Addressing cervical cancer screening disparities through advances in artificial intelligence and nanotechnologies for cellular profiling

Abstract

Almost all cases of cervical cancer are caused by the human papilloma virus (HPV). Detection of pre-cancerous cervical changes provides a window of opportunity for cure of an otherwise lethal disease when metastatic. With a greater understanding of the biology and natural course of high-risk HPV infections, screening methods have shifted beyond subjective Pap smears toward more sophisticated and objective tactics. This has led to a substantial growth in the breadth and depth of HPV-based cervical cancer screening tests, especially in developed countries without constrained resources. Many low- and middle-income countries (LMICs) have less access to advanced laboratories and healthcare resources, so new point-of-care (POC) technologies have been developed to provide test results in real time, improve the efficiency of techniques, and increase screening adoption. In this Review, we will discuss how novel decentralized screening technologies and computational strategies improve upon traditional methods and how their realized promise could further democratize cervical cancer screening and promote greater disease prevention.

I. INTRODUCTION

Cervical cancer remains a significant health threat to women worldwide as the fourth most diagnosed cancer with over half a million new cases each year.^1,2 Cervical dysplasia invariably occurs via human papillomavirus (HPV) infection;³ continued presence of high-risk HPV subtypes catalyze transformation into invasive cervical cancer.⁴ Mechanistically, HPV encodes eight to ten proteins, including E6 and E7, which promote abnormal cellular growth by inactivating tumor suppressor proteins (e.g., p53 and pRB), cyclins, and cyclin-dependent kinases.³ E6 or E7 proteins encoded by high-risk HPV subtypes (notably 16 or 18) bind with increased affinity and hence inactivate their targets to a greater extent than low-risk HPV E6 or E7 proteins.³ High-risk HPV subtypes thus become high value screening targets before sustained cellular damage occurs. Of the approximately 311,000 annual deaths from cervical cancer, more than 85% of these occur in low- and middle-income countries (LMICs).^5,6 Primary prevention through vaccinations toward pre-specified HPV subtypes and secondary prevention through screening to detect and treat precancerous lesions must act congruously to create successful prevention programs.^7,8 In LMICs, where HPV vaccines are not widely introduced or adopted, decentralized HPV screening through low-cost and point-of-care (POC) technologies is urgently needed.⁹ In this Review, we highlight how emerging advances in nanotechnologies and machine learning can (1) complement or supplant existing methods of cervical cancer screening and (2) directly circumvent existing screening obstacles in LMICs.

II. CONVENTIONAL CERVICAL CANCER SCREENING

The Papanicolaou (Pap) test, introduced in the 1940s by George Papanicolaou, helps pathologists examine the morphology of exfoliated cervical cells.¹⁰ Throughout this period, both the Pap test and liquid-based cytology were the gold standards of cervical cancer screening. Importantly, they significantly decreased cervical cancer morbidity and mortality within resource-rich countries. About 80% of cervical cancer can be prevented by well-organized, high-quality screening programs using Pap smears with three- to four‐year screening intervals.¹¹ Despite these accomplishments, the method has yet to similarly impact developing countries.¹² An effective cytology-based screening program necessitates high-quality cytology laboratories, properly trained personnel, and repeated screening at regular intervals due to the low sensitivity of a single Pap test; such ingredients are still not achievable across many LMICs.^12,13 Consequently, other screening options are being explored and piloted in both developed countries and LMICs.^14–16

For resource-challenged areas with pathology bottlenecks, visual inspection with acetic acid (VIA) and Lugol's iodine (VILI) has been explored as a cost-effective alternative to cytology screening.^17–19 Visual screening provides real-time results, establishing a “screen-and-treat approach” where women are not lost to follow-up. Additionally, VIA and VILI can be performed by nurses, midwives, and paramedic staff after a short training course,^19,20 making the method more accepted across many LMICs. The VIA method showed an overall sensitivity of 80% and a specificity of 92%.²¹

While VIA and VILI have gained traction in LMICs, certain limitations have challenged their outcomes.²² Specifically for older women, a degenerating cervical epithelium and difficult visibility of the transition zone lowers the accuracy of these tests. In addition, VIA-based screening still relies on healthcare provider subjectivity and lacks definitive quality assurance control.²¹ In all, concerns with (a) accuracy, sensitivity, and reproducibility; (b) high inter-operator variability; and (c) scale-up challenges have limited the utility and clinical penetration of VIA and VILI within national screening programs.²³ The variable interpretation of the subjective clinical exams may also lead to over- or under-treatment.²⁴

