Computational methods for non-invasive cardiovascular image analysis
Uday Kurkure · 2008
Non-invasive cardiovascular imaging has found significant utility for extraction of relevant quantitative descriptors of clinical importance for cardiovascular risk prediction and diagnosis. Magnetic resonance imaging (MRI) has been the preferred modality for evaluation of left ventricular function and structure, whereas computed tomography (CT) has been used extensively for coronary artery calcification analysis. However, in clinical practice, the analysis of these images is performed manually, which is tedious, prone to human error and becomes prohibitive with the large amount of data being produced. This dissertation presents a novel multi-class multi-feature fuzzy connectedness method to segment objects with complex appearance and background. In addition, novel computational methods for automated segmentation of the left ventricle and detection of the coronary calcifications from the MRI and CT images, respectively, are presented. Since our results are comparable with those of human experts and demonstrate the possibility of fully automatic analysis of left ventricular functioning and coronary artery calcification, they pave the way for employing automated methods to analyze massive amounts of data from a population screening regimen.