Machine learning for predicting fractional flow reserve based on optical coherence tomography in intermediate coronary stenosis

Abstract

Abstract Background Machine learning approaches using intravascular optical coherence tomography (OCT) to predict fractional flow reserve (FFR) have not been previously investigated. The objective of the study was to evaluate a machine learning method to estimate FFR based on intravascular OCT images in intermediate coronary lesions. Methods Data from both OCT- and wire-based FFR methods were obtained for lesions of the left anterior descending artery in 125 patients. Based on the total number of lesions, training and testing groups were partitioned at a ratio of 5:1. For the training group, 36 features, including 16 clinical and lesion characteristics, and 21 OCT features, were used to model machine learning-FFR. machine learning-FFR values were then derived for the testing group and compared with wire-based FFR values in terms of a diagnosis of ischemia (FFR <0.8). Results Clinical and lesion characteristics and OCT features between the training and testing groups were similar. During the machine learning modeling of the training group, six important features of machine learning-FFR were identified: minimal luminal area, percentage of the stenotic area, lesion length, proximal luminal area, pre-procedural platelet count, and hypertension. machine learning-FFR values showed a good correlation (r=0.853, P<0.001) with wire-based FFR values (Figure 1A). The diagnostic power of an FFR value less than 0.8, sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of machine learning-FFR values for the testing group were 85.7%, 100%, 100%, 77.8%, and 90.5%, respectively (Figure 1B). Additionally, OCT-based machine learning-FFR values showed a good diagnostic accuracy compared with other image-based FFR values. Conclusions The OCT-based machine learning-FFR method can be used to simultaneously acquire information on both image and functional modalities using one invasive procedure, suggesting that it may be used to optimize treatments for intermediate coronary artery stenosis, as well as save time and cost. Funding Acknowledgement Type of funding source: Public grant(s) – National budget only. Main funding source(s): Funded by the Korean government (MSIT) (no. 2017R1A2B2003191)