A Fully Automated Deep Learning Model for Quantifying Coronary Plaque at Coronary CT Angiography

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Deep learning (DL) models for coronary plaque quantification via coronary CT angiography (CCTA) are underutilized in clinical practice.
• Developing automated DL tools is crucial for advancing cardiovascular diagnostics.

Purpose of the Study:

• To create a fully automated DL model, PlaqueSegNet, for coronary plaque volume (PV) quantification.
• To assess the prognostic capability of PlaqueSegNet in predicting major adverse cardiac events (MACEs).

Main Methods:

• Developed PlaqueSegNet using a training dataset of 1409 patients undergoing CCTA.
• Externally validated the model on four independent datasets, including paired CCTA/intravascular US (IVUS) and diverse CT scanner data.
• Evaluated prognostic value using the Harrell C-index in three distinct patient cohorts.

Main Results:

• PlaqueSegNet demonstrated high agreement and reproducibility (intraclass correlation coefficients >0.90) for PV quantification compared to IVUS and expert readers.
• The model achieved C-indices ranging from 0.64 to 0.74 for predicting MACEs across different cohorts with median follow-ups of 2.3 to 5.3 years.
• The DL model showed strong performance across various datasets, including those with different CT scanners and imaging techniques.

Conclusions:

• PlaqueSegNet offers fully automated, accurate plaque volume measurements from CCTA.
• The model's quantification closely aligns with expert assessments and IVUS.
• PlaqueSegNet possesses significant prognostic value for future MACEs, supporting its potential clinical utility.