Color doppler echocardiographic image analysis via shape and texture features

Doppler imaging allows evaluation of blood flow patterns, direction, and velocity. The color (red, blue, and mosaic) signify the direction of the blood flow. By analyzing this color Doppler, it is possible to detect heart diseases like mitral and aortic stenosis, mitral, tricuspid, and aortic regurgitation, and Left Ventricle (LV) hypertrophy. We present 3 methods to extract low level features namely color histogram mean, standard deviation, skewness, kurtosis, and texture features such as energy, entropy, contrast, homogeneity of the region of interest (ROI) in a color Doppler echocardiographic image. The first method is based on conventional K-Means algorithm to segment the image. A modified fast K-Means implemented using SQL is the second method presented in this paper. Finally, segmentation is achieved through pixel classification approach which is found being the most efficient. The proposed technique decomposes the image into foreground pixels representing color information and background pixels which are grayscale. Thus, we apply morphological operations, Gaussian blur, and threshold to obtain a distinct object for quantitative measurements. Without doing any modifications to the foreground pixels, we compute histogram statistics of the shape and texture features. Hence, with our proposed method both qualitative and quantitative analysis can be done. Our technique is applied on variety of color Doppler patient images and the results show that it is computationally efficient and abnormality detection is satisfactory.