Effectiveness of Initial Segmentation Technique for Mitral Valve in Off-Line Cardiac Ultrasound Images

Abstract

The current MATLAB-based mitral valve segmentation method is manual, allowing users to define the region of interest and adjust the segmenting angle, but it suffers from being operator-dependent, leading to inconsistent and low reproducibility results. Additionally, edge-based segmentation methods are prone to over and under-segmentation issues. This project addresses these drawbacks by implementing an automatic clustering technique for mitral valve segmentation. Clustering partitions pixels based on feature similarity discover inherent data structures without requiring prior training and avoid the complexity and slow convergence associated with spatially dependent methods like region growth, potentially improving segmentation accuracy and efficiency. This project evaluates the effectiveness of an initial segmentation technique for the mitral valve in random cardiac ultrasound images, focusing on realigning potential risks, analysing, and understanding valve movement through clustering-based segmentation methods. The study preprocesses images using Canny-Edge Detection and applies clustering techniques for segmentation, followed by developing a GUI prototype using MATLAB for user performance testing. Results demonstrate mitral valve movement over four cardiac cycles in healthy and unhealthy hearts, measuring diameters during the opening and closing phases. Healthy hearts showed a minimum opening diameter of 0.32 cm and closing diameter of 0.30 cm, while unhealthy hearts had a minimum opening diameter of 0.25 cm and closing diameter of 0.28 cm. This project enhances the understanding of image processing in cardiac health assessment, particularly non-invasive mitral valve evaluation using segmentation techniques.