Integrated Multi-Omics and Whole Slide Images for Survival Prediction in Glioblastoma Using Multiple Instance Learning and Co-Attention

Glioblastoma multiforme (GBM) is the most aggressive adult brain tumor and presents significant treatment challenges due to its poor prognosis and heterogeneity. Despite the rapid development of deep learning, integrating multi-omics with whole slide images (WSIs) for survival prediction remains difficult. This study aims to improve GBM prognosis by integrating WSIs with multi-omic data through the incorporation of biological pathway knowledge. Utilizing multiple instance learning and co-attention mechanisms, we initiated the integration of multi-omic data informed by biological pathways, leveraging existing knowledge of molecular interactions. The proposed model was evaluated using data from 214 GBM patients from The Cancer Genome Atlas. This dataset included 447 WSIs and multi-omic features such as 927 RNA sequencing gene expressions, 1,168 copy number alterations, and 1,489 DNA methylation patterns. These multi-omic features were organized into nine biological pathways, each selected based on their relevance to GBM, ensuring a targeted and biologically informed strategy for survival prediction. Our results show that the proposed model outperforms existing benchmarks by at least 4.5%, highlighting the potential of incorporating additional biological knowledge into the integration of multimodal data to improve GBM survival prediction.