ESMO GI 2026: AI Biomarkers and Clinical Value in Cancer Care

Experts at the ESMO Gastrointestinal Cancers Congress 2026 highlighted how artificial intelligence is moving beyond image recognition to deliver clinical value across surgery, pathology, and image-guided intervention. They also presented the ESMO basic requirements for AI-based biomarkers in oncology (EBAI), a framework that classifies AI-based biomarkers into three tiers and defines minimum validation requirements for each.

In colorectal cancer, the MSIntuit algorithm detected microsatellite instability directly from routine haematoxylin and eosin (H&E) slides with 96–98% sensitivity (Nat Commun. 2023;14:6695), while deep learning models have been developed to predict lymph node metastasis from pre-treatment CT imaging in gastric cancer (eClinicalMedicine. 2024;75:102805). According to experts, the field is entering a new phase where clinical value is proving as important as technical performance.

In surgical AI, proof-of-concept data using the AI platform EUREKA X quantified preservation of loose connective tissue during robotic rectal cancer surgery. In a matched case-control study of 44 patients, AI-derived measures of tissue preservation were associated with postoperative urinary dysfunction; when combined with patient age, the model achieved an area under the curve of approximately 0.81. Researchers suggested that future AI systems could provide real-time feedback during surgery, noting that 'the future is not AI versus surgeons. It is AI-enhanced surgeons.'

In image-guided intervention, MRI-guided interstitial brachytherapy is expanding treatment options for liver tumours unsuitable for thermal ablation, enabling visualisation of lesions often invisible on CT. In the prospective MR BRIGHT study (Z Gastroenterol 2025; 63(01): e45), MRI-guided brachytherapy achieved significantly higher local recurrence-free survival than CT-guided treatment (95.1% versus 79.9%; p=0.027), while substantially reducing low-dose radiation exposure to healthy liver tissue (15% versus 43%). Machine learning algorithms could accelerate MRI image reconstruction and reduce artefacts, making real-time MRI guidance increasingly practical.

Pathology may be where AI's transition from research to routine practice is most evident. Experts described AI as the product of four converging developments: an expanding biomarker landscape, increasing pathology workloads, widespread adoption of digital pathology, and persistent diagnostic variability. The EBAI framework outlines how algorithms are progressing from standardising existing biomarkers to predicting molecular alterations directly from H&E slides and ultimately discovering entirely new biomarkers, including emerging models capable of recognising when they should not make a prediction (Ann Oncol. 2026;37(3):414–430).