[Artificial intelligence-based automated assessment of coronary flow reserve from angiography and the impact of...

Objectives: To explore the feasibility of a coronary angiography-based method developed with artificial intelligence which was able to automatically and quickly calculate coronary volumetric blood flow and coronary flow reserve (CFR), and explore the differences in CFR after injection of different vasodilators. Methods: This was a observational study screening patients with suspected coronary artery disease who underwent coronary angiography From June to September 2022 in Fuwai Hospital, Chinese Academy of Medical Sciences. Patients without obstructive coronary artery disease in the left anterior descending artery (<50% diameter stenosis by visual) and accompanied by coronary slow flow phenomenon (Thrombolysis in Myocardial Infarction flow grade ≤2) were enrolled. According to pre-specified coronary angiography acquisition protocol, one angiographic image in optimal projection was acquired for each of the following five states: baseline when none of the vasodilators was injected (resting state), intracoronary injection of 200 microg nitroglycerin (nitroglycerin-induced hyperemia), intracoronary injection of 100 microg adenosine (adenosine-induced hyperemia), 5 minutes after cessation of adenosine injection (resting state 2), and intracoronary injection of 4 mg nicorandil (nicorandil-induced hyperemia). Coronary volumetric blood flow and CFR were assessed in a fully automatic manner at an independent core laboratory. One-way repeated measures ANOVA was used to detect the differences in coronary volumetric blood flow at five states and the differences in CFR after injection of different vasodilators. Results: A total of 21 eligible patients were included. The age was (62±9) years, and 5 (24%) were female. Coronary volumetric blood flow at five states and CFR after injection of different vasodilators were successfully calculated in all patients, with a feasibility of 100% (21/21) for CFR. Resting coronary volumetric blood flow was (80.6±12.4) ml/min. Using this as a reference, the volumetric blood flow increased to (167.7±30.5) ml/min under nitroglycerin-induced hyperemia (adjusted P<0.001), and remained at (171.5±23.1) ml/min under adenosine-induced hyperemia (adjusted P<0.001). The volumetric blood flow under resting state 2 was (83.8±15.6) ml/min, returning to baseline level (adjusted P=0.94). Under nicorandil-induced hyperemia, the coronary volumetric blood flow increased again to (182.9±28.3) ml/min (adjusted P<0.001). CFR was 2.09±0.29, 2.15±0.27, and 2.29±0.29 after injection of nitroglycerin, adenosine, and nicorandil, respectively(P=0.034). Using CFR after adenosine injection as a reference, CFR after nicorandil injection was higher (adjusted P=0.044). Using the coronary volumetric blood flow under resting state 2 as the baseline flow for CFR calculation, there was no statistically significant difference compared to the CFR calculated using the volumetric blood flow under resting state (all P>0.05). Conclusions: Preliminary findings confirm the high feasibility of rapid, automated assessment of coronary volumetric blood flow and CFR from a single angiographic projection, as well as good reproducibility in calculating baseline volumetric blood flow. In patients with coronary slow flow, the CFR after nicorandil injection is significantly higher than that after adenosine injection.