Differentiating low tumor burden from oligometastatic disease in colorectal cancer: a call for individualized therapeutic approaches

Abstract

Metastatic colorectal cancer (mCRC) remains a major clinical challenge; however, tumor burden significantly influences treatment outcomes. In this review, we explore the biological and clinical relevance of low tumor burden (LTB) in mCRC. The primary challenge in defining LTB mCRC lies in establishing a standardized definition that extends beyond the current focus on oligometastatic disease. Patients with LTB mCRC exhibit distinct clinical characteristics that may impact both prognosis and therapeutic response. Evidence suggests that LTB patients often respond better to systemic therapies and may derive potential benefits from targeted and immunotherapy approaches. However, establishing a clear definition is crucial for consistent patient stratification, and for guiding research and selecting the most appropriate therapeutic strategies, particularly in the context of emerging treatments such as immunotherapy. Recent studies using advanced imaging modalities, liquid biopsies, and lactate dehydrogenase (LDH) measurements offer novel approaches to evaluate tumor burden more accurately. These developments, coupled with emerging evidence that patients with LTB may benefit from immunotherapy, highlight the need for further research focused on LTB mCRC patients. Additionally, artificial intelligence (AI) could enhance tumor detection, automate three-dimensional (3D) volume quantification, extract radiomics-based prognostic information, and integrate multimodal data. These capabilities may enhance our ability to stratify patients and guide treatment decisions, potentially leading to better outcomes for mCRC patients. Future studies should focus on refining the definition of LTB, validating these new assessment techniques, and evaluating their impact on treatment outcomes in mCRC patients.

HIGHLIGHTS

• •LTB mCRC exhibits distinct clinical characteristics that may influence prognosis and therapeutic response.
• •A clear definition is key for patient stratification and choosing the most appropriate therapeutic strategies.
• •Advanced imaging modalities, liquid biopsies, and LDH measurements offer novel approaches for evaluating tumor burden.
• •AI may enhance 3D volume quantification, extract radiomics-based prognostic information, and integrate multimodal data.
• •Future studies should refine the LTB definition, validate assessments, and evaluate impact on mCRC treatment outcomes.

Introduction

The ‘Hallmarks of cancer’ comprise the acquired biological capabilities by human cells when transitioning from the state of normalcy to malignancy. In its current definition, eight characteristics are depicted as core hallmarks, including evading growth suppressors, evading apoptosis, unlimited replicative potential, sustained angiogenesis, activating tissue invasion and metastasis, reprogramming cellular metabolism, and avoiding immune destruction.¹

Metastatic capability can be defined as the ability of cancer cells to physically disseminate from the primary tumor to distant tissues, and our understanding of this phenomenon has progressively evolved. In the early 20th century, Halsted proposed a model in which the spread of breast cancer cells followed a precise, contiguous fashion: from the primary tumor to the lymphatic structures, and finally to distant sites.² For many years, patients diagnosed with metastatic cancer were immediately and irrevocably considered incurable, as widespread disease was traditionally and uniformly associated with a poor prognosis. However, this perspective began to shift when Hellman and Weichselbaum introduced the concept of oligometastatic disease (OMD).³ OMD can be interpreted as an intermediate entity between localized tumors and widespread systemic disease, in which tumor cells exhibit a restricted capacity for metastasis, resulting in a limited number and sites of metastases. This model indirectly implies that, for certain patients, radical treatment with curative intent may be considered even in the presence of confirmed distant metastases.⁴ Subsequent advancements in stratification strategies, considering both patient characteristics and tumor features, have shifted this paradigm. Recent research has demonstrated that a subset of patients with OMD may indeed have the potential for curative treatment. Improved imaging techniques and molecular profiling have enabled more precise identification of these patients, opening the possibility for targeted interventions such as surgery, stereotactic body radiotherapy, or systemic therapies, which can achieve prolonged survival and even cure in selected cases.

