable screening tools.

While LoC platforms address hardware integration and workflow simplification, recent advances have focused on improving the signal robustness and on-site interpretability of CRISPR diagnostics. Lee et al. introduced CRISPR-enhanced digital testing (CreDiT), a system-level strategy that couples CRISPR-based nucleic acid recognition with engineered signal processing to enable reliable on-site detection [100]. CreDiT employs a one-pot reaction scheme in which recombinase polymerase amplification (RPA) and CRISPR–Cas12a detection occur simultaneously at a constant temperature, allowing target DNA amplification and CRISPR-mediated collateral cleavage of fluorophore–quencher reporters to proceed in a single closed tube (Fig. 9D). Importantly, although potential interference between Cas12a activity and RPA primers is a concern in one-pot designs, kinetic analysis demonstrated that the faster amplification dynamics of RPA dominate, enabling efficient signal generation without compromising amplification efficiency. The entire assay is performed under isothermal conditions and delivers results within approximately 35–40 min, including sample preparation. Beyond biochemical optimization, CreDiT places a strong emphasis on device- and signal-level engineering. The platform integrates a compact heating module with a low-cost fluorescence detection unit and employs digital signal processing based on Walsh–Hadamard transformation to enhance the signal-to-noise ratios and suppress background fluctuations. This approach enables a robust quantitative readout, even with simplified optics, overcoming the common limitation of portable fluorescence-based CRISPR diagnostics. Analytically, CreDiT achieved a reported LOD of ~ 40 copies/mL, outperforming conventional qPCR by approximately one order of magnitude, and demonstrated a broad dynamic range spanning multiple orders of magnitude. Cellular validation using HPV-16-positive CaSki cells further showed that the system could reliably detect signals corresponding to sub-single-cell equivalents, highlighting its suitability for samples with low viral loads. Clinical validation was performed using cervical and vaginal swab specimens, which were benchmarked against standard clinical PCR assays. CreDiT demonstrated high diagnostic accuracy, with area-under-the-curve values exceeding 0.97 for major hr-HPV genotypes, such as HPV-16, HPV-18, and HPV-45, and showed strong concordance with clinical reference tests. In addition, the platform incorporates simplified sample preparation and lyophilized reagent formulations, supporting a short hands-on time and reduced cold-chain dependence, which are key requirements for decentralized testing environments. Nevertheless, the system still relies on dedicated heating and fluorescence readout hardware, and subtype resolution may be influenced by differences in clinical reference classifications, indicating that further optimization and large-scale validation are required before widespread deployment.

Taken together, the HPV-targeted detection platforms reviewed here differ substantially across analytical performance, operational deployability, sampling strategy, and clinical validation stage. Colorimetric and LFS-based approaches offer low instrumentation burden and are well suited for community-level or decentralized POC use, but they may be limited by lower analytical sensitivity and restricted multiplexing capacity [90–96]. CRISPR-Cas12a systems coupled with isothermal amplification provide high analytical sensitivity and multiplex genotyping capability, but broader deployment will require dedicated readout hardware, reagent stabilization, and workflow simplification [97–100]. Lab-on-chip and ISFET-based platforms provide integrated workflows with electronic readout and dual-marker multiplexing, but they depend on microfabrication and device-level standardization that may limit near-term scalability [99]. Non-invasive urine- and pad-based sampling strategies lower participation barriers and improve accessibility, but they introduce preanalytical variability that can reduce sensitivity relative to clinician-collected specimens [103–105]. The optimal choice among these approaches will therefore depend on local healthcare context, including available infrastructure, required sensitivity, acceptable cost, and intended level of deployment (Table 2).

Table 2.

Reported clinical performance of molecular biomarkers and their diagnostic platforms

Biomarker	Endpoint	Sensitivity	Specificity	Available diagnostic platform	Ref.
HPV DNA test	CIN2+/CIN3+	Higha	Lowera (transient infections inflate positives)	PCR / signal-amplification HPV DNA assays (e.g., Cobas HPV, Onclarity)	[8]
HPV-16/18 partial genotyping (normal cytology)	CIN3+ (~ 2 year risk)	-b	- b	HPV DNA assay with partial/extended genotyping	[49, 50]
E6/E7 mRNA	CIN2+	Mid-90%	Higher than HPV DNA	Transcript-based (E6/E7 mRNA) assay	[129]
p16/Ki-67 dual stain	CIN3+	Low-90% (mid-90% in low-grade/normal)	low-60% ( ~ + 10 pp vs. cytology)	Immunocytochemistry dual-stain; paper-based p16ᴵᴺᴷ⁴ᵃ immunosensor (POC)	[55, 56, 108]
FAM19A4/miR124-2 methylation	CIN3 / cancer	High-70% (CIN3); mid-90% (cancer)	high-70%	Methylation-specific PCR	[57, 58]
Methylation + cytology	CIN3+	mid-80%	high-60%	Methylation-specific PCR + cytology	[58]

ᵃ Qualitative as described in text; no pooled numeric estimate is given in this review

ᵇ Values for partial genotyping are reported as cumulative CIN3 + risk, not test sensitivity/specificity: with normal cytology, ~ 2-yr CIN3 + risk is ~ 20% (HPV-16+), ~ 10% (HPV-18+), and ~ 5% (other hrHPV); persistent HPV-16 over 12 mo rises to ~ 30% [49, 50]

Sensitivity/specificity bands reflect the ranges reported in the cited studies. Endpoints differ across rows (CIN2+, CIN3+, risk, cytology grade) and should be compared with that in mind. Each platform is detailed in Table 3

As detection platforms become increasingly sensitive and deployable, attention has shifted toward noninvasive sampling strategies that can fully leverage these technologies by lowering participation barriers and improving real-world screening uptake. Recent studies have systematically explored non-invasive HPV sampling strategies with the shared goal of improving screening accessibility while preserving analytical validity. Urine-based sampling has emerged as one of the most extensively evaluated alternatives to clinician-collected cervical specimens for HPV testing. First-void urine (FVU) contains exfoliated cervical and vaginal cells enriched with HPV DNA, enabling reliable molecular detection when appropriate collection protocols are applied [103]. Clinical studies have demonstrated that HPV detection from FVU achieves substantial concordance with cervical samples, although sensitivity remains strongly influenced by urine dilution, collection timing, and DNA stability during storage. To address these limitations, subsequent studies have introduced urine concentration strategies, including centrifugal filtration and molecular enrichment, which significantly improve analytical sensitivity and reduce false-negative rates, particularly in samples with low viral loads [104]. These approaches underscore that urine is a biologically valid matrix for HPV detection; however, its diagnostic performance is highly dependent on preanalytical processing, which may complicate large-scale implementation in resource-limited or decentralized settings.

