Deep learning-based detection of functionally significant stenosis in coronary CT angiography

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Intermediate coronary artery stenosis requires functional significance assessment.
• Invasive fractional flow reserve (FFR) measurement is the gold standard but costly and burdensome.
• Non-invasive FFR prediction from coronary CT angiography (CCTA) offers a potential alternative.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) method for non-invasively predicting FFR from CCTA scans.
• To assess the DL method's performance in identifying functionally significant coronary artery stenosis.
• To reduce the need for invasive FFR measurements in clinical practice.

Main Methods:

• A deep learning model was trained using CCTA scans from 569 patients.
• The model extracts coronary artery features, including lumen, attenuation, and calcium, using a CNN.
• A second network predicts FFR and classifies stenosis significance, integrating tree characteristics like bifurcations.

Main Results:

• The DL method achieved an Area Under the Receiver Operating Characteristics Curve (AUC) of 0.78.
• Performance was validated on multi-center, multi-vendor test sets.
• The method outperformed existing non-invasive approaches not requiring manual segmentation correction.

Conclusions:

• The proposed DL method accurately predicts FFR from CCTA, aiding in functional significance assessment.
• This non-invasive approach shows potential to decrease unnecessary invasive FFR procedures.
• The study highlights the utility of AI in non-invasive cardiac diagnostics.