Regarding colorectal cancer (CRC), 20% of patients present with metastatic disease at diagnosis, while 25% of those with early-stage CRC will experience recurrence after initial radical treatment.⁵ A proportion of patients present with metastases confined to a single organ or a few organs, and if complete surgical resection (R0) is feasible, either at presentation or after systemic therapy, it may improve outcomes and be potentially curative, with 5-year survival rates ranging from 20% to 45%.6, 7, 8, 9, 10 Currently, patients may be considered for both surgery and locoregional treatments based not only on the technical feasibility of achieving complete eradication, but also on disease- and patient-related factors incorporated into the decision-making process.11, 12, 13, 14 In the context of CRC, metastatic sites such as the liver, lung, peritoneum, lymph nodes, and ovary may be considered for surgical or locoregional approaches.^7,12,15,16 For cases ineligible for curative surgery or locoregional interventions, assessing disease burden at diagnosis is crucial, as it may influence therapeutic choices, treatment response, and the associated toxicity profile.

The purpose of this review is to propose the existence of low tumor burden (LTB) disease as a concept distinct from OMD in metastatic CRC (mCRC), with its own prognostic and therapeutic implications. We aim to discuss the challenges in defining and managing this intermediate state, explore opportunities arising from advances in tumor burden assessment, and examine the potential role of immunotherapy in patients with microsatellite stability (MSS) LTB CRC. In addition, we consider the future role of artificial intelligence (AI) in enhancing tumor burden assessment and management. This review aims to integrate current knowledge and highlight ongoing research, providing insights into potential strategies to improve outcomes in patients with mCRC.

The Challenge: the consensus on a definition

In the evolving landscape of mCRC management, defining the boundaries of ‘LTB’ and distinguishing it from OMD constitutes a critical challenge. While OMD emphasizes the potential for complete resection or locoregional treatment, the definition of LTB should not rely solely on the feasibility of radical interventions. To make the distinction between the two conditions more practical, we propose that OMD refers to potentially resectable mCRC, whereas within the context of unresectable mCRC, LTB should be identified as a distinct entity with specific clinical and radiological features (Figure 1).

Figure 1.

Schematic representation of different concepts of metastatic disease, illustrating the evolution from the original definition of oligometastatic disease to the current distinction between oligometastatic and low tumor burden.

The most recent ESMO guidelines for the treatment of mCRC¹⁷ define OMD as fulfilling all the following criteria: (i) one to five metastatic lesions, occasionally more if complete eradication is feasible, (ii) involvement of up to two metastatic sites, (iii) a controlled primary tumor (optionally resected), and (iv) all metastatic sites must be safely treatable with locoregional therapy. Notably, the previous ESMO recommendations, published in 2016, excluded from the OMD classification cases with lesions affecting multiple bones and/or brain metastases, whereas the most recent guidelines no longer differentiate based on the site of metastases.¹⁸ Furthermore, the oligometastatic condition can be observed at different points in the disease course (Figure 2). The European Society for Radiotherapy and Oncology–European Organisation for Research and Treatment of Cancer (ESTRO–EORTC) consensus classification differentiates between ‘induced OMD’ (i.e. cases with a previous history of polymetastatic disease) and ‘genuine OMD’ (i.e. those with no such previous history). Genuine OMD can be further classified into de novo and repeat OMD.¹⁹ De novo OMD may be subclassified into synchronous or metachronous disease. Ultimately, the classification includes three additional subtypes: oligorecurrence, oligoprogression, and oligopersistence.

Figure 2.

Consensus subclassification of oligometastatic disease (OMD). The oligometastatic state may occur at different points during the course of the disease.

We could speculate that the various scenarios classified as OMD may, in fact, represent distinct disease entities, each with unique biological characteristics and treatment response patterns, rather than falling under a single umbrella term. However, we must acknowledge that the current definition of OMD is primarily based on the number and location of metastases, without considering their volume, shape, or heterogeneity. In addition, existing OMD criteria do not incorporate histopathological, biological, or patient-related factors that could significantly influence tumor characterization and prognosis. These limitations highlight the need for a more comprehensive approach to categorizing patients with limited metastatic disease.

