Segmentation of coronary calcifications with a domain knowledge-based lightweight 3D convolutional neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Cardiovascular diseases are a leading global cause of mortality.
• Coronary artery calcifications (CACs) are key biomarkers for cardiovascular disease risk.
• Manual CAC segmentation is time-consuming and variable; automated methods using deep learning are advancing.

Purpose of the Study:

• To develop and validate a novel, automated approach for CAC segmentation and calcium scoring using a lightweight 3D convolutional neural network.
• To address the limitations of existing methods by incorporating anatomical context for improved differentiation of coronary arteries.

Main Methods:

• A lightweight three-dimensional convolutional neural network (3D CNN) architecture was developed for automated CAC segmentation.
• The model was trained and evaluated on computed tomography (CT) datasets for segmentation and calcium scoring.
• Performance was quantified using Dice score coefficients and validated on an external cohort.

Main Results:

• The proposed 3D CNN achieved high Dice scores for CAC segmentation: 0.93 (foreground), 0.93 (LAD), 0.93 (LCX), 0.84 (LM), and 0.89 (RCA).
• The method outperformed existing state-of-the-art architectures in accuracy and efficiency.
• External validation confirmed the model's generalization capabilities across different clinical scenarios.

Conclusions:

• The lightweight 3D CNN offers an efficient and accurate automated solution for CAC segmentation and calcium scoring.
• This approach surpasses current state-of-the-art methods and demonstrates significant potential for clinical application in cardiovascular risk assessment.
• The model's robust generalization highlights its utility in diverse healthcare settings.