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

Researchers have developed a method to quantify cross-modal interactions in multimodal glioma survival prediction using InterSHAP, a Shapley interaction index-based metric¹. This approach allows for the analysis of how different data modalities, such as imaging and genomic data, interact to improve survival prediction models. By adapting InterSHAP to Cox proportional hazards models, the study provides evidence for additive signal integration, where the combination of multiple data modalities leads to better predictions than any single modality alone. The findings suggest that the interactions between different data modalities are not necessarily synergistic, but rather additive, providing a more nuanced understanding of how multimodal deep learning models work. This matters to practitioners because it highlights the importance of carefully evaluating the interactions between different data modalities in multimodal machine learning models, particularly in high-stakes applications such as cancer prognosis.