Automated vessel-specific coronary artery calcification quantification with deep learning in a large multi-centre...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiomics

Background:

• Coronary artery calcification (CAC) is a key indicator of cardiovascular risk.
• Vessel-specific CAC provides additive prognostic value beyond global CAC assessment.
• Automated analysis methods are needed to improve efficiency and accuracy in CAC scoring.

Purpose of the Study:

• To evaluate the accuracy of a deep learning (DL) model for vessel-specific CAC quantification.
• To assess the prognostic implications of DL-derived vessel-specific CAC for major adverse cardiovascular events (MACE).
• To validate the DL model on both electrocardiogram (ECG)-gated and attenuation correction (AC) computed tomography (CT) datasets.

Main Methods:

• A DL model was trained on 3000 gated CT scans and tested on 2094 gated and 5969 non-gated AC CT scans.
• Vessel-specific CAC was analyzed in the left main/left anterior descending (LM/LAD), left circumflex (LCX), and right coronary artery (RCA).
• Agreement with expert assessment was measured using linear weighted Cohen's Kappa; MACE risk was assessed via hazard ratios (HR) during follow-up.

Main Results:

• The DL model demonstrated strong to excellent agreement with expert ground truth for vessel-specific CAC on both gated and AC CT.
• Higher CAC burden in the LM/LAD (>400 Agatston units) was significantly associated with the highest risk of MACE on both CT types.
• The model showed good performance across different CAC severity categories and coronary territories.

Conclusions:

• Automated, vessel-specific CAC assessment using DL is accurate and rapid on both gated and AC CT.
• DL-based vessel-specific CAC analysis offers significant prognostic value for predicting MACE.
• This technology has the potential to improve cardiovascular risk stratification in clinical practice.