Artificial intelligence (AI) has become an integral force in the clinical landscape of gastrointestinal (GI) oncology. Recent advances in model architectures ranging from traditional machine learning and convolutional neural networks (CNNs) to transformer-based foundational models and graph neural networks (GNNs) have enabled the extraction of complex features from diverse data modalities, including endoscopic images, radiology, pathology whole-slide images, and multi-omics profiles. In this review, AI models are systematically classified into supervised learning, unsupervised clustering, multimodal fusion, and interpretable modeling. The advantages of each model are delineated in unravelling tumor heterogeneity, anatomical characteristics, and treatment-relevant biomarkers. Furthermore, three types of clinical application are emphasized: (1) early screening and lesion localization via segmentation or anomaly detection; (2) molecular subtyping and patient stratification for diagnosis with risk assessment; (3) therapy guidance through response prediction and personalized treatment planning. We also discuss major challenges on the application of AI in integration of heterogeneous clinical data, model generalizability across centers, and the interpretability of predictions. Collectively, this review highlights the transformative potential of AI in better understanding tumor biology and its clinical value in advancing personalized medicine for GI cancer patients.
Liu et al. (Mon,) studied this question.