AMF-U-Net enhances 3D brain tumor segmentation using adaptive multimodal fusion

This study presents AMF-U-Net, a multi-stream residual 3D U-Net designed for precise brain tumor segmentation from multi-modal MRI scans. The model processes T1, T1ce, T2, and FLAIR images through separate encoders and uses a Modality Fusion Module to weight and combine modality-specific features, improving boundary awareness and segmentation accuracy. Residual connections contribute to more stable training, addressing challenges like intensity heterogeneity and unbalanced voxel class distribution.

Originally published by gnews