III. MOLECULAR SCREENING METHODS

In response to the above-mentioned challenges, more effective screening methods leverage the underlying biology of cervical cancer with well-benchmarked diagnostic platforms. The demanding infrastructure needed for Pap testing and its inferior sensitivity are circumvented by direct and non-subjective testing for high-risk HPV subtypes. Over the last decade, HPV DNA testing has been incorporated into cervical cancer screening efforts within developed countries.²⁵ In 2018, the U.S. Preventive Services Task Force issued draft recommendations to include HPV screening every 5 years for women aged 30 to 65 years.²⁶ As compared with Pap testing, HPV testing has greater sensitivity (39% improvement), while only 3% less specific than Pap testing for detecting cervical intraepithelial neoplasia.²⁷ Here, we will provide critical reviews on molecular screen methods and highlight commercially available tests (Table I) and next-generation technologies under development.

TABLE I.

Overview of commercial HPV screening tests. qPCR, Quantitative polymerase chain reaction; TMA, Transcription-based amplification; NASBA, Nucleic acid sequence-based amplification.

Name	Biomarker Type	Genotyping	Detection Method	Test Timesa	FDA Approved (Year)	Company
Digene Hybrid Capture 2 High-Risk HPV DNA Test	DNA	Detection of 13 high-risk HPV types (16/18/31/33/35/39/45/51/52/56/58/59/68), but cannot determine the specific HPV type present	Hybridization, chemi-luminescence		Yes (1991)	QIAGEN
careHPV Test Kit	DNA	Detection of 14 high-risk HPV types (16/18/31/33/35/39/45/51/52/56/58/59/66/68)	Hybridization, chemi-luminescence		No	QIAGEN
Xpert HPV	DNA	Identification of HPV 16 and 18/45, and concurrent detection of the remaining 11 high-risk types (31/33/35/39/51/52/56/58/59/66/68)	qPCR, fluorescence		No	Cepheid
Cobas HPV Test	DNA	Identification of HPV 16 and 18, and concurrent detection of the remaining 12 high-risk types (31/33/35/39/45/51/52/56/58/59/66/68)	qPCR, fluorescence		Yes (2011)	Roche
Cervista HPV HR	DNA	Detection of 14 high-risk HPV types (16/18/31/33/35/39/45/51/52/56/58/59/66/68), but does not discriminate between the HPV types	Invader, fluorescence		Yes (2009)	HOLOGIC
Cervista HPV 16/18	DNA	Detection of HPV 16 and 18	Invader, fluorescence		Yes (2009)	HOLOGIC
Onclarity HPV Assay	DNA	Identification of HPV 16, 18, and 45, and concurrent detection of the remaining 11 high-risk types (31/33/35/39/51/52/56/58/59/66/68)	qPCR, fluorescence		Yes (2018)	BD
Aptima HPV Assay	mRNA	Detection of E6/E7 mRNA from 14 high-risk HPV types (16/18/31/33/35/39/45/51/52/56/58/59/66/68), but does not discriminate between the HPV types	TMA, fluorescence		Yes (2011)	HOLOGIC (Gen-Probe)
Aptima HPV 16 18/45 Genotype Assay	mRNA	Detection of E6/E7 mRNA from HPV 16, 18, and 45; can differentiate HPV 16 from HPV 18, and/or 45	TMA, fluorescence		Yes (2012)	HOLOGIC (Gen-Probe)
HPV-Proofer	mRNA	Detection of E6/E7 mRNA from HPV 16, 18, 31, 33, and 45	NASBA, fluorescence		No	PreTect
NucliSens EasyQ HPV	mRNA	Detection of E6/E7 mRNA from HPV 16, 18, 31, 33, and 45	NASBA, fluorescence		No	bioMérieux
OncoE6 Crvical Test	Onco-protein	Identification of E6 protein from HPV 16 and 18	Sandwich assay, lateral flow		No	Arbor Vita

^a Estimated single assay time: , 0∼1 h; , 1∼3 h; , 3∼5 h; , > 5 h.

