| The objective of this dissertation is to develop innovative schemes to extract useful and reproducible information from medical image sequences to help physicians with clinical diagnosis as well as physiological and pathological studies. Two types of very commonly encountered problems are addressed. The first problem is the extraction and reconstruction of structure of interest when the available intensity information is insufficient. To resolve this problem, shape information is incorporated. We extract the left ventricle chamber by the fusion of an adaptive K-means clustering method and active contour models, and reconstruct the airway trees by incorporating topology analysis. The second problem is the evaluation of motion and deformation, which are usually tightly coupled with organ functions, and are considered as sensitive indicators of diseases. Physics-based models are proposed for cardiac motion analysis to take advantage of their abilities to characterize the physical process. Continuum mechanics theory is incorporated to develop an algorithm for warping and registering lung volumes at different breathing stages. Furthermore, the lung-warping model is extended to assess the clinically significant structure-function relationship of ventilation. Experimental results show that these proposed schemes are promising in various biomedical image processing applications. |
