| This dissertation concerns the development and validation of an ultrasound imaging system and novel image analysis algorithms applicable to multiple imaging modalities. The ultrasound imaging system will include a framework for 3D volume reconstruction of freehand ultrasound: a mechanism to register the 3D volumes across time and subjects, as well as with other imaging modalities, and a playback mechanism to view image slices concurrently from different acquisitions that correspond to the same anatomical region. The novel image analysis algorithms include a noise reduction method that clusters pixels into homogenous patches using a directed graph of edges between neighboring pixels, a segmentation method that creates a hierarchical graph structure using statistical analysis and a voting system to determine the similarity between homogeneous patches given their neighborhood, and finally, a hybrid atlas-based registration method that makes use of intensity corrections induced at anatomical landmarks to regulate deformable registration. The combination of the ultrasound imaging system and the image analysis algorithms will provide the ability to monitor nerve regeneration in patients undergoing regenerative, repair or transplant strategies in a sequential, non-invasive manner, including visualization of registered real-time and pre-acquired data, thus enabling preventive and therapeutic strategies for nerve regeneration in Composite Tissue Allotransplantation (CTA). The registration algorithm is also applied to MR images of the brain to obtain reliable and efficient segmentation of the hippocampus, which is a prominent structure in the study of diseases of the elderly such as vascular dementia, Alzheimer's, and late life depression. Experimental results on 2D and 3D images, including simulated and real images, with illustrations visualizing the intermediate outcomes and the final results are presented. |
