Deep Learning-based Feature Representation For Myocardial Ischemia Classification

Cardiovascular diseases have been a fatal threat to human health, among which many are related with myocardial ischemia. So it’s very important to diagnose myocardial ischemia, which also has impact on therapy and prognosis. With the development of medical imaging technologies, quite a number of them have been widely used in clinical applications, such as electrocardiogram (ECG) and magnetic resonance imaging (MRI). It’s proved that these technologies play an important role in aiding diagnosis.In the context of abundant medical data, it has been an increasingly urgent problem to tackle these data effectively. Traditionally, people turn to clinical empirical judgment or automatic analysis of medical data. These ideas have limitations, for example, it will be such a tedious work for doctors if dealt manually, also automatic analysis has been concentrating on abstraction of general features while always closes eyes on personalized information.To this end, a novel algorithm based on deep learning is proposed, and basically this algorithm makes full use of multi-modality medical data and enables feature extraction with multiple aspects. The main contribution of this work can be summarized as(1) Setup a union framework for data collection, including 64-leads body surface potential records and computed tomography, and build personalized heart-torso model sequentially. Model the cardioelectric field with boundary element method, introduce high-order total variation to the ECG inverse problem, then solve the ECG inverse problem with second order Tikhonov regularization.(2)Combine ECG with CT and MRI in myocardial ischemia analysis, and make good use of the electrophysiological information and myocardial motion information simultaneously, which is embedded in a deep-learning based framework.(3)Transfer deep learning theory to feature representation of myocardial ischemia, and propose a reasonable algorithm to perform myocardial classification.