A. Current generation tactics

1. Cobas HPV test

The Cobas HPV test (Roche Molecular Systems, Pleasanton, CA, USA) utilizes real-time polymerase chain reaction (PCR) and nucleic acid hybridization to amplify target DNA and detect up to 14 types of high-risk human HPV, including HPV 16 and HPV 18, which account for the majority of cervical cancers. While HPV 16 and 18 are detected individually from their own channels, the other 12 oncogenic HPV types are pooled together in a single channel. There is an additional channel for detecting human β-globin gene for sample adequacy as a positive β-globin result indicates the presence of human cells in the vial. To reduce false positives, the AmpErase enzyme is included to correct possible cross-contamination and separate target molecules from amplified products.

The Cobas HPV test was previously evaluated in cervical intraepithelial neoplasia (CIN) grade 2 or worse (≥CIN2) disease within an overall population of women older than 20 years of age.²⁸ For the overall population, the assay sensitivity for ≥CIN2 was 90% with specificity at 58%. Meanwhile, for women with atypical squamous cells of undetermined significance (ASCUS), the sensitivities for ≥CIN2 and ≥CIN3 were 93% and 96.3%, respectively. The specificities in this population were 70.6% and 69.5% for ≥CIN2 and ≥CIN3, respectively.

Unlike many other tests, a major factor rendering the Cobas HPV test a gold standard for cervical cancer screening is the ability to individually discriminate the two most common high-risk HPV types, HPV 16 and HPV 18. When compared with Sanger sequencing and Qiagen Hybrid Capture 2, the analytic accuracy was superior for the three different populations tested: ASCUS (aged ≥21 years), intraepithelial neoplasia (aged ≥30 years), and overall (aged ≥25 years). While the clinical value of the Cobas HPV is reflected in the results of this²⁸ and other validation studies,^29,30 its cost efficiency and sustained feasibility for rural settings in developed or developing countries remains under debate. For example, the test requires a Cobas ×480 instrument for specimen preparation and the use of either the Cobas 6800 or 8800 systems for analyses. Along with the proprietary and costly reagents needed for both PCR and nucleic acid hybridization, the assay becomes increasingly impractical for laboratories in LMICs.

2. CareHPV test

Due to the limitations of subjectivity and under-/over-treatment residing in VIA for wide-scale cervical cancer screening, careHPV (QIAGEN, Gaithersburg, MD, USA) arose from a private-public collaboration with the goal of an easily affordable and usable HPV screening approach.⁵ The test reflects a signal-amplification assay targeting DNA from different HPV subtypes, including HPV 16, HPV 18, and others. The assay consists of denaturing DNA from lysed cells and hybridizing them to full-length complementary RNA. Monoclonal antibodies coated on paramagnetic beads then attach to the DNA/RNA hybrids, and alkaline phosphatase-linked anti-hybrid antibodies bind and detect the hybrids. Finally, a chemiluminescent substrate is added to induce the hybrids to emit light; the extent of emission correlates with the amount of HPV DNA present in these complexes.^31,32 The careHPV platform benefits from its ability to be operative by less skilled personnel in resource-limited settings without need for room temperature control or flowing water. In addition, the technology allows for cervical self-sampling that could capture greater parts of marginalized populations, since prior studies have shown self-collected and clinician-collected samples provide similar results.⁵

To test careHPV screening performance across a variety of settings, the technology was introduced in Nicaragua, India, and Uganda, and compared to VIA and Pap testing.⁵ In each site, at least 5000 women were selected and enrolled for testing based on recruitment strategies targeting each specific population. Each woman underwent clinician-collected vaginal sampling, Pap testing, and VIA; in addition, a self-collected cervical sample was provided. For samples from 16,951 eligible women screened, cervical careHPV testing offered the highest sensitivities (81.5% for CIN2+ and 85.3% for CIN3+) and specificities (91.6% and 91%, respectively) overall. For vaginal careHPV tests, sensitivities were 69.6% for CIN2+ and 71.3% for CIN3+, while specificities were 90.9% and 90.5%, respectively.⁵ Both the VIA and Pap tests offered inferior performance, highlighting the potential for including low-cost, user-friendly options such as careHPV into screening algorithms. Overall, HPV DNA testing reflects a viable option for population-based screening programs and self-collected samples could offer efficiencies for healthcare providers and patients.

