La Trobe University researchers developed SÉMIL, an AI model that analyzes digital pathology slides to identify stage 2 bowel cancer patients at greater risk of relapse. The system evaluates tumor growth patterns at the invasive front, a prognostic feature that can be difficult for pathologists to classify consistently.
SÉMIL was trained using 388 stage 3 bowel cancer cases and validated across more than 1,220 stage 2 patients. The model identified a higher-risk group with roughly twice the risk of relapse within five years, independent of established clinical risk factors. Researchers also found that combining SÉMIL’s assessment with a pathologist’s evaluation improved risk classification.
For clinicians, the model could eventually support more individualized follow-up and treatment decisions without requiring additional tissue samples or costly tests. The research reflects growing interest in AI-assisted pathology for cancer risk stratification.
AI Risk Models
La Trobe University Launches Its BowelRec AI Tool
Trend Themes
-
AI-assisted Pathology — Digital slide analysis is creating new avenues for more consistent cancer risk stratification by augmenting pathologist expertise with scalable machine learning insights.
-
Predictive Cancer Modeling — Risk models that identify relapse probability from existing clinical images are reshaping personalized oncology by reducing dependence on additional invasive tests.
-
Hybrid Clinical Intelligence — Combining clinician assessment with AI-generated classifications reveals opportunities for more precise decision support in complex diagnostic workflows.
Industry Implications
-
Oncology — Cancer care is being transformed by tools that support individualized monitoring and treatment planning based on relapse risk prediction.
-
Digital Pathology — Pathology platforms are expanding beyond digitization toward prognostic analytics that extract clinically meaningful patterns from routine tissue slides.
-
Medical AI — Healthcare AI developers are finding growth potential in specialized models that improve decision confidence within high-stakes diagnostic environments.