Automatic systole-diastole classification of mitral valve complex from RT-3D echocardiography based on multiresolution processing

Abstract

Mitral valve repair is one of the most prevalent operations for various mitral valve conditions. Echocardiography, being famous for its low-cost, non-invasiveness and speediness, is the dominant imaging modality used for carrying out mitral valve condition analysis in both pre-operative and intra-operative examinations. In order to perform analysis on different phases of a cardiac cycle, it is necessary to first classify the echocardiograhic data into volumes corresponding to the systole and diastole phases. This often requires tedious manual work. This paper presents a fully-automatic method for systole-diastole classification of real-time three-dimensional transesophageal echocardiography (RT-3D-TEE) data. The proposed method first resamples the data with radial cutting planes, segments the mitral valve by thresholding, and removes noise by median filtering. Classification is then carried out based on the number of identified mitral valve regions. A multiresolution processing scheme is proposed to further improve the classification accuracy by aggregating classification results obtained from different image resolution scales. The proposed method was evaluated against the classification results produced by a cardiologist. Experimental results show that the proposed method, without the use of computationally intensive algorithms or the use of any training database, can achieve a classification accuracy of 91.04%.