Myocardial Ischemia Detection Based On Deterministic Learning And Transformer

The early detection of myocardial ischemia is an important and challenging problem in the field of cardiovascular disease.The electrocardiogram(ECG)contains rich physiological and pathological information reflecting heart function,and is the most widely used method for detecting myocardial ischemia clinically.However,many patients with myocardial ischemia do not exhibit specific ischemic changes in their ECGs,but rather show normal or nearly normal patterns,which poses a significant challenge to the application of existing ECG-based detection methods in clinical settings.In addition,the patient’s cardiovascular risk factors(such as gender and age)are of great value in the diagnosis of myocardial ischemia.Cardiodynamicsgram(CDG)analysis is a novel method of analyzing ECG based on dynamic modeling of the ST-T segment using deterministic learning.This method can effectively detect myocardial ischemia in ECGs that appear normal or nearly normal.Further exploration of richer dynamic features based on CDG information plays an important role in improving the performance of myocardial ischemia detection.This study focuses on using deterministic learning to dynamically model the ECG signal and obtain ECG dynamic information.We further employ multi-scale analysis methods and Transformer models to extract valuable electrocardiographic features for detecting myocardial ischemia.Additionally,we integrate cardiovascular risk factors to construct a myocardial ischemia detection model.The main contributions of this article are summarized as follows:(1)An effective myocardial ischemia detection model is established based on the deterministic learning and the multi-scale analysis method.First,deterministic learning is used to dynamically model the ECG signal and obtain its dynamic information.Then,the ECG signal is decomposed into multiple scales using discrete wavelet analysis,and dynamic features are extracted from the wavelet coefficients in different frequency bands.Next,a hybrid feature selection algorithm based on filter and wrapper methods is proposed to select important dynamic features and obtain a representative low-dimensional feature subset.Different myocardial ischemia detection models are trained using the selected features.Finally,the performance and generalization ability of the models are validated on completely different datasets.Results show that the myocardial ischemia detection model based on multi-scale electrocardiographic dynamic features has good detection accuracy and clinical generalization ability.(2)Based on Transformer,this study proposes a multi-modality myocardial ischemia detection model that automatically extracts dynamic ECG features and incorporates cardiovascular risk factors.First,an ECG embedding module is designed to convert ECG dynamic information into ECG embeddings.The Transformer model is then used to automatically extract effective features from the ECG embeddings.Secondly,a feature fusion module is designed to combine the ECG features with other modal features,such as cardiovascular risk factors,for multi-modal myocardial ischemia detection.Finally,the model’s generalization performance is validated on a clinical dataset.The results show that the Transformer model can automatically obtain effective ECG dynamic features and achieve similar generalization ability to manually engineered features.(3)This section establishes the low code machine learning training platform.The platform aims to promote the application of machine learning models in the medical field and reduce the complexity and functional redundancy of the machine learning model development process.The platform provides a visual interface that allows users to easily create,train,and test various machine learning models with simple configurations,improving developer efficiency.The platform also supports medical professionals without machine learning development experience to apply machine learning technology more conveniently in actual clinical work.More importantly,the platform supports users to explore research on deterministic learning in the field of heart disease.In conclusion,this paper aims to further explore the features for detecting myocardial ischemia in CDG and to combine them with clinical information to construct a myocardial ischemia detection model with higher generalization performance.In addition,a visual machine learning training platform is established to promote the application of AI-based ECG analysis in the field of cardiology.