ABI-Net: Attention-Based Inception U-Net for Brain Tumor Segmentation From Multimodal MRI Images

Abstract

Magnetic Resonance Imaging (MRI) is widely used for glioma evaluation, but manual segmentation is impractical due to the large data volume. Automated techniques are necessary for precise clinical measurements. U-Net has shown promise in volumetric segmentation, but brain tumor segmentation remains challenging due to tumor diversity in type, location, and structure. This study introduces ABI-Net, an advanced U-Net variant integrating Attention-based Inception blocks for improved segmentation of brain tumor sub-regions in 3D multimodal MRI images. ABI-Net leverages the Inception module for spatial feature extraction and an attention mechanism to enhance cancerous region detection. Trained on the BraTS 2020 dataset, ABI-Net achieved dice scores of 0.8354, 0.8505, and 0.8782 for enhancing tumor (ET), tumor core (TC), and whole tumor (WT), respectively, outperforming state-of-the-art models. On the validation dataset (125 patients without segmentation masks), ABI-Net obtained average dice scores of 0.8189, 0.8401, and 0.8673 for ET, TC, and WT. ABI-Net provides an accurate, efficient solution for automated brain tumor segmentation, with significant potential for clinical applications, including diagnosis, treatment planning, and patient monitoring.