Development and validation of a novel deep learning coronary artery plaque quantification model

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary Computed Tomography Angiography (CCTA) is vital for assessing coronary artery disease.
• Manual plaque volume quantification is time-consuming and impractical for routine clinical use.
• Automated plaque volume assessment using deep learning offers a promising alternative.

Purpose of the Study:

• To develop and validate QuantiPlaque, a novel deep learning software for automated segmentation of total and calcified coronary plaque volume.
• To assess the accuracy and reliability of QuantiPlaque compared to expert manual annotation.

Main Methods:

• Development of a deep learning model for automated plaque segmentation.
• Validation using 115 CCTA scans with expert manual annotation as the reference standard.
• Statistical analysis including intraclass correlation coefficient (ICC), Pearson's r, Spearman's rho, and Bland-Altman analyses.

Main Results:

• The deep learning model demonstrated strong correlation and agreement with expert annotations for both total plaque volume (ICC 0.95) and calcified plaque volume (ICC 0.90).
• High correlation coefficients (Spearman's rho ≥ 0.87, Pearson's r ≥ 0.90) were observed for per-patient analyses.
• Per-vessel analysis showed strong correlation in LAD and RCA, with weaker correlation in the LCx territory.

Conclusions:

• The QuantiPlaque deep learning model accurately quantifies total and calcified coronary plaque burden.
• The software exhibits strong correlation and agreement with expert annotations, making it suitable for routine clinical use.