Coronary physiology instantaneous wave-free ratio (iFR) derived from x-ray angiography using artificial intelligence
Miguel Nobre Menezes1, João Lourenço Silva2, Beatriz Silva1
1Structural and Coronary Heart Disease Unit, Cardiovascular Center of the University of Lisbon (CCUL@RISE), Faculdade de Medicina, Universidade de Lisboa, Lisbon, Portugal; Serviço de Cardiologia, Departamento de Coração e Vasos, CHULN Hospital de Santa Maria, Lisbon, Portugal.
The Journal of Invasive Cardiology
|March 5, 2024
Summary
Artificial intelligence (AI) models were developed to estimate instantaneous wave-free ratio (iFR) from coronary angiography (CAG) images. These models show promise in potentially reducing invasive procedures due to their high negative predictive value.
Area of Science:
- Cardiovascular medicine
- Medical imaging analysis
- Artificial intelligence in healthcare
Background:
- Coronary physiology assessment traditionally relies on invasive methods.
- Current coronary angiography (CAG)-derived physiology methods often use fluid dynamics algorithms.
- Artificial intelligence (AI) approaches for coronary physiology are underexplored.
Purpose of the Study:
- To develop and evaluate AI models for estimating instantaneous wave-free ratio (iFR) from CAG images.
- To explore an alternative to traditional computational fluid dynamics algorithms for coronary physiology.
- To assess the feasibility of AI in simplifying and increasing the use of coronary physiology assessments.
Main Methods:
- Developed three distinct AI models to classify coronary lesions based on iFR values (≤0.89 positive, >0.89 negative).
- Trained and tested models using data from consecutive patients who underwent invasive iFR measurements.
- Compared AI model predictions against true iFR measurements to evaluate performance metrics.
Main Results:
- Three AI models were developed, with Model 3 showing the best overall performance (accuracy 69%, NPV 88%, PPV 44%).
- Performance varied by coronary vessel; for instance, models showed high accuracy and NPV for the right coronary artery but low PPV across all vessels.
- Sensitivity and specificity also varied, with specific models excelling in certain vessels (e.g., Model 2 for LAD sensitivity, Model 1 for RCA specificity).
Conclusions:
- AI models were successfully developed for binary iFR estimation from CAG images.
- The high negative predictive value (NPV) of these models suggests potential to avoid unnecessary invasive procedures.
- This study provides proof of concept for AI in coronary physiology, warranting further research and development.
