This study investigates the influence of CT slice thickness on the accuracy of deep learning (DL)-based automatic coronary artery calcium (CAC) scoring software. The retrospective study included 844 subjects who underwent ECG-gated CAC scoring CT scans with 1.5 and 3 mm slice thicknesses. Using the commercial DL-based software (AVIEW CAC), which employs a 3D U-Net architecture, automatic CAC scoring was compared to manual CAC scoring, the reference standard. The reliability of the automatic CAC scoring was evaluated using intraclass correlation coefficients (ICCs). Results showed excellent agreement between automatic and manual scoring, with ICCs of 0.982 for 1.5 mm and 0.969 for 3 mm datasets. The agreement on CAC severity categories (Agatston scores) was also excellent, with weighted kappa (κ) statistics of 0.851 for 1.5 mm and 0.961 for 3 mm scans, indicating better agreement for thicker slices. The study concludes that automatic CAC scoring is highly reliable, though slightly less so for thinner slices.
목록으로