3. Xpert HPV test

The Xpert HPV test (GeneXpert; Cepheid, Sunnyvale, CA, USA) is an easy-to-use test that allows for same-day results thus introducing “screen-and-treat” programs into LMICs.³³ The test, which detects at least 14 different high-risk HPV types, comprises proprietary reagents, primers, and probes, as well as a human reference gene and an internal probe check control. The cartridges are disposable with readouts in approximately an hour, allowing for more efficient patient visits. Also, the test utilizes the Cepheid GeneXpert platform previously offered in LMICs for tuberculosis detection, lending potential feasibility to the approach.³⁴

A Papua New Guinea study involving 1005 women aged 30 to 59 years old instructed them to self-collect cytobrush specimens.³⁴ In addition, gynecological examinations enabled clinician-collected cervical cytobrush specimens. The researchers noted a 12.3% presence of all high-risk HPV types combined, mostly driven by other HPV types at 9.0%; HPV 16 and HPV 18/45 were prevalent in 3.5% and 1.6% of instances, respectively. Interestingly, the self-collected and clinician samples achieved a high correlation of findings. Overall percentage agreements (OPA) between HPV 16 and HPV 18/45 reached 99.4% and 98.5%, respectively, while the OPA for all high-risk HPV types combined were 93.4%. Some disagreements occurred between the Xpert HPV tests for HPV 16 and HPV 18/45. Yet, many more false positives were noted for the other high-risk HPV types. Not unlike the careHPV test, self-cervical sampling reflects another potential source of decentralized specimen collection for the centralized screening process.

4. Aptima HPV Assay

Aptima HPV Assay (Hologic, San Diego, CA, USA) detects HPV E6/E7 mRNA from 14 types of high-risk HPV. Compared to DNA-based tests, Aptima HPV Assay shows similar sensitivity and improved specificity. This assay mainly involves 3 steps: (1) capture of target mRNA using capture oligomers linked to magnetic microparticles; (2) target mRNA amplification with transcription-mediated amplification; (3) detection of amplicons using the Hybridization Protection Assay.³⁵ Hologic offers an additional product, Aptima HPV 16 18/45 Genotype Assay, to further genotype HPV types 16, 18, and/or 45 from women with Aptima HPV Assay positive results. Aptima HPV Assay and Aptima HPV 16 18/45 Genotype Assay are both approved by the U.S. Food and Drug Administration (FDA). Ge et al. compared the performance of Aptima HPV Assay with the Cobas HPV test for patients with biopsy-confirmed high-grade cervical lesions or worse lesions.³⁶ Both methods showed high sensitivities of 97%. However, Aptima HPV Assay exhibited lower positive rates than Cobas HPV test in benign and low-grade cervical lesions specimens, which resulted in the significantly higher specificity of Aptima HPV Assay compared to Cobas HPV test (88% vs 72%).

5. OncoE6 Cervical Test

The OncoE6 Cervical Test (Arbor Vita, Fremont, CA, USA) is based on a lateral flow method to detect HPV 16/18 E6 oncoproteins in high grade cervical lesions with higher specificity than most technologies commercially available. The test utilizes genotype-specific mouse monoclonal antibodies in a strip format to detect high levels of the E6 oncoprotein for the two high-risk HPV types. The technology was shown to have high specificity of 98.9% compared to careHPV and VIA. However, this test did have a much lower sensitivity of 53.5% for CIN3+, and 70.3% for lesions positive for HPV 16/18/45.³⁷

A more recent study of the test investigated results between self-cervical brushing samples and physician-collected swabs among 20 patients.³⁸ Specifically, three samples were taken from each patient: vaginal lavages from a self-sampling device, cervical swabs, and cytobrushings. Results among all three groups were mostly unvarying and consistent. As the study was more focused on determining the technical feasibility of the technology using self-samples rather than larger population-based analyses, its incorporation into LMICs needs further testing.

B. Next-generation technologies

Given the reduced access to advanced laboratories and resources in LMICs, newer technologies to minimize costs, produce real-time results, and improve both the efficiency and ease of cervical cancer screening have gained research traction (Table II).

TABLE II.

Overview of experimental HPV screening tests. LAMP, loop-mediated isothermal amplification.

