Quantifying Cross-Modal Interactions in Multimodal Glioma Survival Prediction via InterSHAP: Evidence for Additive Signal Integration

This project is a reference implementation for an academic paper (arXiv:2603.29977). Its primary value lies in the mathematical adaptation of InterSHAP—originally designed for classification—to the Cox Proportional Hazards (CPH) framework used in survival analysis. From a competitive standpoint, the project has a low defensibility score (2) because it currently lacks a community (0 stars) and is a specialized research tool rather than a product. However, its findings are highly relevant to the medical AI field: it challenges the common assumption that multimodal deep learning models benefit from complex 'synergistic' interactions, suggesting instead that the benefits are largely additive. This insight could influence how future multimodal architectures are designed in oncology. Frontier labs like OpenAI or Google are unlikely to target this specific niche (glioma-specific InterSHAP), though they might eventually include survival-specific interpretability tools in their broader healthcare offerings (e.g., Med-Gemini). The displacement horizon is set at 1-2 years, as this is a specific technique that could be superseded by more advanced interpretability methods (like higher-order interaction metrics or more efficient Shapley approximations) or integrated into larger frameworks like MONAI or H2O.ai.