Automatic Segmentation and Ventricular Border Detection of 2D Echocardiographic Images Combining K-Means Clustering and Active Contour Model

Accurate analysis of 2D echocardiographic images is vital for diagnosis and treatment of heart related diseases. For this task, extraction of cardiac borders must be carried out. In particular, automatic quantitative measurements of Left Ventricle (LV), Right Ventricle (RV), Left Atrium (LA), Right Atrium, Valve size, etc. are essential. We believe that automatic processing of these echo images could speed up the clinical decisions and reduce human error. In this paper we focus on automatic segmentation of echocardiographic images of different views (Long Axis View, Short Axis View, Apical 4-chamber View) to extract ventricle and atrium borders for detecting heart abnormalities. A novel approach of combining the K-Means clustering algorithm for segmentation and active contour model for boundary detection is proposed. Since conventional K-Means implementation is not time efficient, we propose a novel algorithm called fast K-Means SQL based on (i) TRUNCATE-INSERT instead of DELETE-INSERT for table updates (ii) denormalized database design and tuning (iii) minimal table joins, to accelerate the image segmentation. Thus, with this approach an image of resolution 400×250 takes just 16 seconds, whereas the conventional method takes 950 seconds. After the operator selects the initial contour in the appropriate part of the echocardiographic image, the deformable contour (snake) converges to the boundaries of the region of interest (ROI). Once the shape of the ventricle or atrium is extracted, we apply coordinate geometry to compute all the necessary parameters required for clinical decision. Normally, ultrasound images are embedded with speckle noise; hence we first apply median filter and then the image segmentation. Experiments are conducted using relatively large set of images obtained from a cardiology hospital. The results show that our proposed method is computationally efficient and 2D measurements are accurate.