Detection Of Mental Stress-induced Myocardial Ischemia Using Auto Strain And Deep Learning Assisted Quantitative Myocardial Contrast Echocardiographic Perfusion Imaging

Background:Angina with coronary artery disease(CAD)is a major problem affecting patients with cardiovascular disease worldwide,nearly 70% of whom are diagnosed with no obstructive coronary artery disease(ANOCA)by invasive coronary angiography,and the proportion of female patients is much greater than that of men.Studies have shown that mental factors are closely related to the occurrence and development of CAD,and coronary microvascular dysfunction(CMD)is confirmed to be the common pathogenic mechanism of both ANOCA and mental stress-induced myocardial ischemia(MSIMI).In this study,we aimed to explore the diagnostic value of new techniques,Auto Strain and perfusion quantification with myocardial contrast echocardiography(MCE),for MSIMI involved in female patients with ANOCA by detecting the changes of left ventricular systolic function and myocardial blood flow(MBF)response during laboratory structured mental stress test.We also intended to testify the optimization of quantitative MCE perfusion analysis workflow using automatic myocardial segmentation based on deep learning neural network.Methods : A total of 72 women with ANOCA aged 18-75 years were prospectively recruited,and 23 matched healthy controls were selected to perform echocardiography and positron emission computed tomography(PET)during the structured mental stress test.Traditional left ventricular systolic function evaluation,2D Auto Strain measurement,quantitative MCE perfusion analysis using manual software(Narnar)and automatic myocardial segmentation based on deep learning neural network(U-net+bi-Conv LSTM)were employed during echocardiographic examination.Left ventricular ejection fraction(LVEF),global longitudinal left ventricular strain(LVGLS),plateau signal intensity(A),flux rate constant(β),and semi-quantitative index of MBF(A×β)were assessed respectively.Parameter variations during the mental stress test were observed and ΔGLS,β reserve and A×β reserve were calculated when stress values were compared to those at rest.Receiver operator characteristic(ROC)curves were applied to estimate the diagnostic value of ΔGLS,β reserve and A×β reserve for MSIMI involved in female patients with ANOCA.Results:1)Decrease in LVGLS in MSIMI+ group(P<0.05)by Auto Strain accurately reflected the subtle decline of left ventricular systolic function,while LVEF did not change significantly,during mental stress test in female patients with ANOCA.ROC curve analysis showed that changes of LVGLS could effectively predict MSIMI diagnosed by PET among women with ANOCA [cut point ≤-6.1%,area under the curve(AUC)0.78,P<0.0001].2)Quantitative analysis of MCE perfusion imaging displayed the dynamic MBF changes in female patients with ANOCA during the mental stress test,suggesting the corresponding relationship between MSIMI and reduced or even declined MBF response under mental stress(P<0.05).Parameters of quantitative MCE perfusion analysis by Narnar,βreserve and A×β reserve,showed good diagnostic accuracy in the detection of MSIMI defined by PET(β reserve: cut point ≤0.93,AUC 0.76,P<0.0001;A×β reserve: cut point ≤0.8,AUC0.85,P<0.0001),with A×β reserve helping improve the diagnostic efficiency of LVGLS changes(AUC 0.85,P<0.0001),supporting quantitative MCE as a substitute for routine MSIMI surveillance in women with ANOCA.3)Quantitative MCE perfusion analysis by automatic myocardial segmentation,U-net+bi-Conv LSTM,also revealed the declined MBF response in female patients with ANOCA during the mental stress test(P<0.05).Quantitative parameters,β reserve and A×βreserve,by U-net+bi-Conv LSTM demonstrated good consistency with those by Narnar using Bland-Altman analysis.ROC curve analysis showed that β reserve and A×β reserve by deep learning neural network had good diagnostic performance for MSIMI defined by PET in women with ANOCA(β reserve: cut point ≤1.4,AUC 0.79,P<0.0001;A×β reserve: cut point≤0.6,AUC 0.82,P<0.0001).Conclusions : Our study establishes a set of new echocardiographic methods to accurately assess MSIMI involved in female patients with ANOCA,including Auto Strain,quantitative MCE and deep learning neural network,and provides technical support for early diagnosis and intervention of MSIMI.Auto Strain could greatly improve the tedious workflow and significantly shorten the analysis time for LVGLS compared with conventional manual measurements,and quantitative MCE perfusion imaging could reveal the underlying mechanism of MSIMI with MBF response depicted under mental stress,while automatic myocardial segmentation based on deep learning neural network could greatly reduce the workload of manual segmentation.The study helps to solve the difficulty in MSIMI diagnosis and therefore promotes the early screening and routine monitoring of MSIMI involved in female patients with ANOCA.It also builds a basis for the widespread application of Auto Strain and quantitative MCE techniques in clinical practice.