Name	Biomarker Type	Genotyping	Detection Method	Test Timesa	Point-of-care	Reference
Paperfluidic molecular diagnostic chip	DNA	Detection of HPV 16	LAMP, lateral flow		Yes	48
AIM-HPV	DNA	Detection of HPV 16 and 18	Hybridization, microholography		Yes	44
AmpFire Multiplex HPV Assay	DNA	Identification of HPV 16 and 18, and concurrent detection of the remaining 13 high-risk types (31/33/35/39/45/51/52/53/56/58/59/66/68)	Isothermal amplification, fluorescence		Yes	43
enVision	DNA and RNA	Detection of HPV 16 and 18	enzyme–DNA nanostructures		Yes	42
OncoE6/E7 (8 Type) Test	Onco-protein	Detection of oncoprotein E6 of HPV 16, 18, 31, 35, 45, and oncoprotein E7 of HPV 33, 52, and 58	Sandwich assay, lateral flow		Yes	49
Multiplexed fluorescence platform	Antibody	Detection of antibodies to HPV 16 E7 oncoprotein	Antigen-antibody reaction		Yes	41

^a Estimated single assay time: , 0∼1 h; , 1∼3 h; , 3∼5 h; , > 5 h.

1. Multiplexed fluorescence platform for detecting antibodies to HPV 16 E7

The development of HPV-related cancers is associated with IgG antibodies, primarily to the oncoproteins E6 and E7.³⁹ The antibodies to HPV E7 were more frequently detected in women with invasive cervical cancer (30.3%) than women with CIN 2/3 (19.5%) and CIN 0/1 (6.6%).⁴⁰

A POC multiplexed fluorescence screening platform has been explored for detecting antibodies to HPV 16 E7 oncoprotein in patient plasma.⁴¹ Here, inexpensive interference filters and readout electronics were used to help leverage time integration of output signals for improved accuracy. The setup [Figs. 1(A) and 1(B)] comprises LEDs as excitation sources, 2 × 2 fluorescent labeled biorecognition sites on a microscope slide, charge-integration amplifier readout circuit, and photodiodes as the detectors. For antibody detection, HPV 16 E7 was printed on 3-aminopropylethoxysilane functionalized glass; patient plasma was then incubated with the immobilized protein. HPV 16 E7-specific antibodies in patient plasma samples were captured by the immobilized antigen, and then detected using a secondary antibody conjugated to 1-μm fluorescent microspheres and DyLight549.⁴¹ This hand-held device comprised inexpensive off-the-shelf components thus suitable for resource-limited budgets. On the other hand, the relatively complex workflow and relatively long time-to-result will limit its widespread use unless further streamlining occurs.

FIG. 1.

Multiplexed fluorescence platforms for HPV screening. (A) The schematic of multiplexed fluorescence platform. (B) Some key components and assembly of this platform. Upper-left is the 3D-printed optical assembly; upper-right is a 25-mm filter typically used; lower-left shows components packaged in a 3D-printed enclosure; lower-right is a filter dielectric stack deposited on a substrate. (C) The schematic and photograph of enVision microfluidic platform (Scale bar indicates 1 cm). Figures 1(A) and 1(B) reproduced from Obahiagbon et al., Biosens Bioelectron 117, 153 (2018). Copyright 2018 Elsevier.⁴¹ Figure 1(C) reproduced from Ho et al., Nat Commun 9, 3238 (2018). Copyright 2018 Springer Nature.⁴²

2. Enzyme-assisted nanocomplexes for visual identification of nucleic acids (enVision) platform

A molecular platform named enVision (enzyme-assisted nanocomplexes for visual identification of nucleic acids) purportedly enables visual and modular detection of HPV nucleic acids (both DNA and RNA) without target nucleic acid amplification [Fig. 1(C)].⁴² Detection occurs through three functional steps: target recognition, target-independent signal enhancement, and visual detection. A modified DNA aptamer bound to a Taq DNA polymerase recognizes target nucleic acids. Presence of complementary target DNA enables hybridization and dissociates to activate Taq DNA polymerase activity. The active polymerase target independently elongates a universal nanostructure with biotin-modified deoxynucleotides (dNTPs). Visual signals are enzymatically produced by incorporating horseradish peroxidase (HRP) onto the signaling nanostructures, which could be detected directly by eye and/or quantified through smartphones. This assay was implemented into a configurable microfluidic platform consisting of independent assay cassettes and a common signaling cartridge intended to make the process straightforward and easy to use.⁴²

