Session PB7.4

Similarity Measures Evaluation in Contrast Enhanced Echocardiography Motion Detection and Registration

EP Rodrigues, LO Murta*

University of São Paulo
Ribeirão Preto, Brazil

Myocardium wall motion or blood flow analysis can be evaluated from a sequence of contrast echocardiography images, which captures the heart motion as a set of frames. The blood flow is related to contrast replenishment after an echo pulse which destroys the contrast effect on the image and its replenishment can be viewed as pixel intensity behavior. The anatomic pixel location must be the same in a time sequence in order to allow blood flow estimation. Since heart moves elastically, the frames must be aligned by an elastic registration method to allow correct flow analysis.
Elastic registration proposed in this work is a region matching algorithm based on difference entropy similarity metric. This method allows local displacement field determination which is applied to frames sequence alignment, using middle loop frame as template. Although contrast ultrasound images have well defined structures when compared to usual ultrasound, they are still noisy and low resolution, which make it difficult to register. Therefore contrast echocardiographic image registration method must be robust enough to deal with noise, low resolution and elastic deformation to work.
The registration parameters have shown important influence to registration method. A searching window about 31x31 pixels has shown the best results. A Gaussian smoothing in displacement field has shown best results with kernel of size 11x11 pixels with variance of one square pixel, which privileges local corrections. The considered number of bins in image histogram was experimentally accessed in this study, and the best results were obtained for bin size 15 in difference entropy similarity metric. Difference entropy is found to be more precise and robust when compared to mutual information and normalize cross correlation.

(Abstract Control Number: 197)