Enhancing rectal cancer liver metastasis prediction: Magnetic resonance imaging-based radiomics, bias mitigation, and regulatory considerations

Abstract

In this article, we comment on the article by Long et al published in the recent issue of the World Journal of Gastrointestinal Oncology. Rectal cancer patients are at risk for developing metachronous liver metastasis (MLM), yet early prediction remains challenging due to variations in tumor heterogeneity and the limitations of traditional diagnostic methods. Therefore, there is an urgent need for non-invasive techniques to improve patient outcomes. Long et al’s study introduces an innovative magnetic resonance imaging (MRI)-based radiomics model that integrates high-throughput imaging data with clinical variables to predict MLM. The study employed a 7:3 split to generate training and validation datasets. The MLM prediction model was constructed using the training set and subsequently validated on the validation set using area under the curve (AUC) and dollar-cost averaging metrics to assess performance, robustness, and generalizability. By employing advanced algorithms, the model provides a non-invasive solution to assess tumor heterogeneity for better metastasis prediction, enabling early intervention and personalized treatment planning. However, variations in MRI parameters, such as differences in scanning resolutions and protocols across facilities, patient heterogeneity (e.g., age, comorbidities), and external factors like carcinoembryonic antigen levels introduce biases. Additionally, confounding factors such as diagnostic staging methods and patient comorbidities require further validation and adjustment to ensure accuracy and generalizability. With evolving Food and Drug Administration regulations on machine learning models in healthcare, compliance and careful consideration of these regulatory requirements are essential to ensuring safe and effective implementation of this approach in clinical practice. In the future, clinicians may be able to utilize data-driven, patient-centric artificial intelligence (AI)-enhanced imaging tools integrated with clinical data, which would help improve early detection of MLM and optimize personalized treatment strategies. Combining radiomics, genomics, histological data, and demographic information can significantly enhance the accuracy and precision of predictive models.

Key Words: Metachronous liver metastasis; Radiomics; Machine learning; Rectal cancer; Magnetic resonance imaging variability; Bias mitigation; Food and Drug Administration regulations; Predictive modeling

Core Tip: The early detection of metachronous liver metastasis (MLM) in rectal cancer patients remains a challenge due to tumor heterogeneity and limitations in imaging methods. Long et al’s magnetic resonance imaging (MRI)-based radiomics model, integrated with machine learning algorithms, offers a novel non-invasive solution for improving MLM prediction. This approach has the potential to enhance personalized treatment planning and patient outcomes. However, addressing sources of bias such as MRI variability and patient heterogeneity, along with aligning the model with Food and Drug Administration regulations on artificial intelligence-based medical technologies, will be critical to successful clinical integration.

TO THE EDITOR

Colorectal cancer (CRC) is projected to continue rising in incidence and mortality, particularly in transitioning nations and countries with low-to-medium Human Development Index (HDI) levels, as well as in developed nations experiencing generational shifts. In addition, the impact of coronavirus disease 2019 has led to a notable surge in advanced-stage colorectal cancer diagnoses, with ongoing repercussions anticipated. Among patients with metastatic cancer, the liver is the most common site of metastasis. Rectal cancer, typically associated with worse survival outcomes due to its anatomical location, distinct histological features, and increased risk of local recurrence, underscores a strong need for innovative non-invasive diagnostic tools[1-6].

RADIOMICS AND MACHINE LEARNING IN ONCOLOGY

Long et al[7] developed an magnetic resonance imaging (MRI)-based radiomics model that integrates multiparametric MRI data with machine learning algorithms to predict liver metastasis in rectal cancer patients. The model combines imaging data with clinical variables to improve prediction accuracy. The study focuses on predicting metachronous liver metastasis (MLM), showcasing the role of machine learning in diagnostics by enhancing the precision of metastasis predictions. One example is how the model leverages high-resolution imaging data, which may detect subtle textural differences that traditional methods miss, underscoring the advantage of non-invasive prediction techniques[7]. However, this reliance on high-resolution imaging introduces challenges, such as the need for standardized imaging protocols to minimize variability across scanners and institutions. Differences in image acquisition settings, such as contrast agents, resolution, and scanning parameters, can affect the reproducibility of results. Additionally, the processing and analysis of such detailed imaging data often require advanced computational resources and expertise, potentially limiting accessibility in low-resource settings.

BIAS MITIGATION IN RADIOMICS MODELS

MRI parameters and patient heterogeneity can introduce biases into predictive models. For example, MRI scans performed at different institutions may use varied imaging protocols, leading to inconsistent image quality, which affects the model’s accuracy and reproducibility. Such challenges are common in artificial intelligence (AI) and machine learning-enabled devices, where variations can influence results. Standardizing data and validating models across multiple centers is essential to ensuring clinical applicability. Bias reduction is foundational to model accuracy, which is vital for improving patient outcomes and reducing healthcare costs[8].

EXPLAINABLE AI IN RADIOMICS

Explainable AI (XAI) plays a pivotal role in making complex AI decision-making processes more understandable for clinicians. By promoting trust, transparency, and interpretability, XAI ensures clinicians can verify the reliability of AI systems[9]. For instance, tools like Shapley additive explanations allow clinicians to identify the most influential features in prediction outcomes, offering insights into how specific MRI features correlate with MLM risk, which promotes clinician confidence in the tool’s predictions. Shapley is particularly well-suited for radiomics applications due to its ability to manage non-linear interactions in high-dimensional data, which is common in MRI-based analyses. Additionally, complementary methods for XAI, such as local interpretable model-agnostic explanations, also provide feature importance scores similar to Shapley, offering diverse approaches for explaining AI decisions. The 2016 initiative of the Defense Research Projects Agency aimed to develop AI models with clear and interpretable decision-making pathways[10], bridging the gap between AI models and clinical practice.

REGULATORY CONSIDERATIONS FOR AI IN HEALTHCARE

The Food and Drug Administration (FDA)’s guidelines on AI emphasize transparency, bias mitigation, and continuous monitoring to ensure safe clinical deployment. For instance, if a radiomics model predicts MLM in a diverse population, the FDA may require evidence that the model has been tested and validated across demographic groups to avoid performance disparities. The FDA emphasizes enrolling diverse populations and standardizing demographic data collection in clinical studies to ensure unbiased treatment effects and equitable performance of medical devices, including radiomics models. By adhering to these FDA guidelines, developers can ensure that their radiomics models are rigorously tested and validated across diverse demographic groups, thereby minimizing performance disparities and promoting equitable healthcare outcomes. Advancements in AI, especially with multimodal data integration, push diagnostic boundaries, necessitating adaptive regulatory practices to support these innovations[11]. Ethical and legal considerations associated with AI limit broad application and reproducibility, and regulatory compliance is crucial to safeguarding patient outcomes as AI becomes more prevalent in healthcare.

CONCLUSION

The integration of MRI-based radiomics with machine learning marks a milestone in predicting liver metastases in rectal cancer patients. This model addresses deficiencies in early detection and offers a non-invasive means of predicting metastasis. Ongoing research will likely focus on refining these AI-powered models by addressing biases and ensuring regulatory compliance[7,11]. Future model enhancements, such as integrating genomics or histology, could improve prediction accuracy and model robustness, aligning with multimodal AI advancements in oncology. As AI continues to evolve in oncology, it has the potential to significantly enhance personalized medicine.