While the current definition of OMD focuses on the possibility of achieving eradication of all metastases and/or a status of ‘no evidence of disease’, LTB refers to a state of metastatic disease that is not necessarily resectable but in which the total volume of all lesions is low. However, defining the threshold for what constitutes low remains a challenge. Despite a growing body of literature on this phenomenon,20, 21, 22 a precise definition for LTB is still lacking. Table 1 presents a practical classification proposal, in which the thresholds of 3 cm as the maximum lesion diameter and three metastatic sites are arbitrary and warrant further investigation. Nonetheless, these cut-offs reflect those used in other studies evaluating OMD and/or diseases with more favorable prognostic characteristics.23, 24, 25, 26, 27

Table 1.

Proposal for a practical distinction between oligometastatic colorectal cancer and low tumor burden colorectal cancer

	Oligometastatic disease in CRC	Low tumor burden in CRC
Definition	•One to five metastatic lesions, occasionally more (if LTs are feasible) •Up to two metastatic sites •All metastatic sites must be safely treatable by locoregional treatment	•Total size of lesions is small (≤3 cm maximum diameter) •No formal limit of number of lesions •Ideally no more than three sites of metastases •Metastasis without clinical impact (nor symptoms or alterations in biochemistry)
Primary tumor	Controlled (optionally resected)	Not necessarily controlled (if synchronous metastases and symptomatic patient, then evaluate palliative treatment of the primary tumor)
Therapeutic intervention aim	Curative (nonevidence of disease)	Disease control at first (depending on the response to systemic treatment, evaluate surgery or LTs)
Presentation at diagnosis of metastatic disease	Resectable or borderline resectable disease	Unresectable disease
Role of FDG-PET	Rule out other sites of metastasis not detected by primary staging methods (computed tomography scan or magnetic resonance imaging)	Evaluate the metabolic activity of metastases and their biological aggressiveness
Timing of metastatic disease presentation	•De novo OMD •After a ‘free therapeutic intervention’ interval of at least 6 months	LTB can refer to the disease state at initial diagnosis or may be achieved following a locoregional treatment (e.g. targeting liver metastases) aimed at enhancing the response to ICIs.
Serological markers	LDH and ctDNA VAF should not be considered as criteria to exclude patients from curative interventions	LDH and ctDNA VAF values may be useful tools in the assessment of LTB. No detection of ctDNA or maximum VAF < 10% are more likely to be characteristic of LTB

CRC, colorectal cancer; ctDNA, circulating tumor DNA; FDG, fluorodeoxyglucose; ICI, immune checkpoint inhibitor; LT, locoregional treatment; LTB, low tumor burden; OMD, oligometastatic disease; VAF, variant allele frequency; PET, positron emission tomography.

Santorsola et al.²⁸ reviewed the reporting of tumor burden in phase III clinical trials of metastatic disease, including patients with non-small-cell lung cancer, breast and CRC. They found that only 20 out of 70 (28.6%) trials defined low burden disease, and just 18 (25.7%) trials used disease extent as a stratification factor during randomization. The definition of LTB was mostly based on the number of involved organs, without considering other relevant factors of disease extent. A more comprehensive assessment of tumor burden should consider not only the number and size of metastases but also the extent and grade of organ involvement. For example, by considering only the number of affected organs, we fail to distinguish between a single large metastatic lesion and multiple smaller metastases in the liver, scenarios that may have distinct biological behaviors and clinical implications. Moreover, the specific organs involved can impact clinical outcomes differently. These findings highlight the need for a more standardized definition of LTB.

Despite the lack of a universally accepted definition, tumor burden has been considered in different clinical scenarios across multiple cancer types. Consistently, LTB, measured at different timepoints, has been associated with better outcomes.29, 30, 31, 32 In the context of resectable mCRC, a tumor burden score developed for patients undergoing curative-intent surgery for CRC liver metastases combined both tumor size and the total number of lesions.³³ This score, defined as the distance from the origin on a Cartesian plane incorporating maximum tumor size (x-axis) and number of liver lesions (y-axis), effectively identified distinct prognostic groups, with higher scores correlating with worse long-term outcomes. Similarly, a scoring system for mCRC patients, including both resectable and nonresectable cases, considered both objective factors (e.g. size and number of metastases) and subjective factors (e.g. clinical judgment).³⁴ This study showed that median overall survival (OS) was superior in the LTB group based on both subjective and objective scoring criteria.