The enVision platform was further tested for clinical effectiveness on clinical endocervical brush samples by examining HPV 16 and HPV 18 L1 loci. Results were compared with Cobas HPV, considered a conventional gold standard.⁴² High detection accuracies were found with 92.9% sensitivity and 90.5% specificity for HPV 16, and 83.3% sensitivity and 100% specificity for HPV 18. The assay also identified previously undetectable infections when more locus-specific nanostructures (i.e., L1, L2, and E1) were included. Given its minimal equipment setup and reduced costs, the platform may prove useful for visual detection of emerging infectious diseases in resource-low settings. Visual detection minimizes additional equipment needs, but introduces subjectivity; environmental conditions further challenge reproducibility of readouts. For instance, when quantifying through smartphones, varying environmental light intensities could influence results within and across experiments.

3. AmpFire Multiplex HPV Assay

The AmpFire Multiplex HPV Assay developed by Atila Biosystems detects 15 high-risk HPV genotypes while simultaneously genotyping HPV 16 and HPV 18 in a single tube. The multiplex assay uses sequence-specific primers to target HPV genotypes of interest and amplify their respective sequences in an isothermal amplification system. Once amplified, specific molecular beacon probes are bound to the products to create a detectable fluorescence signal. The key component that separates this assay from other commercial ones is that the tests detect HPV in formalin-fixed, paraffin-embedded (FFPE) samples. While other commercial tests require liquid-based samples or clinician-collected cytobrushes, testing using FFPE samples has not been widely explored. In addition, since the assay requires only two microliters of FFPE samples treated with buffers and solutions, sample processing is simplified as there is no need for complex instruments. After treatment, the samples are added to a reaction tube consisting of primer and reaction mixes and incubated in a real-time PCR system that records the fluorescence signals; readouts are generated in a couple of hours.

A recent preclinical and clinical study compared AmpFire multiplexed assay performance to both the Roche Cobas HPV Linear Array HPV assays.⁴³ A total of 214 clinical specimens were procured from FFPE blocks of cervix/vulva and oropharynx tissue samples archived at the Hospital Clínico Universitario Virgen de la Arrixaca in Spain and Memorial Sloan Kettering Cancer Center in New York. Overall, 19 samples were positive for HPV 16, 6 for HPV 18, and 18 for non-16/18 high-risk HPV. The AmpFire multiplex assay detected HPV in 23.7% of 156 cervical tissues and 9.8% of 51 oropharyngeal samples. The positive percent agreements were 100% for both HPV 16 and HPV 18, and 94.4% for non-16/18 high-risk HPV. Meanwhile, negative percent agreements were 100% all across the board. The limit of detection was around 2–20 copies per reaction for each high-risk HPV genotype. Since it takes an additional hour to completely genotype non-16/18 high-risk HPV, researchers are modifying the assay to target additional potential oncogenic HPV subtypes. Analyses of other sample types such as oral rinse or urine samples could extend its potential impact. Generally, the assay minimizes the time and labor needed for sample processing and amplification with its FFPE samples and isothermal amplification method, respectively, thus paving avenues for POC validation across resource-low settings.

4. Artificial Intelligence Monitoring for HPV (AIM-HPV)

In addition to advancing hardware technologies, an Artificial Intelligence Monitoring for HPV (AIM-HPV) platform leverages deep learning tactics to facilitate POC analyses (Fig. 2).⁴⁴ Here, digital microholography readily produces high-quality image data, even at sub-micron levels, through a simple, lens-free optical system. The platform detects target nucleic acids within cervical brushings introduced into a disposable DNA extraction kit. A library of functionalized microbeads binds to HPV 16 and 18 DNA targets to form dimers if present; unique holographic signatures are produced. Circumventing prior need for cloud computing and attendant challenges in LMICs, deep-learning algorithms were developed and directly loaded onto the device for complete on-site analytics.⁴⁴

FIG. 2.

