Colorectal cancer liver metastases: A radiologic point of view

Abstract

Colorectal cancer (CRC) remains one of the leading causes of cancer-related mortality worldwide. Despite advances in early detection and treatment, approximately half of patients with CRC develop liver metastases (LM), complicating therapeutic strategies and reducing survival rates. Radiological imaging is critical in managing colorectal LM by guiding detection, staging, treatment planning, and response evaluation. This letter to the editor provides a comprehensive overview of both traditional and emerging imaging modalities, including computed tomography, magnetic resonance imaging, and positron emission tomography, and their specific roles in clinical decision-making. It further explores advanced techniques such as radiomics, artificial intelligence, and radiogenomics, which integrate quantitative imaging features with genetic and clinical data to enhance prognostication and tailor personalized treatment approaches. Specific examples of how these innovations are applied in treatment response assessment and pre-surgical planning are highlighted. The discussion also emphasizes the need for large-scale prospective clinical trials and standardized protocols to validate current predictive models and fully integrate these advanced methodologies into clinical practice.

Key Words: Colorectal cancer; Liver metastases; Radiology; Radiomics; Artificial intelligence; Radiogenomics; Precision oncology

Core Tip: This letter underscores the pivotal role of radiological imaging in managing colorectal liver metastases (CRLM). It details advanced imaging techniques, radiomics, artificial intelligence, and radiogenomics, illustrating how integrating imaging biomarkers with genomic and clinical data enhances detection, treatment planning, and prognostication. Such innovations pave the way for precision oncology, enabling more personalized and effective therapeutic strategies for CRLM patients.

TO THE EDITOR

Colorectal cancer (CRC) is one of the leading causes of cancer-related mortality globally[1-4]. Despite advances in early detection and treatment, approximately 50% of patients with CRC develop liver metastases (LM), which significantly impacts survival and complicates treatment options[5-8].

As the liver is often the first and sometimes only metastatic site due to direct venous drainage from the gastrointestinal tract, radiological imaging plays a pivotal role in managing colorectal LM (CRLM), from detection and staging to treatment planning and monitoring treatment response[9-12].

In this editorial, we provide an overview of the key subtopics that will be discussed: the role of imaging modalities in detection and staging; advanced radiological techniques for treatment planning and response evaluation; the application of radiomics and machine learning in CRLM management with specific examples; the integration of artificial intelligence (AI) and radiogenomics for prognostication and treatment personalization; and finally, the use of imaging in pre-surgical planning and surveillance. This structured summary is intended to help readers follow the discussion and understand how each principle is specifically applied in CRLM.

The role of imaging modalities in detection and staging

Accurate detection and staging of LM are fundamental for determining treatment pathways. Contrast-enhanced computed tomography (CT) and magnetic resonance imaging (MRI) with iodinated or intracellular (hepatocyte-specific) contrast agents are the most commonly used imaging modalities[13,14]. CT is often the initial modality due to its accessibility and speed, making it valuable for staging and for assessing distant metastasis. However, MRI has a higher sensitivity and specificity for detecting smaller, less vascular liver lesions, which can remain elusive on CT[15-17]. Thanks to its unique multiparametric capabilities, such as diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) mapping, MRI can reveal subtle changes in tumour microenvironment, providing a more nuanced assessment of metastatic lesion[18].

Positron emission tomography (PET)-CT and PET/MRI provide additional layers of information by assessing metabolic activity. While not all CRLM lesions are FDG-avid, PET imaging complements CT and MRI by detecting occult extrahepatic metastases, which may alter therapeutic plans[19-22].

Advanced radiological techniques for treatment planning and response evaluation

The assessment of therapeutic response in CRLM has traditionally relied on size-based criteria, such as response evaluation criteria in solid tumours. However, as treatments like chemotherapy, targeted therapy (anti- vascular endothelial grow factor) and radiofrequency ablation may induce changes in the vascularization of metastases or necrosis without immediate tumour shrinkage, these criteria do not always correlate with therapeutic efficacy[23,24]. In this context, MRI offers significant advantages; techniques such as DWI and ADC mapping provide detailed insights into changes in cellular density and can identify necrotic regions within lesions, thereby serving as reliable indicators of treatment response even in the absence of tumour size reduction[25-28].

Radiomics and machine learning models add another dimension to response evaluation by analysing imaging biomarkers that reflect tumour heterogeneity. Studies have shown that textural features extracted from CT and MRI images can identify treatment response patterns beyond mere size reduction, offering a more comprehensive understanding of therapy effects[29,30]. Radiologists can now evaluate subtle changes in tumour biology that precede macroscopic alterations, which is particularly useful in early response assessment and in cases where conventional metrics fail to capture complex therapeutic effects[31-34].

Moreover, minimal residual disease (MRD) detection via circulating tumour DNA (ctDNA) analysis has demonstrated the capacity to identify signs of recurrence 3.8 to 10 months earlier than conventional imaging modalities (e.g., CT and MRI), thereby facilitating timely clinical intervention[35,36]. MRD status can be dynamically monitored postoperatively on a regular basis (e.g., every 3 months); if ctDNA remains undetectable, the risk of recurrence within 2 years is minimal, whereas a positive ctDNA result necessitates an intensified imaging surveillance protocol (e.g., every 2 months) to enhance the probability of successful secondary surgical resection. Notably, early intervention based on ctDNA monitoring has been associated with a 72% rate of patients with recurrent LM undergoing a second radical treatment, which significantly prolongs survival[37].