In the context of unresectable mCRC, a pooled analysis of the TRIBE and TRIBE2 trials investigated the prognostic impact of metastatic spread at baseline by distinguishing OMD from non-OMD patients.²⁷ Within the OMD group, the researchers further stratified patients based on tumor burden, classifying them as ‘LTB OMD’ (defined as no more than three metastases per organ, each with a maximum size of 3 cm) and ‘non-LTB OMD’. While patients with OMD had longer progression-free survival and OS compared with non-OMD, the difference between low and non-LTB OMD groups was not statistically significant, although LTB was associated with a trend toward improved OS. The distinction between low and non-LTB OMD, although guided by criteria commonly used to select patients for locoregional treatments, seems somewhat arbitrary.

In our proposed definition of LTB, we refer to a condition that, compared with unresectable non-LTB mCRC, is characterized by a better prognosis due to slower tumor growth kinetics and a metastatic volume that does not result in clinical symptoms or biochemical alterations. However, while we do not consider LTB to be a surrogate for ‘indolent disease’, we believe it should reflect the total amount of tumor lesions (even though a definitive threshold has not yet been established). We do not, however, include clinical parameters such as ‘treatment response’ or ‘molecular profile’ as defining criteria for LTB. Nonetheless, assuming a less aggressive disease course, we hypothesize that once these patients are identified and chemosensitivity is confirmed through initial clinical and radiological assessments, they may benefit from individualized therapeutic strategies, such as early chemotherapy deintensification or chemotherapy holidays. The absence of clear, standardized criteria for defining LTB in mCRC remains a significant challenge and warrants further research.

The opportunities: how to measure tumor burden

Morphological imaging

Imaging remains the gold standard for assessing cancer spread. Chest–abdomen–pelvis computed tomography (CT) scans are most commonly used due to their speed, cost-effectiveness, and wide availability. However, for liver metastases, magnetic resonance imaging (MRI) is preferred because of its superior soft tissue contrast.

Regardless of the imaging modality used, delineating the boundaries of metastases is essential for calculating their volume. However, manual segmentation of radiological images is labor-intensive and not feasible for routine clinical use. Traditionally, expert radiologists manually identify tumors and measure their largest axial diameter. While volume can be estimated from these two-dimensional (2D) axial measurements, this method relies on the unrealistic assumption that all metastases are perfect spheres, which can result in significant volume calculation errors, particularly for irregularly shaped lesions. A study by Rothe et al. compared different measurement methods for liver metastases and revealed significant discrepancies in tumor response classification between 2D and three-dimensional (3D) assessments.³⁵ The authors also reported that 3D volumetric measurements demonstrated significantly higher inter-rater reproducibility than 2D assessments. Another study showed that 3D volumetric measurements of CRC liver metastases are more sensitive in detecting early treatment response compared with traditional 2D measurements.³⁶

Recent advancements in AI in oncology offer promising solutions to overcome these limitations. AI-assisted detection and delineation of metastases could significantly enhance radiologists’ ability to identify and outline tumors more accurately and efficiently.³⁷ The potential of AI to improve tumor burden measurement is discussed in greater detail in a dedicated section later in this review.

Metabolic imaging

Beyond morphological evaluation, analysis of tumor metabolic activity using fluorine-18 fluorodeoxyglucose (¹⁸F-FDG) positron emission tomography (PET)/CT may aid in defining tumor burden, as it provides information on increased glucose uptake and glycolysis by tumor cells, thereby offering insights into cellular metabolism and biology.³⁸ A study aimed at validating the prognostic value of ¹⁸F-FDG PET/CT-based biomarkers, particularly baseline whole-body metabolically active tumor volume (WB-MATV) and early metabolic response in mCRC, found that patients with high baseline WB-MATV, defined as the sum of metabolically active volumes of all target lesions >100 cm³, had worse median progression-free survival and OS, both in first-line settings and in patients with chemotherapy-refractory disease.³⁹

