Automatic thoracic aorta calcium quantification using deep learning in non-contrast ECG-gated CT images

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Thoracic aorta calcium (TAC) assessment from cardiac CT aids cardiovascular risk prediction.
• Accurate TAC quantification is crucial for patient risk stratification.
• Current methods may lack full automation for TAC detection.

Purpose of the Study:

• To develop a fully automatic system for TAC detection using cardiac CT.
• To evaluate the system's performance in classifying patients into four TAC risk categories.
• To leverage deep learning for enhanced cardiovascular risk assessment.

Main Methods:

• Aortic segmentation using a combination of three UNets trained on multi-planar CT images.
• Classification of calcified lesions using three combined Convolutional Neural Networks (CNNs) on orthogonal patches.
• Validation on a dataset of 1190 non-enhanced ECG-gated cardiac CT studies.

Main Results:

• Successful thoracic aorta segmentation with a mean volume difference of 0.3 ± 11.7 ml and a Dice coefficient of 0.947.
• Accurate classification of lesion candidates and risk categories for 87% of patients (Kappa = 0.826, ICC = 0.9915).
• Demonstrated high performance in both segmentation and risk stratification tasks.

Conclusions:

• An automated system combining UNets and CNNs can accurately estimate TAC from cardiac CT.
• This approach enables precise thoracic aorta segmentation and calcified lesion classification.
• The developed system shows significant potential for improving cardiovascular risk prediction through automated TAC analysis.