AI Model Detects Multiple Genetic Colorectal Cancer Markers from Routine Tissue Slides

A new AI model has successfully detected multiple genetic markers associated with colorectal cancer simultaneously from standard histological tissue slides, marking a significant advancement in cancer diagnostics. The study, published in The Lancet Digital Health, involved nearly 2,000 digitized tissue samples from colon cancer patients across seven independent cohorts in Europe and the United States. These samples included whole-slide images of tissue sections along with clinical, demographic, and lifestyle data. Researchers developed a novel multi-target transformer model capable of identifying a wide range of genetic alterations directly from routine stained tissue slides. Unlike previous deep learning models that focused on predicting one mutation at a time, this new approach analyzes multiple genetic changes together, capturing complex interactions and shared morphological patterns in tumor tissues. The model excels at detecting key biomarkers such as BRAF and RNF43 mutations, as well as microsatellite instability (MSI), a critical indicator for identifying patients who may respond well to immunotherapy. The study found that certain mutations are more prevalent in MSI-high tumors, suggesting that multiple genetic changes collectively influence tissue appearance. The AI model detects these patterns by recognizing visual features linked to underlying molecular changes, rather than analyzing each mutation in isolation. The development and validation of the model brought together experts in data science, computer science, epidemiology, pathology, and oncology. Pathological interpretation was led by Dr. Nic Reitsam from the University Hospital Augsburg, who contributed essential clinical insights. The results showed that the model’s performance matched or even surpassed existing single-target models in predicting various biomarkers across diverse patient groups. Marco Gustav, M.Sc., the study’s first author and researcher at the EKFZ for Digital Health at TU Dresden, emphasized the model’s ability to uncover previously underappreciated genetic associations. “Our findings suggest that the interplay of multiple mutations shapes tissue structure in ways that can be detected by AI,” he said. Jakob N. Kather, Professor of Clinical Artificial Intelligence at EKFZ and senior oncologist at the NCT/UCC of the University Hospital Carl Gustav Carus Dresden, highlighted the clinical potential. “This technology could streamline cancer diagnostics by enabling rapid, cost-effective pre-screening. It may help clinicians prioritize patients for more detailed molecular testing and support personalized treatment decisions.” The research team plans to expand the approach to other cancer types, aiming to transform how molecular information is extracted from routine pathology workflows. The study demonstrates that AI can bridge the gap between molecular genetics and tissue morphology, opening new pathways for precision oncology.