Background: It would be ideal for a non-hyperaemic index to predict fractional flow reserve (FFR) more accurately, given FFR’s extensive validation in a multitude of clinical settings.
Aims: The aim of this study was to derive a novel non-hyperaemic algorithm based on deep learning and to validate it in an internal validation cohort against FFR.
Methods: The ARTIST study is a post hoc analysis of three previously published studies. In a derivation cohort (random 80% sample of the total cohort) a deep neural network was trained (deep learning) with paired examples of resting coronary pressure curves and their FFR values. The resulting algorithm was validated against unseen resting pressure curves from a random 20% sample of the total cohort. The primary endpoint was diagnostic accuracy of the deep learning-derived algorithms against binary FFR ≤0.8. To reduce the variance in the precision, we used a fivefold cross-validation procedure.
Results: A total of 1,666 patients with 1,718 coronary lesions and 2,928 coronary pressure tracings were included. The diagnostic accuracy of our convolutional neural network (CNN) and recurrent neural networks (RNN) against binary FFR ≤0.80 was 79.6±1.9% and 77.6±2.3%, respectively. There was no statistically significant difference between the accuracy of our neural networks to predict binary FFR and the most accurate non-hyperaemic pressure ratio (NHPR).
Conclusions: Compared to standard derivation of resting pressure ratios, we did not find a significant improvement in FFR prediction when resting data are analysed using artificial intelligence approaches. Our findings strongly suggest that a larger class of hidden information within resting pressure traces is not the main cause of the known disagreement between resting indices and FFR. Therefore, if clinicians want to use FFR for clinical decision making, hyperaemia induction should remain the standard practice.
Visual summary. Development and validation of deep neural networks to predict fractional flow reserve (FFR) from resting coronary pressure curves. In a derivation cohort, a deep neural network was trained (deep learning) with examples of resting coronary pressure curves and matching FFR values. After the neural network was trained, its new algorithm was validated using different resting pressure curves. Deep learning-based algorithms did not improve the diagnos-tic accuracy of predicting FFR compared to other non-hyperae-mic indices in a clinically relevant way.