Utilizing the U-Net Network for Segmenting the Left Ventricle from Cine-MRI Scans

In this research, we studied the U-Net segmentation model for identifying and demonstrating addresses related to cardiac left ventricle contour in cine-MRI images. Efficacy of the model is evaluated across different stages such as preprocessing, training, validation and robustness testing along with comparisons to conventional methods and other deep learning architectures. But the preprocessing includes standardized pixel intensities, and by data augmentation using rotation, flip method or scaling techniques. The training and validation stages clearly show the robust consolidation of our model with little overfitting, guaranteeing decent performance on different datasets. The developed model is robust by passing the test against noise/ artifacts, and providing a very good segmentation in extreme situations so that there are no accuracy outliers represented. Comparative analyses show better sensitivity-specificity trade-offs and precision-recall characteristics versus other methods. In general, a U-Net model developed here works well for cardiac MRI analysis and can be used as an effective tool to help automatic LV segmentation accurate and efficient in clinical applications which will improve diagnosis of heart diseases from imaging points.