Deep Convolutional and Recurrent Neural Networks for Detection of Myocardial Ischemia Using Cardiodynamics gram

摘要

Cardiodynamicsgram (CDG) has emerged recently as a new noninvasive electrocardiographic method for early detection of myocardial ischemia (MI). There exists a clear relationship between CDG morphology and MI. However, it is still challenging to visually interpret the relationship between the morphology of CDG and the degree and location of MI. Therefore, we propose a novel deep learning framework to automatically detect the myocardial ischemia by using CDG. In this study, we implemented a convolution and recurrent neural networks algorithm for three dimensional CDG data series classification. The CDGs can be obtained by using the deterministic learning and the hadoop parallel. Finally we achieved an average sensitivity of 91.7% and specificity of 81.5% in National Center for Cardiovascular Diseases. The proposed method is expected to provide a real-time software tool towards assisting the physician in cardiology departments for the early detection of ischemic heart diseases.