AI tool predicts bowel cancer relapse

Researchers at La Trobe University have developed an AI tool called SEMIL (Semantically-Enhanced Multiple Instance Learning) that analyzes routine pathology slides to predict relapse risk in stage-two bowel cancer patients. The study, published in Gastroenterology, analyzed over 1,600 slides and validated findings across 1,220 patients from multiple Australian institutions. The tool assesses tumor growth patterns at the invasive front, a feature difficult for pathologists to classify consistently. When AI assessments aligned with pathologist evaluations, the most accurate risk ratings were produced. The tool is designed to support clinical decision-making, potentially enabling earlier life-saving treatment for high-risk patients while sparing low-risk patients from unnecessary chemotherapy. Bowel cancer is Australia's fourth most common cancer and the second most common cause of cancer death worldwide.