Automatic coronary calcium scoring using noncontrast and contrast CT images

Area of Science:

• Medical Imaging
• Cardiovascular Disease Research
• Artificial Intelligence in Healthcare

Background:

• Coronary heart disease (CHD) risk assessment relies on coronary calcium scoring.
• Manual detection of coronary artery calcifications in noncontrast CT images is labor-intensive for radiologists.

Purpose of the Study:

• To develop a fully automated method for coronary calcium scoring.
• To alleviate the workload of radiologists and cardiologists in CHD risk assessment.

Main Methods:

• Utilized contrast CT images to segment aorta, heart, and coronary arteries.
• Generated patient-specific regions-of-interest (ROIs) in noncontrast CT images for focused calcification detection.
• Employed a support vector machine classifier to refine results and eliminate noise.

Main Results:

• Achieved high sensitivity (0.989) and positive predictive value (0.948) for calcification volume detection.
• Correctly categorized risk for 39 out of 40 patients based on Agatston scores.
• Algorithm validated on 40 datasets from four CT vendors (MICCAI 2014 orCaScore Challenge).

Conclusions:

• Fully automated detection of coronary calcified lesions in noncontrast CT is feasible.
• Integration with contrast CT segmentation provides highly accurate results.
• The method shows significant potential for clinical application in CHD risk assessment.