Noninvasive and fast method of calculation for instantaneous wave-free ratio based on haemodynamics and deep learning

Abstract

Background and Objectives

Instantaneous wave-free ratio (iFR) is a new invasive indicator of myocardial ischaemia, and its diagnostic performance is as good as the “gold standard” of myocardial ischaemia diagnosis: fractional flow reserve (FFR). iFR can be approximated by iFR_CT, which is calculated based on noninvasive coronary CT angiography (CTA) images and computational fluid dynamics (CFD). However, the existing methods for calculating iFR_CT fail to accurately simulate the resting state of the coronary artery, resulting in low computational accuracy. Furthermore, the use of CFD technology limits its computational efficiency, making it difficult to meet clinical application needs. The role of coronary microcirculatory resistance compensation suggests that microcirculatory resistance can be adaptively reduced to compensate for increases in coronary stenotic resistance, thereby maintaining stable myocardial perfusion in the resting state. It is therefore necessary to consider this compensation mechanism to establish a high-fidelity microcirculation resistance model in the resting state in line with human physiology, and so to achieve accurate calculation of iFR_CT.

Methods

In this study we successfully collected clinical data, such as FFR, in 205 stenotic vessels from 186 patients with coronary heart disease. A neural network model was established to predict coronary artery stenosis resistance. Based on the compensation mechanism of coronary microcirculation resistance, an iterative solution algorithm for microcirculation resistance in the resting state was developed. Combining the two methods, a simplified single-branch model combining coronary stenosis and microcirculation resistance was established, and the noninvasive and rapid numerical calculation of iFR_CT was performed.

Results

The results showed that the mean squared error (MSE) between the pressure drop predicted by the neural network value for the coronary artery stenosis model and the ground truth in the test set was 0.053 %, and correlation analysis proved that there was a good correlation between them (r = 0.99, p < 0.001). With reference to clinical diagnosis of myocardial ischaemia (using FFR as the gold standard), the diagnostic accuracy of the iFR_CT calculation model for the 205 cases was 88.29 % (r = 0.71, p < 0.001), and the total calculation time was < 8 s.

Conclusions

The results of this study demonstrate the utility of a simplified single-branch model in an iFR_CT calculation method based on haemodynamics and deep learning, which is important for noninvasive and rapid diagnosis of myocardial ischaemia.