Deep learning based automatic quantification of aortic valve calcification on contrast enhanced coronary CT...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Aortic valve calcification quantification is crucial for assessing aortic stenosis severity and cardiovascular risk.
• Current methods often rely on non-contrast CT, which may not always be available or optimal.
• Accurate calcium scoring guides treatment decisions in patients with aortic stenosis.

Purpose of the Study:

• To evaluate a deep learning-based automated method for quantifying aortic valve calcification using contrast-enhanced coronary CT angiography.
• To compare the performance of the automated method against manual calcium scoring.
• To assess the feasibility of this approach as an alternative to traditional non-contrast CT methods.

Main Methods:

• A retrospective analysis of 177 patients was performed, with data split into development and validation sets.
• A DeepLab v3+ model segmented the aorta, and an XGBoost model refined the aortic valve region.
• Calcifications were identified and quantified using a tailored threshold and a weighted scoring method analogous to the Agatston score.

Main Results:

• The automated method demonstrated excellent agreement with manual Agatston scores (Pearson's r = 0.93, CCC = 0.92).
• For classifying severe aortic stenosis, the approach achieved 88.6% sensitivity, 91.1% specificity, and 90.0% overall accuracy.
• High accuracy was observed in both development and validation datasets.

Conclusions:

• The deep learning model enables accurate, automated quantification of aortic valve calcification from contrast-enhanced CT.
• This approach provides a viable alternative when non-contrast CT is unavailable, potentially reducing operator dependency and radiation exposure.
• The findings support the integration of AI-driven tools for efficient cardiovascular risk assessment.