Generalized Knowledge Distillation for Unimodal Glioma Segmentation from Multimodal Models

Gliomas, primary brain tumors arising from glial cells, can be effectively identified using Magnetic Resonance Imaging (MRI), a widely employed diagnostic tool in clinical settings. Accurate glioma segmentation, which is crucial for diagnosis and surgical intervention, achieved by integrating multiple MRI modalities that offer complementary information. However, limited access to certain contexts often results suboptimal performance of segmentation methods. This study introduces novel generalized knowledge distillation framework designed transfer multimodal teacher model unimodal student via two distinct strategies: graph cascade region attention distillation. The former enables the replicate teacher’s softened output, whereas latter facilitates extraction learning feature information at various levels within model. Our evaluation proposed strategies BraTS 2018 dataset confirms their superior compared with existing