Research On Intelligent Detection Methods For Arrhythmia And Heart Failure

In recent years,cardiovascular-related diseases have become the "number one killer"threatening human life,and attracted people's attention.Arrhythmia and heart failure are both common heart diseases.Timely,accurate diagnosis of these two diseases can provide early treatment or intervention for the sick or high-risk polulation and improve people's quality of life,thus has important social value and research significance.As a non-invasive detection method,ECG has always been an important tool for detecting and diagnosing heart diseases.Especially with the continuous development of intelligent hardware and Internet of Things,the wearable ECG monitor has become a new choice for people to carry out daily heart monitoring.Therefore,how to detect and diagnose the heart diseases based on the signals collected by the wearable ECG monitor(usually single lead ECG signal or continuous heart rate signal)has become a research hotspot in the intelligent diagnosis of cardiovascular disease.Starting from the demand background of intelligent diagnosis of cardiovascular diseases,this paper focused on the pre-processing of ECG signals,the recognition of arrhythmia based on single lead ECG signals,the intelligent diagnosis of heart failure based on continuous RR interval signals,and the intelligent conversion from three-lead ECG signals to standard 12-lead ECG signals.The main contents and innovative achievements of the research include:Firstly,in the pre-processing part of ECG signal,an adaptive de-noising method of ECG signal was proposed based on improved bionic wavelet transform.This method can effectively eliminate the interference in ECG signals,especially baseline drift noise,while maintaining the ECG waveform characteristics.In addition,this paper also used the improved bionic wavelet algorithm to detect the location of R-wave in ECG signal,and obtained ideal results.The above methods can lay the foundation for the subsequent intelligent recognition and diagnosis.Secondly,the paper detected the arrhythmia by improved semi-supervised Affinity Propagation(AP)clustering method based on the actual needs of arrhythmia diagnosis.This method first extracted the characteristics vectors from of the heart beat signal based on the independent component analysis.Then,we improved the AP clustering method by the combination of the constrained K-clustering and constrained seed K-means method,and make it as a semi-supervised AP clustering method.Based on the assumption of having a small number of labeled samples,the semi-supervised neighborhood propagation algorithm was used to cluster the characteristics of heartbeat signal.Finally,the hearts in the same category were clustered.The beats were displayed together and the proposed algorithm was evaluated.This method can avoid decision errors caused by non-optimal initialization,and can work well in practical application.Experiments showed that the detection accuracy of this method reaches 98.4%,which is of great value to the recognition of arrhythmias.Thirdly,this paper detected the heart failure based on RR interval signal.We proposed two method for congestive heart failure(CHF)detection.In the first method,we modified Inception module introduced by GoogleNet via long short term memory network,and used it to detect CHF.In the second method,we extracted the high-dimensional features using deep-learning method,and combined them with the expert features as the final input features.Then,the XGBoost toolbox was used for integrated classification,and the experimental results were compared and analyzed with the existing methods.From the experimental results,the two method both achieved an accuracy of approximately 100%in the detection of severe heart failure,and more than 85%in the detection of non-severe heart failure.Deep learning allows automatic feature extraction in high-dimensional primitive space.It can discover the distributed feature representation of data by combining the underlying features,so as to express the data more concretely.Expert experience contains the wisdom of human beings.Combining the two types of features can identify heart failure more effectively and achieve better recognition effect.Finally,this paper also explored the conversion method from three-lead ECG signal to standard 12-lead ECG signal.In this paper,three-lead ECG signals were reconstructed based on convolution neural network to generate standard 12-lead ECG signals.Compared with the multivariate linear regression model,this method reduced the mean square root deviation by about 71%(patient-specific model)and 12%(general model).