SPASHT: An image-enhancement method for sparse-view MPI SPECT

Single-photon emission computed tomography for myocardial perfusion imaging (MPI SPECT) is a widely used diagnostic tool for coronary artery disease. However, the procedure requires considerable scanning time, leading to patient discomfort and the potential for motion-induced artifacts. Reducing the number of projection views while keeping the time per view unchanged provides a mechanism to shorten the scanning time. However, this approach leads to increased sampling artifacts, higher noise, and hence limited image quality. To address these issues, we propose sparseview SPECT image enhancement (SPASHT), inherently training the algorithm to improve performance on defect-detection tasks. We objectively evaluated SPASHT on the clinical task of detecting perfusion defects in a retrospective clinical study using data from patients who underwent MPI SPECT, where the defects were clinically realistic and synthetically inserted. The study was conducted for different numbers of fewer projection views, including 1/6, 1/3, and 1/2 of the typical projection views for MPI SPECT. Performance on the detection task was quantified using area under the receiver operating characteristic curve (AUC). Images obtained with SPASHT yielded significantly improved AUC compared to those obtained with the sparse-view protocol for all the considered numbers of fewer projection views. To further assess performance, a human observer study on the task of detecting perfusion defects was conducted. Results from the human observer study showed improved detection performance with images reconstructed using SPASHT compared to those from the sparse-view protocol. The results provide evidence of the efficacy of SPASHT in improving the quality of sparse-view MPI SPECT images and motivate further clinical validation.