In addition to liquid biofluids, recent studies have proposed pad-based sampling as a strictly non-invasive and passive collection strategy that captures cervicovaginal secretions over an extended period. In a pilot clinical study, a feminine pad worn for 8 h enabled the successful extraction and amplification of HPV DNA, with detection rates increasing progressively from healthy individuals to patients with precursor lesions and cervical cancer [105]. Notably, the pad-based approach demonstrated high sensitivity for clinically significant disease (CIN2 + and cervical cancer) while achieving near-universal participant acceptance and compliance. By integrating prolonged sampling time with self-collection, this strategy effectively mitigates the dilution and variability inherent to spot urine sampling while avoiding the discomfort and cultural barriers associated with pelvic examination. Collectively, these studies illustrate a continuum of noninvasive sampling innovation in which biological validity, user acceptability, and real-world feasibility are progressively optimized.

Detection of non-HPV biomarkers: protein markers

In addition to HPV nucleic acid and antigen detection, increasing attention has been directed toward non-HPV host-derived biomarkers that directly reflect cellular transformation, metabolic dysregulation, and tumor-associated processes during cervical carcinogenesis. Among these, p16^INK4a has emerged as one of the most clinically established surrogate markers for HPV infection. Overexpression of p16^INK4a arises from the functional inactivation of the Rb pathway by the HPV E7 oncoprotein and is closely associated with high-grade lesions than HPV DNA positivity alone [106]. An immunoassay-based study demonstrated that p16^INK4a could be quantitatively detected with high reproducibility, highlighting its utility as an objective triage biomarker to improve specificity and reduce unnecessary colposcopy referrals in HPV-positive populations [107]. Building on the clinical relevance of p16^INK4a as a surrogate marker of transforming HPV infection, Yokchom et al. developed a paper-based immunosensor that translates this established biomarker into a rapid and low-cost POC screening format (Fig. 10A) [108]. The platform integrates antigen recognition and signal amplification within a single paper-based device, enabling instrument-free visual detection in low-resource settings. The assay employs multi-functionalized gold nanoparticles conjugated with horseradish peroxidase (HRP)–labeled anti-p16^INK4a antibodies, allowing simultaneous target capture and signal generation. Upon binding to p16^INK4a extracted from cervical cells, the immune complexes are immobilized on a nitrocellulose membrane, where signal amplification arises from the combined peroxidase activity of HRP and the intrinsic enzyme-mimicking activity of gold nanoparticles, producing a visible color change through the catalytic oxidation of 3,3’,5,5’-tetramethylbenzidine (TMB). Using HPV-associated cervical cancer cell lines, the sensor generated clear positive signals while maintaining minimal background in p16^INK4a-negative controls, achieving naked-eye detection limits in the range of several hundred to a few thousand cells, depending on the cell type. When evaluated in clinical cervical specimens, p16^INK4a positivity was observed predominantly in precancerous and cancerous samples, yielding an overall diagnostic accuracy of 85.2% with substantial agreement relative to cytological classification, supporting its potential utility as a triage-level screening tool. Notably, the platform offers rapid assay completion within 30 min, requires only a small sample volume (25 µL), and enables straightforward visual interpretation without specialized equipment, addressing the key limitations of HPV DNA-based screening in LRS. However, the assay still relies on cell-based sampling and lysis, and its sensitivity may be insufficient for lesions with very low biomarker abundance, indicating the need for further optimization and large-scale validation before broader clinical implementation.

Fig. 10.

Emerging diagnostic platforms for non-human papillomavirus biomarkers in cervical cancer. In addition to viral nucleic acid detection, host-derived protein biomarkers provide complementary information on cellular transformation, metabolic dysregulation, and tumor-associated processes. (A) p16INK4a detection: Paper-based colorimetric immunosensor integrating antigen recognition and enzymatic signal amplification. HRP-labeled anti-p16INK4a antibodies conjugated to gold nanoparticles enabled simultaneous target capture and catalytic TMB oxidation, producing a visible signal suitable for rapid and low-cost screening. Adapted with permission [108]. Copyright 2018, Elsevier (B) Nitroreductase (NTR) activity detection: A ratiometric fluorescence-based strategy using a molecular probe that undergoes enzymatic reduction in the presence of NTR and NADH, generating dual-emission fluorescence changes (FI530/FI630) is used. This approach enables the functional monitoring of tumor-associated metabolic activity. Adapted with permission [109]. Copyright 2021, Molecules (C) Caspase-3 sensing: A surface-enhanced Raman scattering (SERS)-based platform employing peptide-functionalized Au nanostructures. Caspase-3–mediated cleavage induces nanoparticle aggregation, amplifying the Raman signals and enabling the sensitive detection of apoptotic activity. Adapted with permission [110]. Copyright 2021, RSC (D) Protein phosphatase-1-gamma-2 (PP1γ2) detection: An antibody-functionalized gold nanoparticle assay exploiting localized surface plasmon resonance (LSPR) modulation. Antigen binding alters the optical absorption spectrum, enabling the colorimetric or electronic readout of this urine-detectable host biomarker. Adapted with permission [111]. Copyright 2021, Molecules

While p16^INK4a reflects cell cycle deregulation as a downstream consequence of oncogenic HPV activity, other non-HPV biomarkers aim to capture functional changes in tumor metabolism and the microenvironment. Consistent with this concept, Kim et al. introduced a ratiometric fluorescence probe designed to monitor nitroreductase (NTR) activity as a functional readout of tumor-associated metabolic alterations (Fig. 10B) [109]. Rather than quantifying biomarker expression levels, this strategy focuses on enzymatic activity itself, leveraging the fact that NTR is upregulated under hypoxic and reductive conditions commonly observed in malignant tissues. The probe was constructed using a locked-flavylium fluorophore scaffold bearing a nitroaromatic moiety that serves as a substrate for NTR. Upon enzymatic reduction of the nitro group to an amine in the presence of NADH, the probe undergoes a pronounced internal charge-transfer shift, resulting in dual fluorescence emission at distinct wavelengths. This enables a ratiometric signal readout (FI₅₃₀/FI₆₃₀), in which changes in enzyme activity are quantified by the ratio of the two emission channels rather than the absolute intensity, thereby minimizing signal fluctuations caused by probe concentration, photobleaching, or local environmental effects. In solution, the probe exhibited selective and concentration-dependent ratiometric fluorescence changes in response to NTR activity, with a low detection limit and minimal interference from biologically relevant species present in the sample. Importantly, when applied to live-cell imaging, significantly elevated fluorescence ratios were observed in HeLa cervical cancer cells compared to non-cancerous fibroblasts, indicating increased intracellular NTR activity. Confocal microscopy further confirmed efficient cellular uptake, biocompatibility, and stable signal generation within a short incubation time, supporting the feasibility of NTR activity-based sensing in living systems. A key advantage of this approach is its functional and self-calibrated nature, as ratiometric detection provides robust signal reliability while directly reporting tumor-associated metabolic states rather than static protein abundance. Moreover, because NTR activity is linked to hypoxia and reductive stress rather than HPV infection per se, this strategy offers a complementary biomarker axis that is independent of the viral genotype and viral load. Nevertheless, the study is limited to proof-of-concept demonstrations in cultured cancer cell lines without validation in clinical cervical specimens or tissue-derived samples. In addition, although elevated NTR activity was observed in HeLa cells, the lack of explicit cervical cancer-specific clinical correlation necessitates further investigation to establish diagnostic specificity and translational relevance in the context of cervical cancer screening.

