Fine-Grained Hierarchical Progressive Modal-Aware Network for Brain Tumor Segmentation

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Brain tumors are critical health concerns requiring precise diagnosis and treatment.
• Magnetic resonance imaging (MRI) offers vital multi-modal data for tumor detection but faces fusion and feature extraction challenges.
• Accurate brain tumor segmentation is hindered by blurred boundaries and complex morphology.

Purpose of the Study:

• To introduce FiHam, a fine-grained hierarchical progressive modal-aware network for enhanced brain tumor segmentation.
• To develop novel multi-modal fusion and feature extraction strategies for improved MRI analysis.
• To overcome limitations in current methods for segmenting complex brain tumor structures.

Main Methods:

• Proposed FiHam network with a progressive fusion strategy for modality-specific and integrated feature extraction.
• Implemented a gated cross-attention modal-fusion module for adaptive dual-modal feature integration.
• Incorporated a tiny U-Net in the encoder to capture fine-grained boundary and morphological details.

Main Results:

• FiHam achieved state-of-the-art performance on three large-scale, multi-modal brain tumor datasets.
• Demonstrated significant improvements in segmentation accuracy compared to existing methods.
• Showcased enhanced generalizability across diverse MRI modalities.

Conclusions:

• FiHam effectively addresses challenges in multi-modal MRI fusion and feature extraction for brain tumor segmentation.
• The proposed network architecture significantly enhances segmentation precision and reliability.
• FiHam represents a substantial advancement in automated brain tumor analysis using medical imaging.