Influence of computed tomography slice thickness on deep learning-based, automatic coronary artery calcium scoring...

Area of Science:

• Radiology
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Computed tomography (CT) slice thickness impact on deep learning (DL)-based coronary artery calcium (CAC) scoring is unexamined.
• CAC scoring is crucial for cardiovascular risk assessment.

Purpose of the Study:

• To evaluate the influence of CT slice thickness (1.5 mm vs. 3 mm) on DL-based automatic CAC scoring accuracy.
• To compare DL software performance against manual scoring as a reference standard.

Main Methods:

• Retrospective analysis of 844 patients undergoing ECG-gated CAC scoring CT scans.
• Automatic CAC scoring using DL (3D U-Net) and manual scoring as reference.
• Intraclass correlation coefficients (ICCs) and weighted kappa (κ) statistics assessed reliability and agreement.

Main Results:

• Excellent agreement between automatic and manual CAC scoring for both 1.5 mm (ICC 0.982) and 3 mm (ICC 0.969) slices.
• Excellent categorical agreement for CAC severity, with superior results for 3 mm scans (weighted κ 0.961 vs. 0.851).

Conclusions:

• DL-based automatic CAC scoring demonstrates high accuracy irrespective of CT slice thickness.
• 3 mm slice thickness provides slightly better agreement for CAC severity categorization compared to 1.5 mm slices.