Research On Decision-making Support For Rehabilitation Diagnosis And Treatment Of Stroke

Stroke is an acute cerebrovascular disease.Currently,about 70% to 80% of stroke survivors lose their ability to work with varying degrees.The long rehabilitation cycle and complicated rehabilitation mechanism of stroke bring a great burden on the family and society.The traditional diagnosis and treatment process of stroke need to be carried out based on the experience of physicians,which is greatly affected by the subjective decision of physicians.In order to make up for the lack of methods in traditional diagnosis and treatment,the research of computer-aided decision-making on rehabilitation diagnosis and treatment urgently is extensively developed,from which,we can significantly improve the efficiency of physicians,reduce the rate of misdiagnosis,and provide effective decision support for the physicians with insufficient experience and limited medical divices.In this thesis,the electronic medical record information of patients is used to assist doctors in decision-making on the three aspects,including rehabilitation assisted diagnosis,rehabilitation treatment plan recommendation,and rehabilitation prognosis prediction.At the same time,based on the research results,we constructed the intelligent diagnosis and treatment rehabilitation system for stroke patients.The specific research work is as follows:1.We use the technology of natural language processing to convert patient examination results into word vectors as input,and apply the method based on attention mechanism and two-way GRU network to the assisted diagnosis problem.Compared our model with Multilayer Perceptron,Convolutional Neural Networks,and Long Short-Term Memory,test results show their accuracy rates are 60.8%,35.0 %,46.8%and 49.9% respectively.2.We propose a combination of VGG16 and Unet networks for multi-label classification tasks.In this model,VGG16 is used for feature extraction first,by using Unet network for feature reconstruction,the loss function is changed from cross entropy to mean square error.The choice of algorithm and loss function are also discussed.Test resuls show that the accuracy and recall of the proposed model are 68.6% and 52.9%respectively.3.To predict the score of the last assessment of stroke patients,the secondderivative of the loss function is used as an update method of LSTM parameters.Compared this model with MLP and CNN,their accuracy rates are 99.2%,81.1%,and94.7% respectively.In terms of loss curve oscillation,the proposed model has less curve oscillation than other models,which demonstrates the effectiveness of the proposed method.4.The intelligent diagnosis and rehabilitation system for stroke was constructed based on frameworks such as Springboot,Vue and Bootstrap.Development tools include MySQL and Tomcat.In this system,the functions of rehabilitation treatment,plan management,and rehabilitation decision support are implemented,which can help physicians to carry out the management and decision support for their work.