Ultrasound View Detection

Abstract

The goal of this project is to establish effective ultrasound view detection between apical, parasternal long axis and parasternal short axis view. The chosen approach in this project is to represent a shape of the cavity visible on the echocardiographic view by a model. The model is based on a deformable non-uniform rational B-spline algorithm, which can follow the structures of ultrasound recordings by the use of an extended Kalman filter framework. After the cavity is represented by the model, this model is evaluated by global shape measures. The global shape measures are good shape descriptors with reasonable computational requirements. The tested ultrasound recordings were divided into training set and final set. Both the training set and the final set consist of 56 realistic ultrasound recordings with equal distribution of apical and parasternal view recordings. In total, seven global shape measures were selected with respect to the anticipated shape differences between the views. The global shape measures were calculated and evaluated using the training set. The results have been analyzed for possible shape discrimination between the views. From this analysis the area of the models results as the suitable parameter for the classification between the apical and the parasternal view recordings. No global shape measure capable of classification between the parasternal long axis and the parasternal short axis view was found. The proposed classification was tested using the final set recordings. The success rate of this classification in the final set was 83.9 %.