Robust extraction of input function from H_2(~(15)0) dynamic myocardial positron emission tomography using independent component analysis

摘要

It is hard to extract the input function from the left ventricle on H_2(~(15)0) PET image to estimate the regional myocardial blood flow (rMBF). In this study we applied blind source separation technique by Independent Component Analysis (ICA) to extract input function from the H_2(~(15)0) dynamic myocardial PET. Dynamic PET scans were performed on five dogs at rest and dipyridamole-induced stress. A transverse slice containing the biggest heart was selected and masked so that only the cardiac components were included in the analysis. We assumed that the elementary activities corresponding to the left and right ventricular pool and myocardial tissue as independent sources since their anatomical structures are not overlapped in the space. After principal component analysis, ICA unmixing process was performed using the extended infomax learning algorithm and left ventricular input function was obtained. In all the cases, we could extract the input functions, which had identical shape with those obtained using the manually drawn region of interest, and the rMBFs obtained using them were correlated well (r=0.87, p=0.OOl). Since all the process was, moreover, automatically achieved with very short computation time, it will be useful for the quantification of the rMBF using H_2(~(15)0) PET.