AI-DRIVEN CARDIAC MRI SEGMENTATION USING A HYBRID TRANSU-NET MODEL

The "AI-Driven Cardiac MRI Segmentation Using a Hybrid TransU-Net Model" project introduces a robust and scalable framework for automated segmentation of cardiac structures in MRI scans, leveraging state-of-the-art deep learning models: U-Net, U-Net++, and TransU-Net. Implemented within a Flask-based web application, the system integrates secure user authentication via MySQL and supports the processing of NIfTI-format MRI volumes to segment critical cardiac regions are background, right ventricle, myocardium, and left ventricle with average Dice scores. The pipeline encompasses exploratory data analysis, preprocessing with resizing and normalization, data augmentation, and GPU-accelerated model training using a combined loss function (Weighted CrossEntropy and Dice). A dedicated model comparison dashboard visualizes performance metrics, including training/test accuracy and Dice coefficients, using interactive charts and tables. Accurateprediction capabilities allow users to upload MRI scans for segmentation, with results visualized alongside quantitative metrics. Deployed on a flask framework, the framework ensures efficient computation and scalability. This system serves as a powerful tool for researchers and clinicians, enabling precise cardiac segmentation and comparative model evaluation to advance medical imaging diagnostics and research.

Keywords:

 Cardiac MRI Segmentation, Deep Learning, U-Net, U-Net++, TransU-Net, Medical Imaging, Flask Web Application, MySQL Authentication, Dice Coefficient, Model Evaluation, GPU Computing.