In addition to metabolic alterations, biomarkers associated with cell fate and execution pathways offer additional layers of disease characterization. Caspase-3, a key executioner protease in apoptosis, has been investigated as a downstream indicator of cancer progression and therapeutic responses. Sun et al. reported a nanomaterial-assisted sensing strategy that exploits caspase-3 activity as a functional indicator of apoptotic progression in cervical cancer cells (Fig. 10C) [110]. Rather than targeting static expression levels, the platform was designed to dynamically monitor enzymatic activation associated with cell fate execution, thereby capturing the downstream biological consequences of disease progression or therapeutic intervention. The sensing mechanism is based on surface-enhanced Raman scattering (SERS) using hollow gold nanoboxes functionalized with caspase-3–specific peptide substrates and a Raman reporter molecule. Upon caspase-3–mediated cleavage of the peptide linker, electrostatic changes on the nanoparticle surface induce the controlled aggregation of the nanoboxes, generating plasmonic hot spots that dramatically amplify the Raman signal. Finite-difference time-domain simulations confirmed that these interparticle nanogaps are the primary source of signal enhancement, providing a mechanistic basis for the ultrasensitive detection strategy. Using purified enzymes and cell lysates, the probe exhibited a broad dynamic range from 1 fM to 1 nM, with an exceptionally low detection limit of 0.127 fM. When applied to live-cell experiments, HeLa cells treated with doxorubicin to induce apoptosis displayed a time-dependent increase in the SERS signal intensity, reflecting progressive caspase-3 activation. Importantly, spatiotemporal SERS mapping enabled real-time visualization of intracellular caspase-3 distribution during apoptosis, with quantitative agreement between SERS-based measurements and conventional western blot analysis. This approach offers several notable advantages, including ultrasensitivity, resistance to photobleaching, and the ability to dynamically monitor enzymatic activity within living cells rather than relying on endpoint measurements. By directly interrogating apoptotic signaling, such platforms provide valuable insights into disease dynamics and therapeutic responses, complementing upstream biomarkers related to HPV infection or metabolic dysregulation. However, the applicability of caspase-3–based sensing for population-level screening remains limited. Caspase-3 activation represents a late-stage or treatment-associated event rather than an early indicator of malignant transformation, and the current demonstration is confined to in vitro cervical cancer cell models without validation in clinical specimens. As such, while these nanomaterial sensors are unlikely to replace primary screening tools, they may have potential for monitoring disease progression and therapeutic response, although this remains to be demonstrated beyond in vitro models in clinical specimens.

Parallel efforts have been made to identify cancer-specific protein antigens that enable non-invasive detection without relying on HPV-derived targets. Protein phosphatase-1 gamma-2 (PP1γ2), also known as a non-cervical biomarker antigen (NCB-Ag), has been reported to be selectively expressed in cervical cancer tissues and is detectable in patient urine samples. Basak et al. proposed a concept that bypasses cervical sampling altogether by detecting a tumor-associated host protein in urine, addressing both invasiveness and accessibility limitations inherent to cytology-based screening (Fig. 10D) [111]. To translate this concept into a POC format, the authors developed a gold-nanoparticle-based colorimetric immunosensor that exploits LSPR. Gold nanoparticles were functionalized with a 3,3′-dithiodipropionic acid di(n-hydroxysuccinimide ester) linker to immobilize anti-PP1γ2 antibodies, forming a stable Ab–AuNP composite. Upon specific epitope–paratope interaction with PP1γ2 present in urine, changes in the interparticle distance altered the local dielectric environment, inducing a characteristic hypsochromic shift and an increase in absorbance in the plasmonic spectrum. This LSPR modulation served as an optical readout for antigen recognition. The presence of PP1γ2 in urine was first validated using western blotting and RT-PCR in samples from patients with cervical cancer, CIN, recurrence, and treatment stages, while it remained undetectable in healthy controls. When applied to clinical urine specimens, the AuNP-based assay produced a pronounced blue shift and intensity change selectively in cervical cancer-positive samples, with the signal magnitude correlating with the disease stage. An interference study using healthy urine samples confirmed the high specificity of the sensor. A portable prototype integrating an LED light source and photodiode enabled a direct electronic readout, demonstrating its feasibility for POC operations. Compared with HPV nucleic acid-based detection, the PP1γ2 urine sensor offers several operational advantages: fully non-invasive sampling that bypasses pelvic examination, independence from HPV genotype or viral load, and compatibility with low-resource or self-testing contexts through visual or electronic readout. These features position it as a complementary diagnostic axis rather than a replacement for primary HPV screening. However, several limitations constrain its near-term translational potential. PP1γ2 is present at low urinary concentrations, requiring sensitive plasmonic readout and careful calibration to reduce false-negative risk. Antibody-based recognition may introduce batch-to-batch variability and potential cross-reactivity. In addition, available clinical validation remains limited, and large-scale prospective multicenter studies will be required before this platform can be considered for population-level deployment. In its current form, the PP1γ2 urine sensor is best viewed as an early-stage proof-of-concept rather than a clinically deployable screening tool.

Detection of non-HPV biomarkers: non-protein markers

In addition to protein-centric biomarkers, recent studies have explored non-protein molecular and phenotypic signatures associated with cervical carcinogenesis, motivated by the need to capture disease-associated biochemical alterations that are not directly reflected by protein expression. These approaches aim to bypass the limitations inherent to antigen stability, antibody availability, and inter-individual protein variability, while leveraging small-molecule metabolites or holistic biological responses that emerge during early malignant transformation.

Biothiols, such as cysteine (Cys), homocysteine (Hcy), and glutathione (GSH), are central regulators of redox balance, antioxidant defense, and cellular metabolism [112]. In cancer cells, persistent oxidative stress driven by rapid proliferation, hypoxia, and metabolic reprogramming leads to dysregulated biothiol homeostasis as a compensatory mechanism [112, 113]. Consequently, altered biothiol levels, either intracellularly or in excreted biofluids, have emerged as functional biochemical signatures of malignancy. Unlike protein biomarkers that reflect expression states, biothiols capture dynamic metabolic and redox phenotypes, positioning them as attractive non-protein biomarkers for cancer detection, particularly in non-invasive and low-resource diagnostic settings. The exploration of biothiols as non-protein urinary biomarkers for cervical cancer has progressed through a clear conceptual evolution, moving from single-analyte detection to chemically discriminative sensing of structurally similar thiols.

An early milestone in this trajectory was reported by An et al. in 2022, who introduced the first urine-based in vitro diagnostic platform for cervical cancer using fluorescent cysteine probes [114]. In this study, we observed that urinary cysteine levels were elevated in patients with cervical cancer, reflecting increased antioxidant demand associated with tumor progression. The molecular probe, NPO-B, was designed to selectively react with cysteine via nucleophilic addition, followed by a Smiles rearrangement, resulting in a pronounced fluorescence turn-on signal (Fig. 11A). Importantly, this reaction proceeded rapidly and selectively under physiological conditions, enabling direct fluorescence analysis of untreated urine samples. When applied to a large cohort of approximately 1,500–1,800 clinically characterized urine samples, the assay achieved a true-positive rate of approximately 74% for cervical cancer detection, demonstrating, for the first time, the feasibility of non-invasive, urine-based cancer screening without relying on HPV-derived markers (Fig. 11B). However, because the sensing mechanism was optimized for cysteine alone, the platform could not distinguish cysteine from other structurally related biothiols, such as homocysteine, limiting its ability to capture finer metabolic differences within the biothiol pool.