Artificial intelligence monitoring for HPV (AIM-HPV). (A) The workflow of this assay. (B, C) The photograph and schematic of the AIM-HPV device. (D) The diffraction microscopic image and patterns of polystyrene (PS) beads (blue arrow), silica beads (orange), and PS-silica bead dimer (red). Reproduced with permission from Pathania et al., Theranostics 9, 8438–8447 (2019). Copyright 2019 Ivyspring International Publisher.⁴⁴

Specifically, cervical brushings underwent DNA extraction via a disposable syringe filter. Samples were mixed with polystyrene (PS) and silica beads, respectively, coated with DNA probes complementary to the 3' or 5' ends of target HPV DNA. PS-Silica dimers would eventually form in the presence of target HPV DNA. Customized deep learning algorithms readily quantitated PS, silica, and PS-silica dimer signals.⁴⁴

The diagnostic performance of AIM-HPV was explored in 28 patients at Massachusetts General Hospital (MGH) with abnormal pap smear results.⁴⁴ They underwent both cervical biopsy (colposcopy) and concomitant brushings; strong correlations were noted, thus rendering brushings as viable options moving forward. Moreover, readouts between AIM-HPV and Cobas HPV tests, the gold standard at MGH, achieved perfect concordance. Since use in low-resource settings motivated device design, AIM-HPV is currently undergoing field testing.

5. Paperfluidic molecular diagnostic chip

Paperfluidics exploit paper as substrates for analyses, thus offering lower-cost alternatives to lab-on-a-chip microfluidics.^45–47 They offer the common benefits of microfluidics (e.g., small size, portability, minimal volume requirements), while leveraging paper's fluid transport through capillary action in equipment- and electricity-free fashion.

A paperfluidic molecular diagnostic chip has been developed for HPV 16 DNA [(Fig 3(A)].⁴⁸ This chip combines nucleic acid isolation, isothermal amplification, and lateral flow detection of target DNA. Visual readouts through immunochromatography occur in less than 1 h [Figs. 3(B) and 3(C)]. When compared to qPCR, increased false positives rates were noted. If this weakness can be subsequently improved, the technology could be poised for broader testing in LMICs.

FIG. 3.

Paperfluidic HPV POC tests. (A) The photograph of the integrated paperfluidic molecular diagnostic chip. (B, C) An assay result on the integrated paperfluidic molecular diagnostic chip with different copies of HPV 16 DNA. Left line is the test line, and right line is the control line. NTC; no template control. (D) The principle of OncoE6/E7 (8 Type) Test. The result is positive if purple test line(s) can be visualized along with the control lines (line C). One individual test line corresponds to one type of the oncoproteins (HPV 16, 18, 31, 33, 35, 45, 52, and 58 E6/E7). The result of an HPV 16 E6 positive example was showed. (E) The workflow of OncoE6/E7 (8 Type) Test. Figures 3(A), 3(B), and 3(C) reproduced from Rodriguez et al., Lab Chip 16, 753 (2016). Copyright 2016 the Royal Society of Chemistry.⁴⁸ Figures 3(D) and 3(E) reproduced from Rezhake et al., Int. J. Cancer 144, 34 (2019). Copyright 2019 John Wiley & Sons.⁴⁹

6. OncoE6/E7 (8 Type) test

To detect additional high-risk HPV oncoproteins instead of DNA, the OncoE6/E7 (8 Type) test was developed [(Figs. 3(D) and 3(E)] to expand on the prior OncoE6 platform.⁴⁹ Two additional lateral flow strips were included to detect the oncoprotein E6/E7 of eight prevalent HPV types (HPV 16, 18, 31, 33, 35, 45, 52, and 58).

A recent study compared the performance between liquid-based cytology (LBC), DNA-based genotyping, and OncoE6/E7 (8 Type) Test.⁴⁹ Among HPV-positive patients, OncoE6/E7 (8 Type) exhibited lower positivity compared to DNA-based genotyping (8 Type) and LBC (17.4% vs 58.3% and 50.97%, respectively). Strikingly, OncoE6/E7 (8 Type) detected 100% of high-grade lesions (CIN3+) with superior specificity (85.94%). It was less sensitive (67.7%) but more specific (89.5%) for CIN2+ cases, compared to both DNA-based genotyping (8 Type) and LBC. The OncoE6/E7 (8 Type) Test struck a reasonable balance between sensitivity and specificity, rendering further referrals more efficient for women who need them most.⁴⁹

7. Deep learning-based automated visual evaluation of cervical images

In most low-resource regions, HPV vaccination and screening are largely lacking, while the underwhelming accuracy of VIA may lead to over- or under-treatment in practice. In response, a deep learning–based automated visual evaluation algorithm was developed and applied to digitized cervical images to reduce subjective interpretation during clinical exams.⁵⁰

Testing occurred on archived, digitized cervical images from screening of 9406 women followed for 7 years. The automated visual evaluation algorithm located the cervix within input images and evaluated the severity scores that predict a histologic CIN2 or worse (CIN2+) case (Fig. 4). After exploring several different techniques, faster region-based convolutional neural (Faster R-CNN algorithm)⁵¹ performed best in both speed and accuracy. Compared with original cervigram interpretations, automated visual evaluation exhibited greater accuracy.

