A Convolutional-Transformer Model for FFR and iFR Assessment From Coronary Angiography.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Quantifying coronary stenosis severity from X-ray angiography is challenging, relying on subjective visual assessment.
• Accurate assessment of intermediate coronary stenosis is crucial for guiding cardiac interventions.

Purpose of the Study:

• To develop a hybrid CNN-Transformer model for non-invasive assessment of angiography-based fractional flow reserve (FFR) and instantaneous wave-free ratio (iFR).
• To predict the hemodynamic significance of coronary artery stenosis and provide direct FFR/iFR estimates.

Main Methods:

• A hybrid CNN-Transformer model incorporating regression and classification branches.
• Spatio-temporal factorization mechanism to capture local spatial and temporal interactions.
• Training with classification and regression losses using a single angiography view.

Main Results:

• State-of-the-art performance on a dataset of 778 exams (389 patients).
• Robust identification of stenosis location and correlation of prediction uncertainty with output scores.
• Accurate prediction of hemodynamic significance and FFR/iFR values.

Conclusions:

• The proposed hybrid model offers a robust and accurate method for assessing coronary stenosis severity from angiography.
• This approach supports clinical decision-making for cardiac interventions by providing non-invasive FFR and iFR estimates.
• The model demonstrates interpretability and reliable calibration, overcoming challenges in angiographic imaging data.