Deep learning-based myocardial segmentation using non-contrast-enhanced CT for quantitative SPECT myocardial perfusion imaging

Accurate myocardial segmentation is essential for polar map-based quantitative analysis in SPECT myocardial perfusion imaging (MPI). Commercial software requires significant manual adjustment to ensure accurate segmentation, particularly for studies with severe myocardial defects or high extracardiac uptake. To address this limitation, we propose deep learning (DL)-based myocardial segmentation using non-contrast-enhanced CT images.

Myocardial masks were generated by simply thresholding PET images with uniform myocardial uptake in 125 ¹⁸F-FDG PET/CT studies. An optimal network was configured through the nn-UNet pipeline with CT images as input and the masks as labels. For evaluation, the network was tested on 50 ^99mTc-sestamibi SPECT/CT MPI studies with uniform myocardial uptake. Additional 50 SPECT MPI studies with severe myocardial defects or high extracardiac uptake were selected to demonstrate the impact of using DL-masks on polar map-based quantification. Three nuclear cardiologists visually compared myocardium contours with and without DL-masks in commercial software.

DL-masks achieved a Dice Similarity Coefficient of 0.866 ± 0.035 in the PET/CT validation set and 0.846 ± 0.026 in the SPECT/CT test set. For additional studies, segmental differences in the polar map with and without DL-masks indicated meaningful quantitative differences (~ 10%) in basal segments near the septum, compared to mid-cavity or apical segments. Likert scoring demonstrated that DL-masks improved septal and valve plane placement in commercial software, resulting in accurate myocardial segmentation without manual adjustment.

We demonstrated a DL segmentation model for accurate myocardial segmentation without manual adjustment and corresponding polar map-based quantification, particularly in studies with severe myocardial defects or high extracardiac uptake.