Serological markers

Other indirect methods to measure tumor burden are liquid biopsy and lactate dehydrogenase (LDH) levels. Liquid biopsy refers to the detection of tumor cells or fragments of tumor DNA in body fluids, such as blood, urine, saliva, and cerebrospinal fluid.⁴⁰ Both the primary tumor and metastases can release circulating tumor DNA (ctDNA) into the bloodstream,⁴¹ which has a short half-life, ranging from a few minutes to 2 h, making it a potentially useful real-time marker of tumor dynamics.⁴² ctDNA is a noninvasive tool for the dynamic monitoring of neoplasia, as its rapid clearance supports its use in tracking changes in tumor burden. Moreover, several studies have shown that ctDNA levels positively correlate with tumor burden in solid cancers.43, 44, 45

In patients with metastatic melanoma, a correlation has been observed between plasma levels of ctDNA and metabolic tumor burden measured by ¹⁸F-FDG PET, with ctDNA levels also correlating with treatment response.^46,47 In the mCRC setting, patients with high ctDNA, measured by the variant allele fraction (VAF) of clonal gene alterations detected in ctDNA, tend to have worse outcomes compared with those with low VAF.^48,49 A key factor underlying the association between high VAF and poorer clinical outcomes may be its link to tumor burden, as VAF in the bloodstream reflects the quantity of ctDNA released by tumor cells. Moreover, it has been suggested that liquid biopsy may complement radiological response criteria in evaluating treatment response. However, caution is warranted when interpreting liquid biopsy results, as a negative result (i.e. the absence of detectable circulating mutations) may stem from technical limitations or from the reduced tendency of certain tumor sites to release DNA fragments into the bloodstream.^50,51 In fact, ctDNA release and clearance is a more complex process influenced by multiple factors, such as tumor necrosis rate, vascularity, location and stage of the tumor, each of which may be affect the tumor’s shedding capability.^50,52 For instance, patients with primary tumors located in the brain or metastatic cancers of the kidney, prostate, or thyroid have demonstrated lower rates of ctDNA detection.⁵⁰ A study analyzing the concordance between liquid and tissue biopsies for RAS status in colorectal cancer with a single metastatic site found the highest concordance rates (≥90%) for liver metastases alone, regardless of size and number, as well as for peritoneal and lung metastases alone, provided the lesions met certain thresholds—specifically, size ≥20 mm in longest diameter and/or ≥10 lesions. These findings supports the idea that liquid biopsy results should be interpreted with caution in the presence of peritoneal or lung metastases. Moreover, in the same study, the mean mutation allele fraction (MAF) for liver, peritoneal, and lung metastases was 6.8%, 7.2%, and 2.6%, respectively, indicating a tendency toward lower MAF in peritoneal disease.⁵³ Furthermore, the method of analysis plays a critical role, primarily involving next-generation sequencing or PCR, and the respective lower limits of detection for mutant DNA should also be taken into account. By contrast, liquid biopsy may occasionally detect mutations not derived from the tumor, but instead from phenomena such as clonal hematopoiesis.⁵¹ Taken together, these data suggest that, despite the promise of ctDNA analysis, it likely needs to be integrated with other radiological and/or metabolic methods to achieve a more accurate assessment of tumor burden.

Another serological potential biomarker is LDH, an enzyme involved in cellular energy production by converting pyruvate to lactate. Malignant tumors often exhibit a shift in glucose metabolism (the Warburg effect), whereby glucose is metabolized regardless of oxygen availability, supporting cancer cell proliferation.^54,55 Thus, LDH hyperactivation may promote tumor growth and progression by supplying energy under hypoxic conditions.⁵⁶ Elevated serum LDH levels are associated with poor outcomes in several cancer types, particularly melanoma, prostatic cancer, and renal cell carcinoma.^57,58 Its negative prognostic role has also been confirmed in mCRC, where abnormal baseline LDH levels correlate with worse outcomes.^59,60 In a trial evaluating FOLFOX + vatalanib or placebo as first-line treatment for mCRC, serum LDH levels were significantly associated with tumor burden, measured by the sum of maximum assessable tumor diameters. High serum LDH levels (≥1.5 times the upper limit of normal) were more frequently observed in patients with Eastern Cooperative Oncology Group Performance Status (ECOG PS) 1-2 and in female patients.⁶¹

Hence, LDH can serve as a complementary biomarker but cannot replace more advanced imaging techniques. Rather, it may assist in the interpretation of more complex clinical and radiological findings.

