Research On Prognosis Evaluation Model Of Heart Failure Based On Machine Learning

Heart failure(HF)is a group of complex clinical syndromes that lead to impaired ventricular filling or impaired ejection ability due to abnormal structure or function of the heart.Difficult diagnosis and treatment,poor prognosis,and high mortality are the main features of heart failure diseases.Clinically,it is of great significance to evaluate the prognosis of patients with heart failure and to intervene in the development process,to treat some patients with poor prognosis symptomatically and choose the best treatment plan.In this paper,Medical Information Mart for Intensive Care(MIMIC-III v1.4)is taken as the research object,the clinical signals and ECG signals of patients with heart failure are extracted,and the prognosis evaluation model of heart failure is established based on machine learning technology.The research content of this thesis mainly includes the following points:(1)Establish a prognostic model of heart failure prognosis based on Recurrent Attention Network(RAN).Firstly,use Term Frequency-Inverse Document Frequency(TF-IDF),Word to Vecter(Word2vec)and Latent Dirichlet Allocation(LDA)models to construct the medical record text data feature vectors.Secondly,combine Multi-Kernel Learning(KML)to achieve the fusion of text data feature vectors and structured clinical information features.Finally,using the attention mechanism of the recurrent neural network to establish a predictive model of whether the heart failure patients would die within five years.The experimental results show that the model achieves an accuracy rate of 90.7%,and at the same time,it can also obtain the feature importance ranking that affects the prognosis of death in patients with heart failure within five years.The model can dig out the rich information hidden in the electronic medical records,which provides great help for doctors’ clinical decisions and patient self-management,and reduce the mortality of patients with heart failure.(2)A prognostic evaluation model of Bi-directional Gate Recurrent Units(Bi GRU)with self-attention mechanism is established.First,the clinical signal features and the extracted ECG signal features are input into the Bi GRU and the self-attention model,which obtain the feature vector and the weight value of the feature vector of the heart failure patient respectively.Then,the feature vector obtained from Bi GRU layer and the weight vector output from the self-attention mechanism are dot-multiplied.Finally,the obtained results are passed through the softmax classifier to obtain the prognosis of patients with heart failure.Experiments show that the final accuracy of the model reaches 94.7%,and at the same time,it obtains the characteristic weights that affect the prognosis of heart failure.The model can assist doctors to carry out clinical interventions on patients with poor prognosis,delay the progress of patients’ disease,and help patients choose the best treatment plan.It is of great significance to improve the quality of life of patients and increase the survival rate.