| Ultrasound is being increasingly used for diagnosis and treatment of diseases in several fields of medicine, such as, obstetrics, gynecology, urology, cardiology, surgery, etc. In these different areas, segmentation of anatomical features in ultrasound images (detection of organ boundaries from surrounding tissue structure) is routinely performed for clinical decision making. Interpretation of ultrasound images is still open to subjective bias, is tedious and is associated with a steep learning curve. The use of computer aided detection of features in ultrasound images aims to reduce subjective bias, increase productivity and improve cost-effectiveness in clinical decision making.;This research focuses on the development of image segmentation algorithms specifically for medical ultrasound images to solve two clinical problems related to prostate cancer treatment: (1) pubic arch interference assessment, and (2) prostate boundary delineation. In segmentation of ultrasound images, the choice of the algorithm depends on the complexity of the segmentation task and is therefore very application dependent. This dissertation describes a framework for ultrasound image segmentation and statistical validations of the algorithms that provide solutions for the two aforementioned applications.;In the PAI assessment application, an algorithm was developed, tested on 46 patients and its performance was shown to be clinically acceptable. It is currently in routine clinical use at Seattle Prostate Institute, Seattle WA and has been used to predict PAI in over 1500 patients. In the prostate delineation application, we have proposed a concept of edge guidance for delineation and validated it using 125 images. We have shown that with edge guidance the delineations were more consistent while being at least as accurate as the human experts.;The results illustrate that it is possible to use computer-aided ultrasound image segmentation algorithms to augment human experts' ability in arriving at accurate and rapid clinical decisions. |
