Backpropagation Neural Network for Motion Analysis on Blood-pool Gated Single Photon Emission Computed Tomography

We used backpropagation neural network for left ventricular motion analysis on Tc-99m labeled RBC blood-pool gated single photon emission computed tomography (GSPECT). Phantom images by the model of solid spheres were generated to simulate the left ventricle. Training data sets were selected from the phantom images. After training, the neural network can perform motion analysis on the phantom images and all series of patients´ GSPECT images. The results of motion analysis were displayed in the formats of vector fields superimposed on the original GSPECT images. The GSPECT of one patient with normal left ventricle and two patients with abnormal left ventricular motion were acquired and analyzed. The study showed that back propagation neural network was useful in the evaluation of left ventricular motion in GSPECT images