Reconstructing degraded SPECT myocardial images via deep biophysical models: A modern computational approach

Abstract

Myocardial perfusion imaging (MPI) using single-photon emission computed tomography (SPECT) is a critical tool for diagnosing coronary artery disease, but it is often affected by signal degradation due to soft tissue attenuation. In this study, we utilize a publicly available SPECT MPI dataset to establish a benchmark for the task of attenuation correction (AC) by reconstructing AC images from non-attenuation corrected (NC) inputs in a 2D slice-to-slice manner. We implement and compare the performance of several advanced generative models, including generative adversarial networks (GANs) and diffusion models. These models are trained on both general-domain and medical-domain data to evaluate their reconstruction capabilities. The results show that modern deep learning approaches can effectively generate high-quality AC images, demonstrating promising potential for integration into computer-aided diagnosis (CAD) systems for SPECT MPI.