engleski

Registration and model based analysis of transaortic valvular flow ultrasound images

The dissertation describes a method for registration and model- based analysis of transaortic valvular flow ultrasound images. The image analysis is done from segmented images, where the segmentation is obtained by segmentation propagation from the atlas. The results showed that variability of the automated segmentation relative to the manual is comparable to the intra- observer variability. From the segmentation morphological feature are extracted and it was shown that the symmetry of aortic outflow profile relates to postoperative functional recovery of patients with aortic stenosis, and that it is is better predictor of functional recovery than other routinely used measurements. The accuracy of the segmentation depends on the registration, which in turn depends on the atlas and correct alignment of atlas and image. Thus, methods for atlas formation and definitions of image similarity measure are investigated. We show that the atlas constructed from multiple template images and multiple expert segmentation is fast method for atlas formation which produce atlas whose properties are comparable to atlas obtained by more complex atlas formation methods. It is also discussed under which circumstances the proposed method produces the same result as the classical atlas fusion method. Also, a novel image similarity measure, named absolute joint moments, is proposed. It was shown, that absolute joint moments can be observed as a generalization of the correlation, but also, via cumulant expansion of the probability density function as the approximation of the mutual information. Experimental results also showed that absolute joint moments combine good properties of both correlation coefficient and mutual information.

medical image analysis ; Doppler ultrasound imaging ; cardiac outflow velocity profile ; image registration ; template matching ; model-based segmentation ; segmentation propagation ; atlas formation/construction ; similarity measure ; absolute joint moments

nije evidentirano