Leveraging the Untapped Potentials of Incomplete MRI Sequences for Glioma Grading and IDH Mutation Status Prediction

Motivation: The absence of MRI sequences poses significant challenges to reliable predictive modeling in multiparametric magnetic resonance imaging (mp-MRI) radiomics for glioma grading and predicting isocitrate dehydrogenase (IDH) mutation status. Goal(s): This study developed a disentangled-learning-based incomplete sequence completion enhanced robust network (DISCERN) to impute missing features and learn latent fusion representations, which are used for accurate, noninvasive glioma grading and IDH status prediction. Approach: Validation was performed across multi-center datasets and simulations of various clinical scenarios with differing missing rates to assess DISCERN's resilience to incomplete sequences. Results: DISCERN achieved robust performance in both glioma grading and IDH prediction. Impact: The DISCERN model demonstrates significant potentials for real-world clinical applications in noninvasive glioma grading and IDH mutation status prediction with incomplete mp-MRI data, offering a robust tool for clinical decision-making and personalized treatment planning.