Research On Dialysate Decision Support Method Based On Deep Learning

Patients with end-stage renal disease depend on hemodialysis to maintain their basic state of life.Hemodialysis makes use of dialysate to exchange dialysate and solute of dialysis patients’ blood in dialyzer with the help of principles of dispersion,ultrafiltration and convection,so as to correct the electrolyte and acid-base balance in the body,remove excess water,and maintain the stability of internal environment of body.Studies have found that the scheme of potassium and calcium ions in dialysate is closely related to the prognosis of patients,so it is of great significance to develop appropriate dialysate schemes for different patients.With the development of computer technology and electronic medical records,researchers at home and abroad began to use machine learning techniques to build decision models for various medical problems in the field of dialysis.However,machine learning techniques,especially deep learning techniques in machine learning techniques,have not been widely studied on the decision support of individualized dialysate scheme.In the actual process of dialysis treatment,doctors judge the physiological development trend of dialysis patients by referring to the multi-dimensional temporal physiological information of dialysis patients,and then develop individualized dialysate schemes for patients.In order to improve the decision-making efficiency of doctors and reduce the decision-making pressure,this paper applies deep learning techniques to model the multi-dimensional temporal physiological data of dialysis patients,predict the dialysate scheme for patients,and provide strong technical support for the formulation of good treatment plans for patients.Therefore,the main research content and innovative work of this paper are as follows:(1)Aiming at the problem of dialysate decision support for dialysis patients,a dialysate decision support model BLSTM-Attention is proposed by combining the Attention Mechanism and Bidirectional Long Short-Term Memory(BLSTM).BLSTM is used to characterize the multi-dimensional physiological information of dialysis patients in both positive and negative time directions,which is consistent with the logic of doctors to browse the physiological information of patients.At the same time,in order to increase the interpretability of the model and improve the ability of model to capture important physiological information records,the attention mechanism in time direction is introduced to learn the contextual information in time direction of multidimensional physiological information.The experiment results show that the BLSTMAttention model significantly improves the performance of dialysate scheme multiclassification prediction compared with other temporal models and non-temporal models.(2)In order to make a more comprehensive use of the multi-dimensional temporal physiological information of dialysis patients,characterize them in time dimension and feature dimension to realize multi-perspective feature fusion,a dialysate decision support model SACNN-BLSTM is proposed.Based on the BLSTM-Attention model,this model combines 1-Dimentional Convolutional Neural Network based on Spatial Attention(SACNN)to extract the representation of physiological information in feature dimension.That is,using 1-Dimensional Convolutional Neural Network to perform multi-window convolution in the feature direction of the dialysis patient’s physiological information to learn multiple representations,and using the spatial attention mechanism to highlight the important features of the feature map to form the new feature map.Then,all the new feature maps formed by the same convolution window are element-wise added to form the channel fusion representation.The different channel fusion representation formed by different convolution window is spliced as the representation in feature dimension of physiological information,and then fuse the contextual information in time direction extracted by the BLSTM-Attention and the personal basic information of patient as the patient’s final representation.The experiment results show that the SACNN-BLSTM can further improve the performance of dialysate multiclassification prediction.