Added prognostic value of incorporating MRI into WSI+RNA-seq glioma survival models

Determine whether adding FLAIR MRI-derived volumetric features as a third modality to a multimodal deep learning survival prediction framework that already integrates haematoxylin and eosin-stained whole-slide histopathology and RNA-seq gene expression for glioma patients provides additional prognostic value beyond the corresponding bimodal histopathology+RNA-seq model.

Background

Multimodal deep learning for glioma survival prediction has largely focused on combining histopathology whole-slide images (WSI) with RNA-seq gene expression, where early fusion has shown strong performance. Although MRI is routinely acquired and captures complementary volumetric and structural information, its incremental contribution when added to a WSI+RNA-seq framework had not been established prior to this work.

The paper frames this as a central unresolved issue in extending bimodal approaches to a trimodal setting that includes FLAIR MRI, motivating an empirical evaluation to assess whether MRI confers measurable prognostic gains over a strong WSI+RNA-seq baseline.