PCA-based active contour model for detection and tracking of the left ventricle in apical echocardiographic sequences

摘要

Detecting and tracking the left ventricle in echocardiographic images is a very hard task due to the presence of noise. It is generally done by a cardiologist expert who traces manually the contour for all frames of the sequence representing the whole cardiac cycle. Our aim is to build a computer-aided system for contour detection in order to minimize the human intervention. To do this, we make use of deformable templates which deform in conformation to salient image features. A special case consists in the active contour model (snake). We are interested in applying such models for this kind of images, and in particular for apical views. We propose a new model based on a combination of the active contour model and the PCA (Principal Component Analysis) algorithm. In fact, we add in the definition of the energy associated to the model a new term issued from the PCA done on a training basis of contours. This term is used to help the snake in order to avoid being trapped in wrong positions due to noise. The results obtained are very satisfying, and the snake converges in almost cases. Comparison is done with manually traced contours done by a cardiologist expert, and with the original snake.