The potential: immunotherapeutic options in patients with low tumor burden CRC

Immunotherapy has marked the beginning of a new era in the management of CRC. However, while mismatch repair deficiency and high microsatellite instability have been recognized as predictive biomarkers for cancer immunotherapy, it has failed to demonstrate clinical benefit in patients with CRC with proficient mismatch repair (pMMR), which are frequently poorly infiltrated by T cells, enriched with myeloid-derived suppressor cells, and exhibit low neoantigen enrichment.^62,63

Preclinical and clinical evidence suggests that tumor volume at the time of therapeutic intervention negatively correlates with the activity of immune checkpoint inhibitors (ICIs).⁶⁴ Preclinical studies have shown that ICIs exhibit greater activity in mice with smaller tumors. In vivo murine models using CT26 colon carcinoma models, which are sensitive to OX40 (a T effector lymphocyte stimulator) have shown larger tumors require the addition of transforming growth factor-β (TGF-β) receptor antagonists to further reduce intratumoral immunosuppression.⁶⁵ By contrast, emerging evidence also indicates that tumor size correlates with sensitivity to ICIs in various solid cancer types, such as metastatic non-small-cell lung cancer and melanoma,^66,67 in which better outcomes in terms of OS and lower risk of hyperprogressive disease have been reported.67, 68, 69 Interestingly, a correlation between ctDNA levels (a potential surrogate for tumor burden, as previously discussed) and treatment outcomes has also been observed in patients with advanced melanoma treated with ICIs. Notably, the difference in OS according to ctDNA levels was smaller in patients receiving combination ICI therapy compared with monotherapy, suggesting that patients with higher levels of ctDNA levels might derive greater benefit from dual-agent treatment.⁷⁰

In addition, the response to immunotherapy in patients with MSS or pMMR CRC is influenced by the site of metastases. In fact, liver metastases have been reported to poorly respond to ICIs in many solid tumors, including mCRC (Table 2).71, 72, 73, 74, 75 This limited response may be attributed to the liver’s immunoregulatory role and the unique composition of its tumor microenvironment, which is characterized by a reduced number of CD8⁺ T cells—potentially due to hepatic siphoning of immune cells.^76,77

Table 2.

Clinical trials investigating the activity of ICIs based on the presence of liver metastases in refractory pMMR/MSS mCRC

Trial registration	Trial type	Population	Immunotherapy agents	Outcomes of clinical trials testing ICI-based combinations in patients with NLM versus patients with LM
NCT04126733 (https://clinicaltrials.gov/ct2/show/NCT04126733)	Phase II	Refractory pMMR/MSS mCRC	Regorafenib + nivolumab	ORR: 21% versus 0% 40-week DCR: 56.5% versus 29.8% mPFS: 15 versus 8 weeks
C-800-01 (NCT03860272; https://clinicaltrials.gov/ct2/show/NCT03860272)	Phase I	Refractory pMMR/MSS mCRC	Botensilimab +/− balstilimab	ORR: 22%a versus 0% DCR: 73% versus 25% mPFS: 4.1 versus 1.4 months
CCTG CO.2692	Phase II	Refractory mCRC unselected for MSI status	Durvalumab + tremelimumab	mDCR: 49% versus 10% mPFS: 2.0 versus 1.82 months
LEAP-01793	Phase III	Refractory pMMR/MSS mCRC	Pembrolizumab + lenvatinib	The trial was negative for the primary endpoint (OS), but data suggest that patients with NLM could benefit more from pembrolizumab + lenvatinib
NCT03539822 (CRC cohort; https://clinicaltrials.gov/ct2/show/NCT03539822)	Phase II	Refractory pMMR/MSS mCRC	Cabozantinib + durvalumab	In the overall mCRC population: ORR: 27.6%. DCR: 86.2%. mPFS 3.7 months The authors supported the activity of cabozantinib + durvalumab regardless of the presence of LM, as 4/6 patients with confirmed response had LM.
MoTriColor Consortium CT394	Phase II	mCRC progressed to more than one chemotherapy line classified MSI-like by genomic expression signature	Atezolizumab + bevacizumab	ORR: 15% versus 0% (among patients with MSS mCRC by standard diagnostic assays)
NCT04362839 (https://clinicaltrials.gov/ct2/show/NCT04362839)	Phase I	Refractory pMMR/MSS mCRC	Ipilimumab + nivolumab + regorafenib	ORR: 36.4% versus 0% mPFS: 5 versus 2 months
Wang et al.95	Retrospective	Refractory pMMR/MSS mCRC	PD-1 or PD-L1 inhibitor	ORR: 19.5% versus 0% DCR: 58.5% versus 1.9% mPFS: 4.0 versus 1.5 months

