Deep Learning-based Automated Coronary Plaque Quantification: First Demonstration With Ultra-high Resolution...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Coronary artery disease diagnosis relies on accurate plaque quantification.
• Ultra-high resolution CT angiography (UHR CCTA) offers detailed plaque visualization.
• Deep learning (DL) tools show promise for automating complex image analysis.

Purpose of the Study:

• To assess the feasibility and reproducibility of a novel DL tool for coronary plaque quantification.
• To evaluate the impact of temporal resolution on plaque quantification accuracy.
• To validate an automated workflow for UHR CCTA plaque analysis.

Main Methods:

• Retrospective analysis of 45 UHR CCTA scans.
• Reconstruction of datasets at 66 ms and 125 ms temporal resolutions.
• Application of a DL algorithm for automated coronary segmentation and plaque quantification.

Main Results:

• The DL algorithm demonstrated high reproducibility and required no manual correction.
• Lower temporal resolution (125 ms) systematically overestimated plaque volume and diameter stenosis compared to 66 ms.
• Significant differences in plaque volume (P<0.05) and stenosis (P<0.01) were observed between resolutions.

Conclusions:

• The novel DL tool is robust and reproducible for coronary plaque quantification in UHR CCTA.
• Temporal resolution significantly influences plaque quantification, with lower resolution leading to overestimation.
• Standardized imaging protocols are crucial for reliable DL-based plaque analysis.