PET Myocardial Flow Reserve Estimation from 4D-Coronary-CT using Deep Neural Network

The myocardial flow reserve (MFR) index proves to be a highly effective means of assessing the severity of myocardial ischemic disease. An MFR value below two commonly indicates impaired coronary artery perfusion function. Nevertheless, the measurement of MFR using the NH3-positron emission tomography (NH3-PET) method is both invasive and time-consuming. In order to alleviate the burden on both patients and doctors, we aim to achieve accurate estimation of MFR through a coronary computed tomography angiography image processing method. In this paper, we propose 2 methods: one is estimation method of MFR using i) coronary flow index (CFI) with multiple linear regression (MLR) and the other one is ii) vertical-time (VT) image with fully connected network (FCN) and convolutional neural network (CNN). We evaluate the performance between the observed and estimated MFR values with the Pearson correlation coefficient. The results show that LAD = 0.31, LCX = 0.19, and RCA = -0.21 in MLR method; LAD = 0.19, LCX =-0.37, and RCA = 0.23 in FCN and LAD = 0.34, LCX = 0.44 and RCA = 0.37 in CNN. The results exhibit superior performance based on VT images-based method compared to the CFI-based method, which show that VT images method is promising.Clinical relevance- MFR is a crucial indicator in the clinical field of cardiovascular diseases, aiding in the early diagnosis, risk assessment, treatment monitoring, and prognosis evaluation of cardiac conditions. Nevertheless, the measurement of MFR using the NH3-PET method is both invasive and time-consuming. Our results show that estimating the MFR through VT images is a promising method.