A Novel Three-Dimensional Clustering Method Incorporated with Ellipsoid Model Fitting for Personalizing Myocardial Perfusion Positron Emission Tomography Image Segmentation

Abstract

The quantification of myocardial blood flow (MBF) from cardiac positron emission tomography (PET) has been used for evaluating the abnormalities of coronary flow reserve for patients with known or suspected microvascular disease. We proposed a novel approach to automatic segmentation of the left ventricle (LV) and generating a personalized myocardial model from 82Rb myocardial perfusion positron emission tomography (PET) images using a scheme of 3-dimension (3-D) clustering and ellipsoid model fitting. The new methods were evaluated using 17 patients with normal myocardial perfusion (normal group) and 20 patients with myocardial perfusion defect (abnormal group) who underwent rest and stress 82Rb PET imaging. For comparisons, PET images were processed using k-means with 4 clusters (k-ms4), fuzzy c-means with 4 clusters (FCM4), and fuzzy c-means with 3 clusters (FCM3). The myocardial volume of both rest and stress studies of each patient was calculated. Paired t-test was used for the assessment of statistical significance in the difference between two measures and the Bland-Altman analysis was used to evaluate the agreement between the two measures. The personized myocardial model was successfully generated for all the 37 patients included in this study. The myocardial volume derived from FCM3 were higher than those estimated from the other methods being compared in this study. There were no significant differences between the k-ms4-derived and the FCM4-derived myocardial volume, whereas there were significant differences between the k-ms4-derived and FCM3-derived myocardial volume. The myocardial volume is calculated from FCM4 and FCM3 were also significantly different. Thus, our methods may have a potential of automatically segmenting the LV myocardial regions from myocardial perfusion PET images, whereas the performance of the new automated methods can be limited by the partial volume effect of the heart with a small LV cavity.