Optimization of feature extraction for automated identification of heart wall regions in different cross sections

In most current methods of evaluating the cardiac function based on echocardiography, the heart wall in an ultrasonic image is currently identified manually by an operator. However, this task is very time-consuming and leads to inter- and intraobserver variability. To facilitate the analysis and eliminate operator dependence, automated identification of heart wall regions is essential. We previously proposed a method of automatic identification of heart wall regions using multiple features based on information of the amplitude and phase of the ultrasonic RF echo signal by pattern recognition. In the present study, we investigate a new method of segmenting an ultrasonic image into the heart wall, lumen, and external tissues (includes pericardium) by two-step pattern recognition. Also, parameters in the proposed classification method were examined for application to different cross sections, i.e., long-axis and short-axis views, by considering differences in the motion and echogenicity of the heart walls. Furthermore, moving target indicator (MTI) filtering to suppress echoes from clutters was improved to enhance the separability in the shortaxis view.