AI methods for myocardium reorientation using SPECT perfusion images

Abstract

SPECT technology is one of the most used myocardial detection technologies at present, which is widely used in the diagnosis and treatment of myocardial ischemia and other diseases. However, the reconstructed images of the existing SPECT technology are often subject to a certain deviation between the imaging technology and the geometric positive direction of the left ventricle, which may lead doctors to make some wrong judgments based on this. The usual method is to manually adjust by experienced doctors, but this operation is time-consuming and laborious. On the basis of previous studies, this study proposes an algorithm for automatically redirecting myocardial images based on artificial intelligence. Convolutional neural network combined with attention mechanism is used to extract features of the input images and make regression prediction of parameters, and spatial transformation network is used to help train and finally complete redirection. This study performed well on the data set of 500 patients, and the average prediction accuracy was over 92%.