Vector Institute Develops CRISPNAM-FG: An Interpretable AI Model for Diabetes Complications

Researchers at the Vector Institute for Artificial Intelligence have developed CRISPNAM-FG, a novel deep learning model designed to predict the risk of diabetes-related foot complications in discharged patients. This innovative approach addresses the critical challenge of interpretability in healthcare AI by combining the Fine-Gray competing risks framework with Neural Additive Models. Unlike traditional black-box deep learning systems such as DeepHit, CRISPNAM-FG offers intrinsic transparency, allowing clinicians to understand how individual features influence risk predictions for competing events, such as foot complications versus mortality. The model achieves high predictive accuracy while maintaining feature-level clarity, thereby fostering trust and facilitating clinical adoption. This breakthrough represents a collaborative effort involving the Vector Institute, GEMINI, Unity Health, and Diabetes Action Canada. By solving the opacity issue inherent in many advanced AI systems, this research demonstrates that accurate prediction and interpretability can coexist, potentially transforming decision-making processes in high-stakes medical environments and improving patient outcomes through trustworthy artificial intelligence applications.