Fig. 11.

Chemical sensing mechanisms for urinary biomarker detection in cervical cancer screening. Host-associated biomarkers, such as biothiols, offer complementary diagnostic value when integrated with noninvasive sampling formats. (A) NPO-B undergoes nucleophilic substitution followed by a Smiles rearrangement in the presence of cysteine (Cys). Adapted with permission [114]. Copyright 2022, Elsevier (B) Practical application of the NPS-B probe for rapid (~ 30 min) urinary testing, demonstrating a true positive rate of 74% in cervical cancer patients compared to an 11.7% false-positive rate in healthy controls. (C) NPS-B demonstrates dual-channel sensing capability, differentiating Homocysteine (Hcy) and Cysteine (Cys) through controlled addition and substitution reactions, resulting in distinct S-form (green emission) and N-form (red emission) products. Adapted with permission [115]. Copyright 2024, ACS (D) Representative fluorescence images and statistical analysis showing significantly elevated signal intensities in the patient group compared to the healthy group for both Cys and Hcy channels (p < 0.05)

Building on this limitation, a subsequent study published in 2024 extended the chemical sensing strategy from biothiol detection to biothiol discrimination [115]. In this later work, the authors recognized that cysteine and homocysteine, despite their close structural similarity, may have distinct biological significance in cancer-associated redox metabolism, and that their selective differentiation could enhance diagnostic specificity. To address this challenge, researchers developed a nitrobenzene-based fluorescent probe engineered to undergo a controlled Smiles rearrangement, in which the reaction pathway and kinetics could be tuned through the rational modification of the electronic substituents and the introduction of a selenium atom (Fig. 11C). This refined chemical design enabled the formation of distinct S-bound- and N-bound adducts upon reaction with cysteine and homocysteine, respectively, producing well-separated fluorescence emission bands. The probe allowed the simultaneous discrimination of Cys and Hcy within complex urine matrices without sample pretreatment. When applied to clinical urine samples from patients with cervical cancer, the assay generated characteristic emission patterns that were not accessible with earlier single-analyte probes (Fig. 11D). Together, these two studies exemplify the evolution of urine-based biothiol sensing from single-marker detection to chemically resolved discrimination, illustrating a rational progression in non-protein biomarker design for cervical cancer screening.

Rather than attempting to quantify individual biothiol species, Li et al. adopted a fundamentally different diagnostic philosophy by treating biothiol dysregulation as a pattern-level biochemical phenotype (Fig. 12A) [116]. Their study was motivated by the observation that cancer cells rarely alter a single metabolite in isolation but instead exhibit coordinated shifts across redox-related small molecules. A sensing platform was constructed using Ag–MoS₂ hybrid nanozymes, which exhibited strong peroxidase-like catalytic activity toward chromogenic substrates. Biothiols inhibited this activity through metal–thiol interactions, with the inhibition strength varying depending on the thiol identity and solution pH. By operating the same nanozyme system across multiple pH conditions, the authors generated a set of cross-reactive colorimetric responses that collectively formed a multidimensional fingerprint of the biothiol environment of the target cells. Multivariate statistical analysis, particularly principal component analysis, was used to classify the fingerprints. Experimentally, the sensor array successfully discriminated between cysteine, homocysteine, and glutathione across a range of concentrations (10–100 µM) and mixture compositions. Importantly, when applied to biological samples, the system differentiated cancer cell lysates, including HeLa cervical cancer cells, from normal cells, demonstrating that cancer-associated biothiol phenotypes could be resolved without targeting a specific molecule. This approach offers notable advantages owing to its marker-agnostic and integrative nature, reducing dependence on single biomarkers, and enhancing robustness against biological variability. Simultaneously, its reliance on statistical pattern recognition rather than direct molecular identification introduces challenges in biological interpretability and requires careful clinical correlation before diagnostic translation.

Fig. 12.

Pattern-based sensing platforms for non-protein biomarkers of cervical cancer. In addition to single-molecule detection, systems-level sensing approaches have been developed to capture the coordinated biochemical and metabolic alterations associated with cervical carcinogenesis. (A) Biothiol phenotype discrimination: A cross-reactive colorimetric sensor array based on an Ag–MoS₂ hybrid nanozyme exhibiting peroxidase-like catalytic activity. Biothiols inhibit nanozyme activity through metal–thiol interactions, with inhibition profiles varying by thiol species and pH. By operating the system under multiple pH environments, a multidimensional response pattern was generated and analyzed using principal component analysis, enabling the discrimination of cysteine, homocysteine, glutathione, and cancer-associated cell lysates without targeting a single analyte. Adapted with permission [116]. Copyright 2024, RSC (B) Volatile metabolite–based phenotypic sensing (N-NOSE): A chemotaxis assay utilizing Caenorhabditis elegans to detect cancer-associated volatile organic compounds in urine. Nematode behavioral responses are quantified into a diagnostic index, enabling the sensitive detection of preinvasive lesions, such as CIN2 and CIN3, through pattern-level metabolic signatures rather than defined molecular identification. Adapted with permission [117]. Copyright 2024, Acc Science

Extending the concept of non-protein biomarkers beyond defined chemical species, a markedly different paradigm has emerged that exploits biological pattern recognition rather than molecular identification of biomarkers. In this approach, the olfactory behavior of Caenorhabditis elegans was leveraged to detect cancer-associated volatile organic compounds in urine (Fig. 12B) [117]. The N-NOSE platform quantifies nematode chemotaxis toward urine samples, translating collective behavioral responses into a diagnostic index. In a clinical cohort, this system reported a sensitivity of 73% (180/245) for CIN and 100% (9/9) for CIN3; however, the latter estimate is based on a very small subgroup and was not accompanied by a correspondingly characterized specificity, so it should be interpreted as preliminary rather than as evidence of near-perfect detection. This strategy exemplifies a shift from single-analyte detection to systems-level phenotypic sensing, capturing emergent disease signatures arising from global metabolic reprogramming. However, despite its impressive sensitivity, this approach has intrinsic limitations. Because the N-NOSE platform relies on an integrated behavioral response rather than the detection of defined molecular entities, the specific chemical drivers underlying the observed chemotactic patterns remain unidentified, rendering the system largely a black-box sensor. This lack of molecular attribution complicates mechanistic interpretation, limits explainability, and makes it difficult to distinguish cancer-specific signals from confounding metabolic or physiological states, such as diabetes, ketosis, infection, or dietary variation. Moreover, the readout is inherently qualitative or semi-quantitative and depends on the behavior of living organisms, which poses challenges in terms of standardization, reproducibility, and large-scale clinical deployment. Finally, although such phenotypic sensing platforms may serve as highly sensitive front-line screening tools, their inability to localize diseases, identify molecular targets, or provide pathway-level information restricts their utility in precision diagnostics, therapy selection, and mechanism-driven translational research.

