Tracking endocardial border motion in ultrasonic images by using neural networks and ARIMA modelling techniques

The problem of tracking cardiac tissue motion in ultrasonic images is studied. This is a very important task in clinical analysis, since it could result in achieving better focusing of ultrasonic scanners and thus in improved diagnosis. Our study is focused on endocardial border motion and two methodologies are employed. Namely, feedforward neural networks and ARIMA modelling techniques. Concerning short term motion tracking, these two approaches give comparable results, while for longer term motion estimation neural networks clearly outperform linear models in capturing the inherently nonlinear dynamics of the process. Although the results presented here are preliminary, the novelty and significance of the study and application should be emphasized