Deep learning attenuation correction (TruCorr) versus dual-position SPECT imaging for myocardial perfusion: diagnostic agreement and workflow benefits
Y B Brodov, S T Tiosano, E R Rosen, N A Naimushin, R G GoldkornAbstract
Background
Dual-position (upright and supine) stress imaging is commonly used to mitigate attenuation artifacts but increases scan time and often necessitates rest imaging. TruCorr is a novel deep learning–based attenuation correction (AC) method that generates corrected images from a single-position acquisition. This study evaluated whether single-position stress MPI with TruCorr AC provides diagnostic performance equivalent to standard dual-position stress imaging and reduces the need for rest scans.
Methods
We analyzed 200 patients who underwent clinically indicated stress/rest MPI on a CZT SPECT (D-SPECT) camera. Stress imaging was acquired in upright and supine positions, followed by supine rest imaging per standard protocol. Supine stress images were further processed with TruCorr. Three experienced nuclear cardiologists (blinded to clinical data) independently interpreted studies using two approaches: (1) dual-position non-AC stress images and (2) single-position TruCorr stress images. Perfusion was graded using a 5-point diagnostic confidence scale; image quality, attenuation artifacts as well as the need for rest imaging were recorded. Interpretation time, scan time, and quantitative total perfusion deficit (TPD) were compared.
Results
Diagnostic agreement between single-position TruCorr and dual-position non-AC imaging was high, with 95% concordance in normal/abnormal classification (κ=0.92). TruCorr significantly reduced attenuation artifacts (10% vs 50%, p<0.001) and improved image quality (4.1 vs 3.8), p=0.002). Equivocal interpretations decreased from 10% to 5% (p=0.03). Rest imaging was recommended less frequently with TruCorr (50% vs 80%, p<0.001), representing a 37.5% relative reduction. Interpretation time was similar between methods, while stress scan time was reduced by approximately 50% with the single-position protocol. Quantitative TPD showed excellent correlation between methods (R²=0.95). Polar maps of non-corrected and TruCorr corrected images are presented in picture below.
Conclusions
Single-position stress MPI with TruCorr deep learning AC provides diagnostic performance equivalent to dual-position imaging while improving image quality and reducing attenuation artifacts. This approach substantially decreases the need for rest imaging and scan time, supporting more efficient stress-only MPI protocols without compromising diagnostic accuracy.