A multi-parametric segmentation tool dedicated to ultrasonic data

Ultrasound images segmentation can bring very useful information for diagnosis but is a difficult task. We propose in this paper a segmentation tool dedicated to ultrasonic data. Our software is based on a segmentation method that involves both multiresolution and multiparametric approaches. A multiparametric approach ensures the robustness of the method. Therefore, the segmentation tool is based on information connected to echo envelope characterization and acoustical characterization, the latter being calculated from the Radio-Frequency (RF) signal. The RF signal allows to calculate local information related to the acoustical properties of the investigated tissues. Several parametric images are build and involved in the segmentation process. The segmentation software is based on a previously presented method (Boukerroui et al. 2001). A Bayesian approach using a multiresolution Markov random field is implemented. The relevance of the several features calculated is shown by the mean of phantoms build with the FIELD II tool. We also present segmentation results on data provided by a technological platform that provides in-vivo multimodality imaging of small animals such as mice and rats. Results on the segmentation of liver mice ultrasonic data are presented.