M$^{2}$SegMamba: Mamba-Based Incomplete Multimodal Learning for Brain Tumor Segmentation With Few Samples.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate brain tumor segmentation is crucial for clinical diagnosis and treatment.
• Multimodal magnetic resonance imaging (MRI) offers rich, complementary data but faces challenges with incomplete modalities and small sample sizes.
• Existing multimodal segmentation methods struggle with data scarcity and missing imaging sequences.

Purpose of the Study:

• To develop a robust framework (M²SegMamba) for brain tumor segmentation using multimodal MRI, specifically addressing incomplete data and small sample challenges.
• To leverage Mamba and Masked Autoencoder networks for both supervised and self-supervised learning in brain tumor segmentation.
• To enhance the interaction and integration of inter-modal and cross-modal image features.

Main Methods:

• Designed M²SegMamba framework integrating Mamba and Masked Autoencoder networks.
• Developed a specialized masking strategy for multimodal brain tumors to optimize feature extraction.
• Implemented a multi-traversal method for enhanced inter-modal and cross-modal feature interaction using Mamba.
• Incorporated TSmamba in skipping connections for efficient multimodal feature integration.
• Utilized auxiliary regularizers in encoder and decoder to improve robustness against incomplete modalities.

Main Results:

• M²SegMamba demonstrated superior performance compared to state-of-the-art methods on brain tumor segmentation tasks.
• The framework achieved high accuracy even with missing modalities in the BraTS 2018 and BraTS 2020 datasets.
• Results indicate significant improvements in handling incomplete multimodal MRI data for segmentation.

Conclusions:

• M²SegMamba offers a robust and effective solution for brain tumor segmentation, particularly in challenging clinical scenarios with incomplete multimodal data and limited samples.
• The proposed framework advances the field of medical image analysis by improving segmentation accuracy and reliability.
• The study highlights the potential of Mamba-based architectures for complex medical imaging tasks.