Deep Learning-Based Automated Segmentation of the Parcellated Corpus Callosum in Brain MRI

This study proposed PCcS-RAU-Net, an enhanced deep learning model that integrates Residual Blocks and Attention Gates (AGs) into a U-Net architecture for segmenting the parcellated Corpus Callosum (Cc) into five anatomical regions: rostrum, genu, mid-body, isthmus, and splenium. The model was evaluated across three prominent datasets: ABIDE, OASIS, and Real Clinical Images (RCI), achieving 97.10%, 96.88%, and 97.11%, in terms of Dice Similarity Coefficients (DSCs), and corresponding Mean Intersection over Union (MIoU) scores of 94.43%, 93.89%, and 94.19% respectively. Additionally, this approach outperformed the classical U-Net, Residual U-Net, and Attention U-Net models, providing robustness and an effective general applicability concerning the different imaging protocols.