Hybrid Wavelet-Mamba diffusion network (HyWaM-Diff) for joint brain tumor segmentation and glioma grading using multi-modal MRI

Abstract Accurate brain tumor segmentation and glioma grading from multi-modal MRI are critical for clinical diagnosis and treatment planning. However, these tasks are frequently complicated by tumor heterogeneity and indistinct boundaries. To address these limitations, we present HyWaM-Diff, a lightweight, multi-task framework designed for simultaneous segmentation and grading. The unified architecture integrates a wavelet-enhanced encoder to preserve structural details, a Mamba module to efficiently capture long-range contextual dependencies, and an uncertainty-aware diffusion decoder to refine boundaries. Evaluated on the BraTS 2020 dataset, HyWaM-Diff achieved strong segmentation performance with Dice scores of 0.90, 0.87, and 0.84 for whole tumor, tumor core, and enhancing tumor, respectively. For glioma grading, the framework attained an accuracy of 0.91, an F 1-score of 0.90, and an AUC of 0.93. Extensive ablation studies and statistical validation confirm the stability and contribution of the proposed framework.