Using echocardiography based deep learning to early detect the infarct related artery in patients with acute coronary syndrome

Abstract

Abstract Funding Acknowledgements Type of funding sources: None. Background In patients with acute coronary syndrome (ACS), echocardiography detected regional wall motion abnormalities (RWMA) facilitates the recognition of ischemic heart disease and infarct related artery. Nevertheless, the differentiation of RWMA relies on the experiences of performers. Notably, in ACS patients without transmural infarction, RWMA may not be visible upon naked eyes. Purpose This study aims to investigate whether the application of 3D Convolution Neural Network could assist clinicians to differentiate patients with and without ACS based on echocardiography detected RWMA. Methods From 2018 to 2021, we collected echocardiographic imaging in 796 patients without ACS (Normal Control; NC), 759 with ACS and detectable RWMA (RWMA) and 267 with ACS but not detectable RWMA (uncertain; UC). The diagnosis of ACS was defined by the obstructive coronary arterial disease (CAD) in coronary angiography. Apical four, two and long chamber viewer were acquired and RWMAs were defined by cardiologists. Cardiac-Echo Net consists the techniques of 3D Convolution Neural Network and 3D MaxPooling. Results After exclusion echocardiographic imaging not qualified for analysis, we collected 40813 and 5928 images for establishing the model of Cardiac-Echo Net. In the final model, areas under the receiver operating characteristic curve are 98.9 and 89.2% for the training and validation, respectively. In the external validation dataset, the sensitivity was 81.8% and specificity was 81.6%. Notably, compared with cardiologists, Cardiac-Echo Net showed a superior accuracy in differentiating NC from RWMA (0.89 v.s. 0.815). Likewise, in differentiating NC from UC, Cardiac-Echo Net has a persistently higher accuracy than cardiologists (0.87 v.s. 0.65). Conclusions Superior to previous deep learning models, this novel one combined several neural-networking from different fields. Cardiac-Echo Net could spontaneously detect the subtle myocardial ischemia in ACS patients without eye-catching RWMA while further external validation is necessary.