DCR, disease control rate; ICI, immune checkpoint inhibitor; LM, liver metastasis; mCRC, metastatic colorectal cancer; mDCR, median disease control rate; mPFS, median progression-free survival; MSI, microsatellite instability; MSS, microsatellite stable; NLM, nonliver metastasis; ORR, overall response rate; OS, overall survival; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1; pMMR, proficient mismatch repair.

^aAmong the 77 patients without LMs, 16 had previously treated LMs and 2 of 16 experienced partial response.

Unfortunately, clinical trials cohorts testing immunotherapy-based strategies in pMMR/MSS are enriched with heavily pretreated patients, many of whom present with liver metastases, what may overshadow the potential beneficial effects of these strategies.^78,79 In fact, it has been proposed that early-stage tumors exhibit higher T-cell infiltration and a lower degree of immune suppression, whereas colorectal tumors become progressively immune-excluded during outgrowth.⁸⁰ This forms the scientific rationale behind the pathological responses observed in the NEST-1 trial, which evaluated with neoadjuvant botensilimab and balstilimab in patients with MSS/pMMR localized colon cancer. Although not yet sufficiently investigated, this could support the hypothesis that patients with LTB mCRC may have a less immune-suppressed tumor microenvironment, making them more likely to experience better outcomes with the use of immune-based strategies.

Taken together, these findings are noteworthy not only because they support stratifying patients enrolled in trials based on the site of metastasis and disease burden, but also because they raise the question of whether liver metastases should be surgical removed or treated with other locoregional therapies as a therapeutical strategy to improve response to ICIs, even in pMMR/MSS mCRC.⁸¹ Moreover, considering the potential positive impact of LTB on outcomes and therapeutic response, the development of immune-based strategies in patients with low-burden mCRC should be encouraged. In line with this, the phase I LCB-2301-001 clinical trial is evaluating NILK-2301, a bispecific antibody targeting CEACAM5 and CD3, in patients with refractory pMMR/MSS mCRC with LTB (EuCT registration number 2023-508058-24-01).

The future: AI for improved low tumor volume stratification in colon cancer

We can therefore describe LTB as a clinical entity that defines a specific group of patients with a unique pattern of tumor spread and distinct prognostic and biological characteristics. In this setting, identifying patients with LTB mCRC relies heavily on tumor volume assessment, primarily through radiological techniques. In the era of AI-enhanced cancer care, the definition of low tumor volume in cancer patients is likely to be redefined (Figure 3). AI applications can improve image quality for improved tumor detection, automate 3D volume quantification, and extract additional prognostic information from imaging data through radiomics. Moreover, AI enables the integration of diverse data sources, allowing for multimodal analyses that could refine the identification of LTB patients and enhance prognostic assessments. This, in turn, may support more robust studies to determine clinically relevant volume thresholds and improve patient stratification for targeted therapies, particularly in the context of emerging immunotherapeutic approaches for CRC.

Figure 3.

