Graph Neural Network for Brain Tumor Segmentation with Missing MRI Modalities

A new research paper on arXiv proposes a graph-based one-stage framework for brain tumor segmentation that handles missing MRI modalities. Multimodal MRI typically uses four key modalities for effective sub-region analysis, but missing modalities are common in clinical practice and degrade existing full-modality methods. Previous approaches often require multi-stage training for different modality scenarios, increasing costs and inadequately addressing interference from missing data. The proposed method introduces modality-specific virtual nodes as supplementary information sources to compensate for missing modalities and enhances robustness against arbitrary modality combinations. The framework is designed as a single-stage training process, reducing computational overhead. The paper is available at arXiv:2605.16880.