| Description |
Statistical shape analysis plays a crucial role in computer vision and medical imaging applications. Shape-based models help improve the performance of object detection/ segmentation tasks and also provide gain critical insights for medical science research. Over time, researchers have adopted various shape modeling strategies, varying from primitive geometry-based models to data-driven machine learning approaches. This dissertation focuses on generative modeling techniques for the statistical shape analysis. Three chapters (2-4) in this dissertation summarize the four research contributions. Chapter 2 presents a Bayesian approach to generate a probabilistic summary from volumetric shape representations. The proposed approach has shown favorable results for varied medical imaging applications. Chapters 3 and 4 lay out the research contributions involved with surface-based shape models. Specifically, the third chapter showcases the technical contributions of surface-based shape models for the purpose of orthopedic research. The fourth chapter focuses on state-of-the-art deep learning algorithms to help generate better statistical shape models. The machine learning approach has helped alleviate the need for manual effort, often performed by clinicians, for the analysis of medical images. The primary contribution of this dissertation is to improve statistical shape modeling, using classic Bayesian learning and currently favored deep learning methods. |
