The Research Of Heart Failure Warning Technology Based On The Fusion Of ECG And Clinical Factors

Heart failure(HF)is one of the most common reasons for death worldwide,which brings huge spiritual and economic burden to patients,family,and society.Establishing a reliable automatic warning model is of vital importance to the effective management of HF disease and it is an increasing focus for physicians and policymakers in recent years.In this paper,the electrocardiogram(ECG),clinical examination variables and demographic data of Medical Information Mart for Intensive Care III(MIMIC-III v1.4)are combined as research objects,and signal processing and machine learning technology are introduced into this field to realize the automated warning and assessment of HF patients,moreover,the risk factors associated with death were analyzed.The main research contents are as follows:(1)A method based on long short-term memory network(LSTM)is proposed to realize the automatic dignosis of HF.In this study,firstly,the ECG signal is subjected to preprocessing such as denoising,heartbeat segmentation,and Z-Score normalization.Then,the synchrosqueezed wavelet transform is used to obtain the intrinsic mode function of the ECG.At last,the energyentropy and the permutation entropy of these modes are feed to LSTM to realize the automatic detection of HF.(2)On the basis of the above diagnostic results,a random forest early warning model worked on particle swarm optimization was established to further warn the death of patients with heart failure within 5 years,which fully integrates the features of ECG signals,clinical examination variables and demographic data.Firstly,the entropy features of HF patients’ ECG signals are extracted again.Then,their demographic data and clinical characters are also extracted.Finally,the fused feature is input into the improved random forest based on particle swarm optimization algorithm.The model can be used to predict the death of patients with HF within 5 years,and further explores the risk factors affecting the 5-year prognosis.The experimental results show that the average accuracy of the HF automatic detection is 97.50%,moreover,in the prognostic assessment of HF patients,the area under the ROC curve is also more than 0.80.Therefore,the method proposed in this study can provide effective guidance for doctors’ clinical treatment,follow-up management,and patients’ self-management to a certain extent.