Automated Detection of Ischemic Stroke-Related Brain Atrophy Using Deep Learning-Based Tissue Segmentation

Ischemic stroke is a major contributor to global mortality and prolonged disability, frequently leading to stroke-induced brain atrophy, which involves the deterioration of white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). Precise segmentation of brain tissues is essential for diagnosing and tracking atrophy; however, manually segmenting magnetic resonance imaging (MRI) images is labor-intensive, subjective, and prone to inaccuracies. To overcome these limitations, this study introduces a comprehensive deep learning-based pipeline for automated brain tissue segmentation in ischemic stroke patients. The pipeline incorporates preprocessing steps such as skull stripping and bias field correction, followed by data augmentation to enhance dataset diversity. This study utilizes two deep learning architectures, U-Net and an enhanced U-Net integrated with ResNet (U-Net + ResNet), are employed for segmentation, with ResNet integration improving feature extraction for complex stroke-affected tissues. The models undergo training and evaluation on a dataset of MRI scans collected from SRM Medical College Hospital, with ground truth labels annotated by a radiologist. The performance of the model is assessed using evaluation metrics such as the Dice Similarity Coefficient (DSC) and Intersection over Union (IoU). Results demonstrate that U-Net+ResNet outperforms U-Net, achieving DSC scores of 0.9482, 0.9723, and 0.9267 for WM, GM, and CSF. The proposed pipeline provides a robust and efficient solution for brain tissue segmentation, enabling improved diagnosis and analysis of stroke-related atrophy. This study advances medical imaging by providing a scalable and reproducible tool for clinical use, ultimately enhancing patient outcomes.