Performance of a deep learning algorithm for the evaluation of CAD-RADS classification with CCTA

Area of Science:

• Artificial Intelligence in Medical Imaging
• Cardiovascular Disease Diagnosis
• Deep Learning for Radiology

Background:

• Coronary artery disease (CAD) diagnosis relies heavily on imaging techniques.
• Coronary Computed Tomography Angiography (CCTA) is a key diagnostic tool.
• Standardized classification systems like CAD-RADS are crucial for reporting CCTA findings.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) for automated CCTA classification.
• To categorize CCTA images into the appropriate Coronary Artery Disease Reporting and Data System (CAD-RADS) risk stratification.
• To compare the diagnostic performance and efficiency of the AI model against expert human readers.

Main Methods:

• A retrospective study included 288 patients who underwent CCTA.
• CCTA scans were classified by expert readers according to CAD-RADS, serving as the reference standard.
• A deep CNN was trained and tested on the CCTA dataset, analyzing diagnostic accuracy across three classification models and comparing analysis time with on-site readings.

Main Results:

• The deep CNN achieved varying diagnostic accuracies across different CAD-RADS classification models, with Model 1 (CAD-RADS 0 vs CAD-RADS>0) showing 86% accuracy and high specificity (91%).
• Model 2 (CAD-RADS 0-2 vs CAD-RADS 3-5) demonstrated the highest sensitivity at 82%.
• AI-driven analysis time was significantly lower (104.3 ± 1.4 sec) compared to physician reading (530.5 ± 179.1 sec), with p=0.01.

Conclusions:

• Deep CNN models can accurately and efficiently automate the classification of CCTA scans into CAD-RADS categories.
• The AI approach offers a substantial reduction in analysis time, potentially improving workflow in clinical settings.
• Automated CCTA classification using deep learning shows promise for enhancing the diagnosis and management of coronary artery disease.