Collectively, these non-protein biomarker platforms differ meaningfully in mechanistic specificity, translational maturity, and standardization feasibility. Single-analyte biothiol probes provide chemically defined and molecularly interpretable readouts, representing a relatively advanced approach within this category. Pattern-based nanozyme arrays capture integrated redox phenotypes across multiple biothiol species without requiring identification of a single molecular target, offering robustness against single-biomarker variability but requiring statistical classification and further interpretability studies. Organism-based N-NOSE sensing captures broader VOC or metabolic patterns and has shown promising sensitivity in pilot clinical settings, but its black-box nature, dependence on living organisms, and behavioral readout variability pose challenges for standardization and regulatory approval. Thus, single-analyte probes may be more immediately translatable because of their defined molecular mechanism, whereas pattern- and organism-based platforms may serve as complementary non-invasive screening tools after further mechanistic characterization and prospective multicenter validation.

Non-biomarker-based diagnostic platforms

In parallel with molecular and biochemical sensing strategies, a growing body of work has explored non-biomarker-based diagnostic approaches that rely on tissue-level optical, electromagnetic, and image-derived signatures rather than discrete molecular targets. The motivation for these approaches lies in the recognition that cervical neoplasia induces global alterations in tissue architecture, biochemical composition, and physical properties, which can be captured through advanced imaging and signal analysis, even in the absence of explicitly defined biomarkers. Such strategies are particularly attractive in low-resource or high-throughput screening settings, where reagent-free, rapid, and operator-independent diagnostics are urgently required.

Several optical and spectroscopic platforms have been developed to exploit disease-associated tissue signatures. Raman spectroscopy and Fourier-transform infrared (FT-IR) spectroscopy enable label-free interrogation of biochemical fingerprints arising from nucleic acids, proteins, lipids, and carbohydrates, with multivariate statistical models or deep learning algorithms used to classify disease states [118, 119]. Wang et al. explored Raman spectroscopy–based biochemical fingerprinting as a label-free approach for the detection and classification of cervical precancerous lesions and cervical cancer, motivated by the need for a rapid, objective, and non-invasive diagnostic alternative to cytology- and HPV-based screening (Fig. 13A) [120]. Rather than targeting a specific molecular biomarker, this study was based on the premise that cervical carcinogenesis is accompanied by progressive and systematic alterations in the biochemical composition of tissues, which can be captured as vibrational spectral signatures. Using confocal Raman spectroscopy, the authors collected spectra from six histopathologically confirmed cervical tissue types, including cervicitis, CIN I–III, cervical squamous cell carcinoma, and cervical adenocarcinoma. Raman signals in the 400–1800 cm⁻¹ range reflected contributions from proteins, lipids, nucleic acids, collagen, and amino acids, with disease progression associated with reproducible changes in the relative intensities of characteristic peaks [121]. These spectral differences were interpreted as manifestations of underlying biochemical remodeling during malignant transformation, such as alterations in lipid content, protein secondary structure, and nucleic acid composition. To translate these complex spectral datasets into diagnostic classifications, Raman spectra were analyzed using a support vector machine (SVM) algorithm. Based on more than 1,100 spectra acquired from 210 tissue samples, the resulting model achieved an overall diagnostic accuracy of 85.7% in distinguishing inflammatory tissue, precancerous lesions, and invasive cancers, with a particularly high classification performance for cervical squamous cell carcinoma and adenocarcinoma. The analysis further revealed statistically significant spectral differences between successive disease stages, supporting the ability of Raman spectroscopy to capture gradual biochemical evolution, rather than binary disease states. Compared with cytology-based screening, Raman spectroscopy offers several potential advantages: label-free and reagent-free operation, objective biochemical classification, and reduced reliance on observer-dependent morphology. It may also capture gradual biochemical evolution across successive CIN stages rather than relying only on binary morphological thresholds. This feature could help address a specific gap in conventional cytology, namely the underdetection of adenocarcinoma, by capturing biochemical rather than purely morphological signatures. However, relative to other emerging platforms, Raman spectroscopy faces significant translational barriers. The current evidence base relies largely on ex vivo tissue sections rather than non-invasive or minimally invasive samples, limiting its immediate applicability to population-level screening. Spectral acquisition requires specialized confocal instrumentation, and classification models require further standardization and external validation before clinical adoption. Therefore, Raman-based fingerprinting is best positioned as a laboratory-based adjunct for objective lesion grading and adenocarcinoma characterization rather than a frontline screening tool.

Fig. 13.

Integration of artificial intelligence (AI) and machine learning (ML) in cervical cancer diagnostics. (A) Raman Spectroscopy with SVM: Integration of a Confocal Raman Micro-Spectrometer with Support Vector Machine (SVM) algorithms. This approach analyzes the molecular vibrational fingerprints of cervical cells, achieving a diagnostic accuracy ranging from 77% to 100%. Adapted with permission [118]. Copyright 2021, Springer Nature (B) FT-IR Spectroscopy with CNN: A serum-based diagnostic platform utilizing Fourier-transform infrared (FT-IR) spectroscopy. Combined with a particle swarm optimization–convolutional neural network (PSO-CNN) model, this system offers a minimally invasive screening tool with a diagnostic accuracy of up to 87%. (C) Digital cytopathology workflow: Whole slide images (WSIs) are digitized and processed using RetinaNet for abnormal cell detection. Features related to detection were then analyzed using a Random Forest Classifier to achieve precise cell classification (e.g., ASC-US, LSIL, HSIL), reducing manual screening errors. Adapted with permission [123]. Copyright 2024, Springer Nature

Complementarily, FT-IR spectroscopy integrated with convolutional neural networks has enabled the rapid classification of cervical cancer and precancerous lesions from serum samples, achieving accuracies above 85% while minimizing sample preparation and assay complexity [122]. Ma et al. reported a serum-based FT-IR spectroscopy platform combined with deep learning for the rapid screening of cervical cancer and its precancerous stages, aiming to overcome the subjectivity, invasiveness, and limited sensitivity of conventional cytology- and HPV-based screening methods (Fig. 13B). Rather than focusing on individual molecular biomarkers, this study was grounded in the concept that cervical carcinogenesis is accompanied by systematic biochemical remodeling in blood serum, which can be captured as disease-specific infrared absorption fingerprints. Using ATR-FT-IR spectroscopy, the authors acquired vibrational spectra from serum samples collected from patients with cervical cancer, CIN I–III, and uterine fibroids. The spectra reflect the collective biochemical contributions of nucleic acids, proteins, and lipids, with disease progression associated with reproducible intensity changes in the characteristic absorption bands. In particular, increased nucleic acid-related peaks and altered lipid- and protein-associated bands were observed along the CIN–cancer continuum, consistent with enhanced cellular proliferation, membrane remodeling, and metabolic reprogramming during malignant transformation.

