Comprehensive Image Processing for Automated Detection of Hypertrophic Cardiomyopathy
M. Sowmya Kini, Rajesh K. Pandey, Arnab Das, SK Malani · IJCER · 2014
Abstract — The research work involves image processing techniques and a novel algorithm for automated detection of Hypertrophic Cardiomyopathy (HCM) from M-mode Echocardiography (Echo) images. The automation of the disease has been tried and established in the past using various techniques, however we propose a novel and relatively simple algorithm to achieve the results. The process of automation includes image enhancement, extraction of relevant features, quantification of disease characteristics and arriving at a decision about the presence or absence of HCM. As part of comprehensive image processing we have carried out a comparative performance evaluation of spatialimage enhancement techniques viz Gaussian filtering, Median filtering, Trimmed Mean filtering and Anisotropic Diffusion (AD) filtering in varying Gaussian noise and Speckle noise environments of SNR ranging from 1 to 20dB. The main features of interest in this application are the ‘edges’ – so we have laid sufficient impetus on retaining the edges while applying enhancement techniques to images. Therefore theperformance evaluation was carried out w.r.t two importantcriteria - Edge Keeping Index (EKI) and Mean Squared Error (MSE). The AD filtering technique was found to outperform the other techniques achieving an average EKI as high as 0.98 and a low MSE of 0.15. The proposed algorithm was evaluated using our database of 24 M-mode images of varying SNR values. The database included 16 images belonging to subjects diagnosed with HCM and rest were of subjects with normal cardiac functioning. Out of a total of 24 images, the proposed algorithm was able to arrive at the correct diagnosis for 20 cases indicating a success rate of roughly 83%. The performance of the proposed algorithmwas good even at SNR values as low as 1.2dB in an AWGN environment and 3.6dB in a speckle noise environment.