Radiomics and machine learning in CRLM management

Radiomics—the extraction of quantitative data from medical images—has gained momentum as a tool to capture and analyse complex tumour characteristics[30,38-43]. For CRLM, radiomic features derived from texture, intensity, and shape provide insights into tumour heterogeneity and microenvironment, enabling the identification of aggressive lesions that may require intensified therapy[44]. Radiomics applied to MRI, particularly using hepatocyte-specific contrast and DWI, has shown promising results in distinguishing between mucinous and non-mucinous metastatic lesions[45]. This is clinically significant, as mucinous metastases tend to respond differently to treatments and carry distinct prognostic implications.

When combined with machine learning, radiomics enables predictive modelling that surpasses traditional visual assessment. For instance, algorithms trained on radiomic features from baseline CT or MRI scans have demonstrated high accuracy in forecasting treatment outcomes and recurrence risk[46-48]. These models are invaluable for early identification of high-risk patients, informing treatment intensification or alternative approaches before conventional signs of progression emerge. For example, a radiomics model based on texture analysis of CT images was applied in a cohort of CRLM patients to predict response to neoadjuvant chemotherapy, thereby directly informing clinical decision-making[49]. This shift toward predictive radiology marks an important step in precision oncology, enabling personalized management plans that align closely with each patient's disease profile.

AI for prognostication and treatment personalization

AI has revolutionized radiology, especially in the prognostication of CRLM[44]. AI algorithms process vast datasets, including imaging and clinical information, to identify patterns associated with disease progression, recurrence risk, and patient survival[50]. In CRLM, AI-driven models have shown high accuracy in predicting post-treatment recurrence, a critical factor in determining follow-up frequency and adjuvant therapy plans[51-53]. Moreover, AI has proven effective in assessing genetic and molecular characteristics of tumours, such as KRAS mutation status, which influences prognosis and treatment response[54,55].

By integrating imaging biomarkers with genetic and clinical data, AI enhances the ability to create highly personalized treatment plans. In addition, a multicentre study in 2024 showed a significant decline in the predictive performance of radiomics models in independent cohorts (c statistic from 0.78 to 0.50), suggesting the need for more standardized datasets[56]. Moreover, large-scale prospective clinical trials should be promoted to verify the clinical utility of these models and gradually incorporate them into clinical guidelines.

Taking a higher-dimensional perspective, combining genomics (such as TMB and PD-L1 expression), metabolomics, and radiomics can build a more comprehensive prediction system, representing a promising avenue for future research.

Radiogenomics in CRLM management

Radiogenomics, which integrates quantitative imaging features with genomic data, represents a cutting-edge approach in CRLM management. For example, specific radiomic features extracted from hepatocyte-specific contrast-enhanced MRI, when combined with molecular markers such as KRAS mutation status, have been shown to improve the stratification of patients in terms of recurrence risk and survival outcomes[5,51,53]. This integrated approach not only enhances prognostication but also informs personalized treatment strategies by predicting response to targeted therapies. Further studies focusing on radiogenomics in CRLM are warranted to validate these promising findings and to standardize data acquisition and analysis protocols[57,58].

Imaging in pre-surgical planning and resectability assessment

Determining resectability is paramount for patients with CRLM, as surgical resection offers the best chance of long-term survival in cases where the disease is confined to the liver. MRI, especially when enhanced with hepatocyte-specific contrast, is highly effective in mapping lesion boundaries, assessing vascular involvement, and optimizing resection margins. Advanced imaging techniques, such as three-dimensional reconstructions and virtual surgical planning, have further improved the precision of hepatic surgeries, ensuring that enough healthy liver tissue is preserved, thus reducing the risk of postoperative complications and recurrence. For instance, in a recent cohort of CRLM patients, 3D imaging techniques enabled precise mapping of tumour boundaries and vascular structures, leading to improved resection margins and a reduction in postoperative complications.

For patients with borderline resectable lesions, radiological imaging plays a key role in determining the need for neoadjuvant therapies to downstage tumours, increasing the likelihood of complete resection. Radiomics and texture analysis offer an additional layer of prognostic information by identifying imaging biomarkers associated with higher recurrence risk or poor therapeutic response, allowing clinicians to tailor pre-surgical interventions more effectively[59].

Abbreviated imaging protocols and surveillance

The demand for imaging in monitoring CRLM has led to the development of abbreviated MRI protocols, which provide sufficient sensitivity for detecting recurrence with shorter scan times and reduced healthcare costs. These protocols, which often include only the most critical sequences, such as T2-weighted and DWI, maintain diagnostic accuracy while reducing patient discomfort and resource usage. However, it is crucial to note that these abbreviated protocols may be best suited for follow-up in stable cases, while full MRI protocols remain preferable for initial staging and complex cases that require detailed lesion characterization[60].

Conclusion

The landscape of radiological assessment for hepatic metastases from CRC is advancing rapidly, with innovative imaging techniques, radiomics, and AI transforming the detection, characterization, and management of CRLM. These advancements hold great potential for improving outcomes by enabling more accurate staging, precise treatment planning, and personalized follow-up strategies. The integration of predictive models, AI-based radiomics, and an expanded focus on radiogenomics—coupled with the critical need for large-scale prospective trials and the promise of combining genomics, metabolomics, and radiomics to form comprehensive prediction systems—represents a new frontier in precision oncology. Moving forward, it is essential to continue investing in multi-center trials and developing standardized protocols to ensure these tools can be seamlessly integrated into routine clinical practice. As these innovations mature, they promise to make radiology an even more powerful contributor to personalized care in CRLM.