To translate these complex spectral signatures into clinically actionable classifications, we implemented a particle swarm optimization–guided convolutional neural network (PSO-CNN) framework. Unlike conventional CNN models with fixed architectures, the PSO-CNN algorithm automatically optimizes both the network structure and hyperparameters, thereby enabling efficient learning from one-dimensional Fourier-transform infrared (FT-IR) spectral data. When evaluated across five disease categories, the optimized PSO-CNN model achieved an average classification accuracy of 87.2%, outperforming classical deep learning architectures, such as LeNet, AlexNet, VGG16, and GoogLeNet. Notably, the model demonstrated robust discrimination across multiple CIN stages and invasive cervical cancer, suggesting potential utility for early-stage stratification that requires confirmation in independent, prospectively collected cohorts. These accuracies, however, were obtained in a single-center retrospective cohort without external or prospective validation, and the reported superiority over standard CNN architectures may partly reflect dataset-specific tuning. A key strength of this approach is its noninvasive, marker-independent diagnostic philosophy, as it does not require antibodies, molecular probes, or prior knowledge of specific biomarkers. By combining FT-IR spectroscopy with adaptive deep learning, this platform enables rapid and objective analysis of serum samples with minimal preparation, making it particularly attractive for large-scale screening and deployment in resource-limited settings. However, the study was conducted on a relatively limited cohort, and the FT-IR spectral features underlying the model predictions remain indirectly interpretable, necessitating further validation and biological correlation. In addition, although the classification performance was strong, clinical translation will require standardized spectral acquisition protocols and prospective validation in larger, multicenter populations. Thus, FT-IR–based serum fingerprinting is best positioned as a complementary screening and triage tool rather than a standalone diagnostic replacement within the emerging landscape of non-protein, phenotype-based cervical cancer diagnostics.

In addition to vibrational spectroscopy, fluorescence-based imaging and smartphone-enabled optical platforms have further expanded accessibility. A standalone smartphone-based fluorescence spectroscopy device captured intrinsic fluorescence changes associated with the metabolic and structural remodeling of cervical tissue, and when coupled with machine learning classifiers, achieved sensitivities and specificities exceeding 94% across multiple clinical sites [124]. By integrating data acquisition, analysis, and classification into a portable system, this approach demonstrates the feasibility of POC, reagent-free cervical cancer screening.

At the macroscopic level, AI-assisted image analysis has been applied to both cytology slides and VIA images, significantly improving diagnostic consistency and sensitivity compared with manual interpretation alone. Harsono et al. evaluated an AI-assisted visual inspection platform (Cerviray AI^®) to improve the objectivity and accessibility of cervical precancer screening in low- and middle-income settings, where VIA remains the primary screening modality [125]. The motivation for this study stemmed from the well-recognized limitations of VIA, namely, its operator dependence and variability in diagnostic performance, despite its low cost and immediate availability of results. By integrating AI into a portable colposcopy system, we sought to standardize the visual interpretation of acetowhite lesions and support clinical decision-making in resource-limited environments. The Cerviray system combines a handheld colposcopic imaging device with AI-based image analysis software trained to classify cervical images into normal, CIN1, CIN2/3, and CIN3 + categories according to the Bethesda classification. In addition to the fully automated AI output, the platform enables telemedicine-based expert review (Cerviray^® expert), allowing remote specialists to reassess uploaded images and provide confirmatory interpretations. In this study, VIA served as the reference standard against which the diagnostic performance of both the AI algorithm and expert-assisted evaluation were assessed. In a cross-sectional pilot study involving 44 high-risk women in West Java, Indonesia, Cerviray AI^® demonstrated a sensitivity of 42.9% and specificity of 100% for detecting precancerous lesions, yielding a positive predictive value of 100% and an area under the receiver operating characteristic curve (ROC AUC) of 71.4%. Although the AI-alone sensitivity was modest, expert-assisted interpretation substantially improved performance, achieving a sensitivity of 71.4%, specificity of 97.3%, and ROC AUC of 84.4%, which is comparable to or exceeds the reported performance ranges for conventional VIA screening. Kappa analysis further indicated substantial agreement between AI-assisted and expert evaluations, supporting the clinical reliability of the platform. A key strength of this approach is its pragmatic integration with existing screening infrastructure, as it preserves the simplicity and low cost of VIA while mitigating subjectivity through AI support and telemedicine connectivity. The portability of the device, rapid result generation, and minimal training requirements make it particularly attractive for deployment in geographically dispersed or underserved regions where access to trained colposcopists is limited. However, this study has some limitations. The relatively low sensitivity of the fully automated AI system suggests that Cerviray AI^® is better positioned as a decision-support or triage tool rather than as a standalone screening replacement. In addition, the small sample size and single-center design restrict generalizability, underscoring the need for larger multicenter validation studies with histopathology-based gold standards. As such, AI-assisted VIA platforms represent an incremental but meaningful advance toward more objective cervical cancer screening, complementing rather than supplanting molecular or phenotype-based diagnostic approaches.

A particularly influential example of image-derived, non-biomarker-based diagnostics is the recent large-scale study by Wang et al., who developed and clinically validated an artificial intelligence cervical cancer screening (AICCS) system for the whole-slide analysis of LBC specimens [123]. This study is significant because it directly addresses one of the central bottlenecks of cytology-based screening: the dependence on highly trained cytopathologists, the time-consuming nature of manual slide review, and the substantial inter-observer variability that contributes to false-negative diagnoses. Rather than proposing AI as a proof-of-concept classifier, the authors aimed to establish a clinically deployable, end-to-end screening system capable of operating at the population scale. The AICCS framework consists of a two-stage deep learning architecture with a patch-level detection model for identifying suspicious cells and a whole-slide classification model for assigning cytologic grades from entire digitized slides (Fig. 13C). The system was trained and validated using multiple datasets, including retrospective, prospective, and randomized observational cohorts, comprising over 16,000 participants. In the prospective evaluation, the model achieved an area under the ROC curve (AUC) of 0.947, with a sensitivity of 94.6%, specificity of 89.0%, and overall accuracy of 89.2% for cytology grade prediction. Most notably, in a randomized observational trial, cytopathologists assisted by the AICCS demonstrated significantly improved diagnostic performance compared with cytopathologists alone, including an absolute sensitivity gain of approximately 13.3%, along with improvements in AUC, specificity, and overall accuracy. A key strength of this approach lies in its fully marker-independent, image-based diagnostic philosophy: the system operates directly on routine whole-slide cytology images without requiring molecular reagents, additional staining, or explicit feature engineering. By embedding deep learning into the existing cytology workflow, this platform offers a scalable solution to workforce shortages, reduces reading time and fatigue, and improves consistency across readers and centers. Importantly, the study design, incorporating multicenter, prospective, and randomized components, provides a level of clinical validation that is rarely achieved in AI-based diagnostic research. Despite its strong performance, the system is fundamentally dependent on high-quality slide digitization infrastructure and remains tied to the cytology sampling process itself, which retains intrinsic limitations in terms of sensitivity. Moreover, while the model improves overall diagnostic accuracy, its decision-making process is not fully interpretable at the biological level, and its generalizability to different preparation protocols and population settings requires further prospective, real-world validation. Accordingly, AI-assisted whole-slide cytology analysis is best viewed not as a replacement for molecular screening or tissue diagnosis but as a powerful, infrastructure-level augmentation of morphology-based screening within the broader ecosystem of non-biomarker, phenotype-driven cervical cancer diagnostics.

