CARDIAC MRI SEGMENTATION METHOD BASED ON MASKS LOCALIZATION

Abstract

MRI analysis of cardiac images is a time-consuming process that requires significant effort and time. This is due to the complexity of cardiac anatomy, the variability of images, and the need for high accuracy in detecting pathologies. Automation of MRI analysis processes is important for improving diagnostic efficiency. High-quality segmentation of MRI cardiac images can serve as a basis for further research and clinical applications.

This study proposes a novel approach to myocardial segmentation in MRI images, which includes three stages: localization, mask generation, and post-processing. The first stage involves the decomposition of masks into separate binary masks for myocardium, left and right ventricles. The second stage involves detailed mask generation based on localized images to ensure accurate determination of the contours of the heart structures. The third stage involves post-processing of the masks to smooth transitions between pixels and preserve details when resizing the image.

To evaluate the accurancy of the results obtained, a series of experiments were conducted using the Automated Cardiac Diagnosis Challenge dataset. Using the proposed approach, it was possible to achieve an accuracy of segmentation of the right ventricle of 0.974, the left ventricle of 0.947, and the left ventricular myocardium of 0.896 for End diastole and the right ventricle of 0.94, the left ventricle of 0.915, and the left ventricular myocardium of 0.92 for End systole.

The proposed approach has the potential for widespread use as a basis for further MRI image research, ensuring the accuracy of cardiac segmentation in MRI images.