Artificial intelligence (AI) for enhanced low tumor volume assessment and prognosis. Current AI applications can improve image quality for better tumor detection, automate 3D volume quantification, extract prognostic information from images through radiomics, and leverage data from different sources to refine low tumor burden patient identification and prognostic evaluation.

Enhancing detection and tumor volume quantification with AI

AI applications can enhance the quality of standard radiological scans, ultimately improving tumor detection.⁸² For instance, generative adversarial networks can be utilized to enhance CT images by integrating information from more accurate, yet less accessible, MRI scans.⁸³ This augmentation not only improves image quality but also increases the sensitivity of imaging modalities to detect subtle, early tumor characteristics that may otherwise go unnoticed.

In addition to improving image quality, AI-based tools facilitate fully automated detection and delineation of tumors, accelerating what traditionally has been a complex and time-consuming task. By automating these processes, we can obtain an accurate representation of disease extent in a clinically feasible and timely manner. This enhanced capability may support the collection of prognostic information and inform more precise decisions regarding the need for extensive surgical interventions, intensified local therapies, or systemic treatments. Furthermore, tools such as TotalSegmentator⁸⁴ enable automatic quantification of whole organ volumes, including the liver. When combined with automated tumor volume quantification, this allows for the calculation of liver tumor ratios. As a result, patient selection for liver resection becomes more accurate and easier to implement.

Radiomics: extracting prognostic information from imaging

Beyond tumor volume, radiomics offers a means of extracting potential prognostic information from medical images. This rapidly evolving field applies advanced computational methods to derive quantitative features from imaging modalities such as CT, MRI, and PET scans. The fundamental concept of radiomics is that medical images are rich in quantitative features that reflect underlying tumor biology, extending far beyond what is apparent through standard radiological assessment.⁸⁵ The radiomics workflow typically involves image acquisition, preprocessing, feature extraction, and model development.^86,87 Feature extraction can be carried out using predefined mathematical formulas (handcrafted radiomics) or through automated learned by neural networks (deep radiomics). While deep radiomics may capture more complex patterns, handcrafted radiomics offer higher interpretability.

In the context of mCRC, several studies have demonstrated the prognostic value of radiomics.⁸⁸ A study using the TRIBE2 trial data developed a CT-based radiomics model able to predict OS (area under the curve of 0.83). The authors found that specific radiomic features representing tumor heterogeneity correlated with shorter survival in patients with liver-limited metastases.⁸⁹ Supporting these findings, another large-scale study across two phase III trials showed that a radiomics signature significantly outperformed traditional response evaluation criteria (RECIST version 1.1)⁹⁰ in prognostic stratification, particularly in refining prognosis for patients classified as having stable disease by conventional criteria.⁹¹

Conclusion

LTB in mCRC should be considered a distinct clinical entity, characterized by a complex landscape of both challenges and opportunities. The primary challenge lies in establishing a standardized definition that extends beyond the current focus on OMD. Defining LTB clearly is essential for consistent patient stratification, for guiding the study and selection of the most appropriate therapeutic strategies, particularly in the context of emerging treatments such as immunotherapy, and for enabling accurate prognostic assessment and personalized clinical management.

Recent studies using advanced imaging modalities, liquid biopsies, and LDH measurements may contribute to more accurate identification and characterization of tumor burden. However, the main limitation remains the absence of validated cut-off values, both radiological and serological. In this context, AI offers a powerful tool to enhance our ability to stratify patients and optimize treatment decisions, through capabilities such as tumor detection, automated 3D volume quantification, radiomics-based prognostic data extraction, and multimodal data integration. Moreover, digital pathology studies may improve our understanding of histological patterns and reveal differences in the composition of the immune compartment that underlie distinct biological behaviors.

In the future, oncologists could benefit from a multimodal AI system that integrates volumetric tumor measurements, radiomic features reflecting tumor heterogeneity, and serological biomarkers to generate a more comprehensive profile of a patient’s disease. This approach could help refine the definition of LTB in CRC and identify patients who may benefit from novel treatments, including immunotherapy. However, these multimodal models for improved CRC stratification remain understudied, representing both a significant gap in current research and an exciting opportunity for future exploration.