Collectively, these non-biomarker-based strategies offer several key advantages, including reagent-free operation, rapid readout, and the ability to capture holistic disease signatures that emerge from complex tissue remodeling processes. However, they also face important limitations, such as dependence on sophisticated instrumentation, sensitivity to data quality and population heterogeneity, and challenges in biological interpretability, compared with molecularly targeted assays. Therefore, these approaches are best viewed not as replacements but as complementary modalities that can augment biomarker-based screening, particularly when integrated into multimodal and AI-assisted diagnostic frameworks.

Conclusion and perspectives

Cervical cancer represents a unique paradox in oncology: it is largely preventable but remains a major cause of cancer-related mortality worldwide. This discrepancy reflects not only biological complexity but also disparities in implementation across different health systems. In this review, we examined cervical cancer diagnostics across the full continuum of the disease, from viral infection and molecular transformation to population-level screening and emerging technologies, within an integrated prevention framework. At the biological level, cervical carcinogenesis is driven by persistent hrHPV infection, followed by progressive host genomic and epigenetic remodeling. The extended preinvasive window provides a critical opportunity for detection; however, effective prevention depends on distinguishing between transforming diseases and transient infections. This need has driven a paradigm shift from morphology-based detection to molecular risk stratification.

Current screening programs rely on a complementary triad: HPV testing as the molecular risk anchor, cytology as a triage tool, and colposcopy-directed biopsy for histological confirmation of disease. Although these modalities have reduced the incidence and mortality of organized programs, their effectiveness remains limited by biological heterogeneity, operational complexity, and variability in implementation.

Despite these advances, several key challenges persist. First, global disparities in screening coverage and care continuity continue to drive mortality outcomes, with loss to follow-up and limited diagnostic capacity remaining as major bottlenecks rather than test availability alone. Second, biological heterogeneity complicates risk stratification, as HPV positivity alone lacks specificity and current triage tools do not fully capture viral oncogene activity, host genomic remodeling, and immune evasion — contributing to both overtreatment and under-detection of high-risk lesions. Third, existing modalities face infrastructural limitations, and although emerging platforms show promise, their translation into population-level impact is constrained by additional real-world barriers: formal cost-effectiveness evaluations are lacking for most next-generation platforms; regulatory frameworks, particularly for AI-based systems, remain unsettled with respect to algorithmic transparency and post-market monitoring; inter-laboratory variability in sample handling and result interpretation limits reproducibility across institutions; and cold-chain requirements, electricity dependence, and population-specific factors including regional HPV genotype distribution and HIV coinfection restrict deployment in low-resource settings. Fourth, it should be emphasized that most candidate biomarkers do not ultimately reach clinical practice [126], and that AI-assisted tools frequently show performance degradation when moved beyond their development cohort [127, 128]; the technologies discussed here therefore remain investigational until validated in multicenter, prospective studies. Taken together, these barriers underscore that next-generation diagnostic development must incorporate cost-effectiveness evaluation, standardized protocols, quality assurance systems, and implementation science endpoints alongside conventional measures of diagnostic accuracy.

Future progress in cervical cancer prevention will depend on aligning biological insights with scalable implementation. First, next-generation diagnostics should advance toward multi-layered risk stratification by integrating viral, host, and contextual biomarkers to distinguish transient infection from transforming disease more accurately. Advances in epigenetic profiling, integration-aware biomarkers, and multi-analyte approaches may enable more precise and clinically actionable risk predictions. Second, decentralized and rapid diagnostic platforms are essential to reduce loss to follow-up and support screen-and-treat strategies, particularly in resource-limited settings. Self-sampling and point-of-care (POC) testing are key components of this transition toward accessible and scalable screening. Third, digital and AI-assisted diagnostics may enhance consistency and efficiency by augmenting expert interpretations rather than replacing them. Finally, diagnostic innovation must be integrated across the entire prevention continuum, including vaccination, screening, and treatment, to achieve a sustained population-level impact.

Translating these directions into practice, the biomarkers and platforms discussed in this review currently occupy three levels of clinical readiness. The most mature are those already embedded in, or readily compatible with, current screening and triage algorithms: primary HPV DNA testing, HPV-16/18 partial genotyping, HPV E6/E7 mRNA assays, and p16/Ki-67 dual staining [23, 24, 47–52, 55, 56]. These are closest to translation because they rely on existing cervical samples and established laboratory workflows. At an intermediate stage are host methylation assays (notably FAM19A4/miR124-2), AI-assisted cytology, CRISPR-based POC systems (e.g., CreDiT, MIRA-Cas12a), and ISFET-based lab-on-chip platforms [57, 58, 97–100, 123]. These show clinically relevant performance and may improve triage efficiency, reader consistency, or decentralized use, but broader adoption awaits prospective validation, standardized thresholds, workflow harmonization, and clear regulatory pathways. Earlier-stage technologies include non-invasive urine- or pad-based sampling, metabolomic/VOC-based sensing, Raman/FT-IR spectroscopy, and organism-based platforms [103–105, 114–122]. Although these may broaden non-invasive or marker-free screening, their clinical role depends on reproducibility, biological interpretability, cost-effectiveness, and integration into care pathways. In summary, near-term translation will be led by established molecular and cytology-based triage tools, while continued validation of AI-assisted, POC, and non-invasive sensing platforms will shape the future of precision cervical cancer screening.

Cervical cancer is one of the few malignancies for which elimination is biologically and technically achievable; however, implementation gaps continue to limit progress. The central challenge has shifted from detection alone to the delivery of integrated, biologically informed, and scalable prevention strategies. This review presents a biology-aligned, target-centric framework that connects molecular transformation with diagnostic strategies and clinical workflows. By synthesizing past advances, current limitations, and emerging innovations, this review provides a conceptual foundation for guiding the development of next-generation screening approaches.

A roadmap toward global cervical cancer elimination will require the convergence of molecular diagnostics, scalable and decentralized testing platforms, and system-level implementation to ensure equitable access to prevention, diagnosis, and treatment. The continued integration of biological insights with clinical and public health strategies offers a realistic path to transforming cervical cancer prevention from a theoretical possibility to a globally achievable outcome.

Table 3.

