Low-Dose Cardiac-Gated Spect Studies Using a Residual Convolutional Neural Network

In recent years low-dose studies have gained increased attention in myocardial perfusion imaging (MPI) using single-photon emission computed tomography (SPECT) owing to its potential radiation risk. However, reducing the injected dose may degrade the image quality and potentially affect the diagnostic accuracy. In this study, we investigate a de-noising approach for low-dose cardiac-gated SPECT images by using a three-dimensional residual convolutional neural network (CNN). The proposed CNN is trained with clinical acquisitions to exploit hierarchically the image features for the task of mapping (noisy) low-dose images into their corresponding standard-dose counterparts. In the experiments, we demonstrated this approach on a set of 119 clinical acquisitions with the imaging dose reduced by four times. The results show that the proposed CNN approach can effectively suppress the overall noise level in the reconstructed myocardium and improve the spatial resolution of the left ventricular (LV) wall. It also outperforms two alternative post-processing methods, including a spatiotemporal non-local means (ST-NLM) filter previously developed for gated SPECT images.