LCMF-Net: A lightweight collaborative multimodal fusion network for brain tumor segmentation.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Accurate brain tumor segmentation is crucial for clinical decisions, but manual methods are inefficient.
• Deep learning methods face challenges in multimodal feature integration and computational cost.
• Existing approaches struggle to effectively combine information from different MRI sequences and adjacent slices.

Purpose of the Study:

• To introduce a novel Lightweight Collaborative Multimodal Fusion Network (LCMF-Net) for high-precision and high-efficiency brain tumor segmentation.
• To address limitations in multimodal feature integration and computational complexity in current deep learning models.
• To develop a practical and reliable tool for clinical brain tumor segmentation.

Main Methods:

• Proposed LCMF-Net utilizes a multi-branch encoder with Cross-Modality and Cross-Slice Attention (CMCSA) for feature enhancement.
• Incorporated a State Space Model-based Fusion (SSM-Fusion) module for adaptive, spatially continuous feature integration.
• Employed a spatial dimensionality reduction strategy and improved Residual Inception Blocks (RIB) for efficient 3D contextual modeling within 2D constraints.

Main Results:

• LCMF-Net achieved a 1.6% improvement in segmentation accuracy compared to state-of-the-art methods.
• Demonstrated a reduction in computational cost by over 50%.
• Exhibited strong segmentation performance with low computational complexity.

Conclusions:

• LCMF-Net offers a practical and reliable solution for clinical brain tumor segmentation.
• The network effectively integrates multimodal MRI features and maintains spatial continuity.
• Achieves a balance between high accuracy and computational efficiency for real-world applications.