Research On Automatic Identification And Classification Of Cardiac Arrhythmia Diseases Based On Deep Learning

Cardiovascular disease is one of the most deadly diseases worldwide,accounting for approximately 31% of all deaths worldwide each year and posing a serious risk to human health.The onset of cardiovascular disease is usually accompanied by arrhythmic disease.In clinical testing,doctors usually complete the diagnosis of arrhythmia disease through electrocardiogram,which is a traditional diagnosis method with low efficiency and prone to misdiagnosis and omission.Therefore,how to make timely and accurate diagnosis of arrhythmias has become a challenge in the prevention of cardiovascular diseases.With the in-depth development of deep learning,related technologies have been widely used in the medical industry,and some studies have attempted to accomplish automatic classification of arrhythmia diseases with high accuracy through deep learning related algorithms,but the diagnosis results are not satisfactory.To address the above problems,the following research work has been systematically carried out in this paper.(1)To address the problem that ECG signals have non-smoothness,this paper proposes a wavelet transform-based and short-time Fourier transform ECG signal processing method.Firstly,by comparing different wavelet basis functions and threshold processing,the high-frequency signal noise is removed,and then the low-frequency noise is filtered out by combining sliding window integration to make the signal further smooth and the quality of ECG signal is improved.Secondly,because the high-dimensional data contains more feature information,the short-time Fourier transform technique is used to convert the signal from the time-domain state to the time-frequency domain state for analysis,which completes the conversion of one-dimensional signal to two-dimensional image signal and helps to remove redundant information and present more powerful waveform features.(2)To address the problem that the feature information extracted by the convolutional layer in Res Net34 is not sufficient,this paper improves the Res Net34 network structure by adjusting the connection of each residual block to improve the information exchange between channels and extract ECG features to a higher degree,followed by using depth-separable convolution instead of standard convolution,which effectively reduces the number of mathematical operations and parameters and improves the operational efficiency.Combined with the experimental comparison,the effectiveness of the improved Res Net34 network model is demonstrated.(3)To further improve the classification performance.The Res Net-PLSTM-CBAM arrhythmia classification model was built.The model can complete the extraction of ECG local features by Res Net module,complete the extraction of ECG global features by using the improved LSTM gate structure P-LSTM,and then complete the extraction of local combined with global key features by using CBAM attention module,and put into the Softmax classifier to complete the classification.At the same time,the Focal loss function is used to effectively optimize the experimental results for the problem of unbalanced data of five types of samples.Finally,by comparing the models and previous methods,we achieve better results and prove the effectiveness of the proposed model.The research results in this paper fully demonstrate that the convolutional neural network and long and short term memory network models in deep learning techniques can well accomplish the task of automatic classification of arrhythmia diseases.The experimental results show that the model proposed in this paper is beneficial to improve the performance of the existing automatic classification of arrhythmia diseases,and has certain theoretical value and application prospects.