Research On The Method Of Assisted Diagnosis Of Heart Disease Based On Deep Learning

Cardiovascular diseases have gradually become one of the major diseases that endanger human health due to their high incidence,high mortality,and relapse.Electrocardiogram(ECG)is a significant diagnostic method for detecting arrhythmia in clinical practice.However,it is very labor intensive to externally evaluate ECG signals,due to their small amplitude,which makes the doctor not only spend a lot of energy in the analysis of the ECG but also cannot making sure that there are no errors.Therefore,the use of deep learning technology for detection and classification research is of great significance to help doctors quickly and accurately diagnose diseases.It was based on performed some pre-processing on the ECG signals and proposed models of Convolutional Neural Network(CNN)and a combination model of Convolutional Neural Network and Long Short-Term Memory(LSTM)in this paper.The main research contents include:(1)Aiming at the one-dimensional ECG data,a series of noise filtering work was performed on the one-dimensional ECG signal,and the R wave in the ECG signal waveform was located by wavelet transform and adaptive threshold method.(2)For two-dimensional ECG data,one-dimensional ECG signals were converted and two-dimensional ECG grayscale images were used as input data in the paper.The original data information can be retained to the greatest extent while eliminating preprocessing tasks such as noise filtering.The data imbalance problem was better optimized by using a specific cropping method to enhance the data of the ECG image.(3)A one-dimensional CNN model was proposed,and five heartbeat types including normal beats were classified according to the Association for the Advancement of Medical Instrumentation(AAMI)standard.Based on the MIT-BIH arrhythmia database data,preprocessed one-dimensional ECG data was used for experimental verification,and 97.87% accuracy,97.25% sensitivity,and 98.50% specificity of classification performance were achieved.(4)A CNN-LSTM model was proposed.The model took the converted twodimensional ECG image as input data and also classified eight heartbeat types including normal beats according to the AAMI standard.Experimental verification using the MIT-BIH arrhythmia database has achieved 99.01% accuracy,97.67% sensitivity and 99.57% specificity of classification performance.In summary,the CNN and CNN-LSTM models and the preprocessing of ECG data proposed in this paper can achieve better classification performance of arrhythmia diseases and provide effective means for early prevention and auxiliary diagnosis of cardiovascular diseases.