Research On Automatic Classification Of Heart Failure Based On ECG Signal

Heart failure occurs when the human heart develops disease and continues to deteriorate.Nowadays,due to the aging population and other problems are increasing,heart failure has also become an important medical problem all over the world.The medical examination of heart failure includes multiple indicators,and the electrocardiogram(ECG)is one of the easiest obtained and one of the most commonly used reference indicators for doctors.It indicates the potential activity of the heart of the patient and can directly display the changes of the heart.This thesis uses the Physio Bank databases and the MIMIC-Ⅲ databases to extract ECG signals to study heart failure.And this thesis establishes an early diagnosis model of heart failure based on deep learning methods,including the diagnosis and classification of heart failure.The main work of this thesis is summarized as follows:(1)Proposing a model based on Convolutional Block Attention Mechanism-Convolutional Neural Network(CBAM-CNN)to diagnose heart failure.The model makes full use of the Convolutional Neural Network(CNN)can automatically extract features and the Convolutional Block Attention Mechanism(CBAM)module adaptively learns the local features,effectively extracts the complex features of ECG signals and performs diagnostic classification.And using ECG signals to design comparative experiments.The model is evaluated and the effect of signal preprocessing on the performance of the model is discussed.The experimental results show that the proposed CBAM-CNN model has good performance for ECG signal classification.The highest accuracy of heart failure diagnosis reaches 97.62%.At the same time,the CBAM-CNN model is sensitive to noise.The accuracy is improved by1.66% after denoising.(2)Proposing a model based on the combination of Convolutional Neural Network-Gated Recurrent Unit Neural Network(CNN-GRU)to classify heart failure.The CNN is combined with a Gated Recurrent Unit Neural Network(GRU)based on the temporal features of ECG signals.The CNN is used as a featureextractor,and the GRU is used as a classifier to achieve four classifications of heart failure.By designing a comparative experiment of the combined network with the single network,the results show that the combined network has a better performance on the classification of heart failure,with a maximum accuracy of92.75%.At the same time,the influence of adding the batch-weighted loss function on the model is discussed.The experimental results show that the batch-weighted loss can reduce the impact of the imbalance of the data set on the classification performance.The accuracy of the model using the batch-weighted loss is improved by 1.11%.