Calcium Scoring at Coronary CT Angiography Using Deep Learning.

Area of Science:

• Radiology
• Cardiovascular Imaging
• Artificial Intelligence in Medicine

Background:

• Separate noncontrast CT scans are often used to quantify coronary artery calcium (CAC) scores before coronary CT angiography (CTA).
• This dual-scan approach increases radiation exposure for patients.
• Quantifying CAC directly from CTA scans would reduce radiation but presents technical challenges.

Purpose of the Study:

• To develop and validate a deep learning method for automatic CAC scoring directly from a single CTA scan.
• To eliminate the need for a separate noncontrast CT scan for CAC quantification.

Main Methods:

• A deep learning algorithm was trained and validated on 292 and 73 patient scans, respectively.
• Virtual noncontrast images generated from spectral CT data were used for algorithm development.
• The method was validated on an independent test set of 240 CTA scans using comparison with semiautomatic Agatston scores from noncontrast CT.

Main Results:

• The deep learning CTA CAC scores showed excellent correlation with semiautomatic noncontrast CT CAC scores (Pearson correlation = 0.96, r² = 0.92).
• Cardiovascular risk categorization agreement was excellent (weighted κ = 0.94), with 93% of scans correctly categorized.
• No statistically significant differences were observed based on scanner, sex, or section thickness.

Conclusions:

• A deep learning-based automatic CAC scoring method accurately quantifies coronary artery calcium from CTA images.
• This method effectively categorizes cardiovascular risk, comparable to traditional noncontrast CT methods.
• The approach offers a radiation-sparing alternative for CAC assessment.