"MMFBC-DBN"

"This repository provides the official PyTorch implementation of MMFBC-DBN (Multi-Modal Fusion and Boundary-Constrained Dual-Branch Network), a lightweight yet high-performance framework for 3D brain tumor segmentation. The model follows a U-shaped architecture and introduces two synergistic pathways: a progressive multimodal fusion pathway that hierarchically aggregates four MRI modalities (FLAIR, T1, T1ce, T2) using HFM with DAFM and CSAM modules, and a boundary-aware pathway that explicitly extracts and reconstructs tumor boundaries through MABE with a learnable 3D Sobel operator and a dedicated Boundary-Decoder. Both branches share a common encoder, are optimized independently via separate decoders, and are reinforced with dual deep supervision. This release includes the complete inference code, pre-trained weights for BraTS 2018, 2019, 2020, 2021, and BraTS-Africa, evaluation scripts, and configuration files. With only 3.21 million parameters, the model achieves state-of-the-art segmentation accuracy and boundary consistency across all five datasets, while maintaining a compact architecture with low GPU memory footprint and fast inference time. The TorchScript-exported models bundle the network architecture and weights, enabling out-of-the-box inference without requiring any source code. This repository enables full reproducibility of our experiments and facilitates further research in multimodal fusion and boundary-constrained medical image segmentation. Please cite our paper if you find this code useful in your work."