A fast K-Means algorithm for the segmentation of echocardiographic images using DBMS-SQL

An efficient K-Means clustering algorithm is proposed using the power of SQL in a relational Database Management environment. Further, this method is applied to segment 2D echocardiography images. We propose this method mainly to improve the speed of segmentation process for further clinical analysis and diagnosis (example: Left Ventricular (LV) boundary detection and other 2D quantitative measurements) of a patient. With few initial image processing steps like median filtering, etc, our proposed K-Means SQL algorithm is applied on the image to identify the segments. Since segmentation alone cannot resolve the anatomical ambiguities present in echocardiogram images, few post segmentation steps like median filtering, Canny edge detection operator, Gaussian sharpening, and morphological operations are applied to obtain venrtricle borders. We identify steps that would normally take more time in conventional K-means algorithm and provide appropriate alternatives in SQL. We compare the conventional K-means algorithm with K-Means SQL algorithm for a set of sample echocardiogram images and the experiments show that our method is superior in terms of quality of segmentation, running time, and scalability.