FIG. 4.

The architecture of the automated visual evaluation algorithm for cervical images. The training and validation included both cervix locator and automated visual evaluation. Reproduced with permission from Hu et al., J. Natl. Cancer Inst. 111, 923 (2019). Copyright 2019 Oxford University Press.⁵⁰

The cases adopted in this method arose from a single cohort study, and the images were captured by several highly trained nurses with outdated film cameras. More images across different cohorts captured with contemporary digital image technologies, (e.g., phone cameras), would be needed to explore real world use case for POC testing. This is addressed in more recently published papers. Xue et al. demonstrated that the automated visual evaluation algorithm could be applied to cervix images taken with a smartphone-based commercial system (MobileODT Eva).⁵² Guo et al. developed a deep-learning algorithm, screening the cervix images' quality and adequate coverage of anatomical areas for the robust use of the automated visual evaluation algorithm.⁵³ These developments show the potential use of a POC system powered by deep-learning algorithms for rapid cervical screening, especially in low- to medium-resource settings.

IV. DISCUSSION

The COVID-19 pandemic has highlighted how societal inequities can disproportionately harm the most vulnerable groups even further. A similar lens can be applied to the long-standing structural barriers precluding access to effective cervical cancer screening. Early intervention can be curative, yet most deaths worldwide occur in disadvantaged settings. The convergence of poverty, tenuous healthcare infrastructures, scarcity of trained pathologists, and shortage of organized screening programs also pose significant barriers. Challenges exist translating the Pap smear, with its attendant requirements for skilled pathology interpretation, into feasible workflows especially for LMICs. In response, more practical methods described above have been explored. Compounding matters, Pap smears are increasingly losing favor to HPV DNA analyses in developed countries evidenced by emerging clinical guidelines. The significant fixed and operating costs of HPV analyses broaden the gap between the “haves” and “have-nots.”

Commercial HPV DNA tests were designed to require less resources, while offering highly precise and accurate detection, for practical utility in underserved settings. Many such tests target DNA from high-risk HPV subtypes, including HPV 16 and HPV 18. Meanwhile, other technologies focus on oncoproteins differentially encoded in the presence of HPV DNA. Turnaround times generally occur within hours; not near real-time yet still actionable within the encounter.

Moving further along the innovation scale, machine learning and nanotechnology-based efforts improve diagnostic readouts through the use of microfluidics and other components that lower test expenses and maintain end-user friendly attributes. Such technologies tend to emphasize antibody or probe binding to target HPV nucleic acids. Resulting images are driven by machine and deep learning algorithms that detect the expression of these hybrids. Such algorithms are often developed within resource-rich laboratories using local clinical specimens or cell lines — both historically derived from patients with ready access to clinical care. Moreover, such algorithms don't often incorporate other specific clinical or environmental variables in order to better generalize across populations. However, such tactics seemingly under-appreciate the additional complexity and interplay between environment and physiology. For example, the higher prevalence of HIV in sub-Saharan Africa could influence genotypic and phenotypic readouts on patient-derived specimens. Hence, deep learning approaches divorced from such broader but relevant variables may prove less reliable or actionable. As such, we contend the integration of machine or deep learning need to occur in partnerships with the intended populations of interest. As an added benefit, this would help investigators empower the individuals or populations they seek to understand.

By integrating these technologies into cervical cancer screening programs, healthcare officials and researchers are working together to eliminate potential barriers preventing women from accessing to them. Cervical self-sampling, a demonstrated feasible tactic, eases any tension or uneasiness that might arise during the collection process while circumventing transportation challenges to centralized locations. Here, the intersection between nano-oncology, bioengineering, and data sciences could offer patients additional innovative avenues for self-collection and analyses of samples. Integrating HPV analyses instead of cytology would position patients into the state-of-the-art screening space alongside their more privileged counterparts.

AUTHORS' CONTRIBUTIONS

Z.Y. and J.F. contributed equally to this work.

DATA AVAILABILITY

Data sharing is not applicable to this article as no new data were created or analyzed in this study.