Diagnostic performance and field-deployability of novel HPV detection technologies. A comprehensive evaluation of recently developed biosensors based on their target biomarkers (DNA, mRNA, and proteins), limit of detection (LOD), and turnaround time, specifically focusing on their translatability to decentralized clinical environments

Type	sample type	target (DNA vs. protein)	amplification	LOD	time-to-result	equipment requirement	POC suitability	clinical samples tested	Ref
LFS Biosensor	Cervical exfoliate cells	DNA (HPV-16, 18)	PCR	700 copies	75 min	Minimal (PCR thermal cycler)	High (Rapid lysis, visual readout)	20 patient samples (100% agreement with qPCR)	[90]
acpcPNA AuNP sensor	PCR-amplified DNA	DNA (HPV-16, 18)	PCR (pre-amplified)	1 nM	Rapid (Visual/ RGB)	Low (Smartphone for analysis)	Moderate (Requires pre-amplified DNA)	SiHa, CaSki, C33A cell lines	[92]
Apta sensor (AuNP)	Clinical & vaccine samples	Protein (HPV-16 L1)	None	9.6 ng/mL	30 min	Very Low (Visual)	High (Equipment-free, rapid)	Clinical & vaccine samples (74.8–85.4% recovery)	[95]
Ab-functionalized AuNP	Cervical samples	Protein (HPV-16/18 L1)	None	0.18 µg/mL (HPV-16)	Rapid	Low (Visual/Redshift)	High (Stored at 4 °C for 90 days)	173 clinical samples (NPV > 96%)	[96]
CRISPR/Cas12a (MIRA)	Cervical specimens	DNA (14 h-HPV types)	MIRA (Isothermal)	2 copies/µL	~ 30 min	Low (Blue/UV light or Fluorescence)	High (Isothermal 37–42 °C)	258 clinical samples (100% Sen/Spe)	[97]
Electrochemical CRISPR	Cervical swabs	DNA (HPV-16, 18 E7)	LAMP (Isothermal)	1.2 × 10⁴ copies	Rapid	Low (Portable potentiostat)	Very High ($2.30/test, gold leaf electrode)	Cervical swab samples	[98]
CMOS Lab-on-Chip	Cervical biopsy	DNA & mRNA (HPV-16/18, hTERT)	LAMP (Isothermal)	10² copies (hTERT)	< 25 min	Low (Handheld LoC device)	High (Real-time pH monitoring, no optics)	Malignant vs. Benign biopsy samples	[99]
CreDiT	Cervical & Vaginal swabs	DNA (HPV-16, 18, 45)	RPA (One-pot Isothermal)	~ 40 copies/mL	35–40 min	Moderate (Integrated heating/ fluorescence)	High (Lyophilized reagents, digital signal)	Cervical/Vaginal swabs (AUC > 0.97)	[100]

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1 ^{(496.1KB, docx)}

Supplementary Material 2 ^{(25.7KB, docx)}

Supplementary Material 3 ^{(3.4MB, docx)}

Abbreviations

acpcPNA

α, β-Peptide nucleic acid (D-prolyl-2-aminocyclopentanecarboxylic acid-based)

AI

Artificial intelligence

AICCS

Artificial intelligence cervical cancer screening

ALTS

ASCUS-LSIL triage study

ASC-US

Atypical squamous cells of undetermined significance

ASCCP

American Society for Colposcopy and Cervical Pathology

ASR

Age-standardized rate

ATR

Attenuated total reflectance

AUC

Area under the curve

AuNP

Gold nanoparticle

BRD4

Bromodomain-containing protein 4

CIN

Cervical intraepithelial neoplasia

CMOS

Complementary metal-oxide-semiconductor

CNN

Convolutional neural network

CreDiT

CRISPR-enhanced digital testing

CRISPR

Clustered regularly interspaced short palindromic repeats

Cys

Cysteine

DNA

Deoxyribonucleic acid

dsDNA

Double-stranded DNA

DTSP

3,3′-Dithiodipropionic acid di(N-hydroxysuccinimide ester)

ECC

Endocervical curettage

FITC

Fluorescein isothiocyanate

FT-IR

Fourier-transform infrared

FVU

First-void urine

GSH

Glutathione

H3K27ac

Histone H3 lysine 27 acetylation

Hcy

Homocysteine

HDI

Human development index

HIF-1α

Hypoxia-inducible factor 1-alpha

HIV

Human immunodeficiency virus

HPV

Human papillomavirus

hrHPV

High-risk human papillomavirus

HRP

Horseradish peroxidase

HSIL

High-grade squamous intraepithelial lesion

hTERT

Human telomerase reverse transcriptase

IARC

International Agency for Research on Cancer

ISFET

Ion-sensitive field-effect transistor

Ki-67

Marker of proliferation Ki-67

LAMP

Loop-mediated isothermal amplification

LBC

Liquid-based cytology

LCR

Long control region

LED

Light-emitting diode

LFS

Lateral flow strip

LMICs

Low- and middle-income countries

LoC

Lab-on-chip

LOD

Limit of detection

LRS

Low-resource settings

LSIL

Low-grade squamous intraepithelial lesion

LSPR

Localized surface plasmon resonance

MIRA

Multicolor isothermal rapid amplification

ML

Machine learning

mRNA

Messenger ribonucleic acid

NADH

Nicotinamide adenine dinucleotide (reduced form)

NCB

Non-cervical biopsy

NGS

Next-generation sequencing

N-NOSE

Nematode-nose

NPV

Negative predictive value

NTR

Nitroreductase

NTCC

New Technologies for Cervical Cancer screening

p16INK4a

Cyclin-dependent kinase inhibitor 2A

p53

Tumor protein p53

PAD

Paper-based analytical device

PCR

Polymerase chain reaction

PD-L1

Programmed death-ligand 1

PNA

Peptide nucleic acid

POC

Point-of-care

PP1γ2

Protein phosphatase-1 gamma-2

pRb

Retinoblastoma protein

PSO

Particle swarm optimization

RGB

Red, green, blue

RNA

Ribonucleic acid

ROC

Receiver operating characteristic

RPA

Recombinase polymerase amplification

RT-PCR

Reverse transcription polymerase chain reaction

SCC

Squamous cell carcinoma

SERS

Surface-enhanced Raman scattering

SVM

Support vector machine

TCGA

The Cancer Genome Atlas

TMB

3,3′,5,5′-Tetramethylbenzidine

UV

Ultraviolet

VIA

Visual inspection with acetic acid

VOC

Volatile organic compound

WHO

World Health Organization

WSI

Whole slide image

Author contributions

JSY and GP contributed equally. JSY: Conceptualization, Investigation, Writing - original draft, Writing - review and editing. GP: Writing - original draft, Writing - review and editing. HS: Conceptualization, Investigation, Supervision, Writing - original draft, Writing - review, and editing. YHK: Conceptualization, Investigation, Supervision, Funding acquisition, Writing - original draft, Writing - review and editing. DK: Conceptualization, Investigation, Supervision, Funding acquisition, Writing - original draft, Writing - review and editing.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (RS-2025-00518296 to D.K.; RS-2024-00406152 to Y.H.K.) and the Bio & Medical Technology Development Program of the NRF funded by the MSIT (RS-2024-00439078 to D.K.; RS-2023-00223591 to Y.H.K.). This study was also supported by the Bio & Medical Technology Development Program (Bio-Conjugate Core Technology Development Project) of the NRF funded by the Ministry of Science & ICT in 2026 (RS-2025-25396539 to G.P.).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.