AGHA-Net: Anatomical Gating and Hybrid Attention 3D Network for Efficient Brain Tumor Segmentation

This repository contains the official codebase, pre-trained model weights, and comprehensive evaluation metrics for the study: "AGHA-Net: Anatomical Gating and Hybrid Attention Network for 3D Brain Tumor Segmentation".

AGHA-Net is a novel, lightweight 3D Convolutional Neural Network (CNN) designed for the automated volumetric segmentation of brain gliomas using multimodal Magnetic Resonance Imaging (MRI) scans (T1n, T1c, T2w, and FLAIR). The architecture introduces Anatomical Gating (AG) modules to filter out background parenchyma noise in skip connections, alongside a Hybrid Attention (HA) bottleneck to dynamically recalibrate spatial and channel-wise features, prioritizing regions of active angiogenesis.