Prediction Of Albumin Levels In Critically Ill Patients Based On Deep Learning

With the development of medical technology,the nutritional health of patients with severe diseases is highly concerned by medical staff.Serum albumin level is related to the mortality and prognosis of severe patients and is one of the indicators concerned by clinicians.To reduce the risk of death in severe patients,it is necessary to monitor serum albumin level and provide timely nutritional support.However,clinical detection of albumin level often requires artificial blood sampling test for patients,frequent monitoring will consume a lot of material and human costs,and increase the discomfort of patients.In order to achieve the goal of free examination of patients’ albumin level,it is particularly important to use artificial intelligence technology to predict serum albumin level.However,the current research on the prediction of albumin level in severe patients is still in the primary stage.Therefore,how to accurately predict the level of albumin and analyze the correlation between patients’ nutritional support and albumin level indicators are urgent problems to be solved.First,a nutritional support standard dataset is constructed.Aiming at the problems of lack of research data,high clinical data and irregular intervals of actual characteristic recording points,relevant data screening criteria are designed according to the requirements of albumin level prediction task.The time series data set including the basic information,disease,laboratory test indicators and nutritional support mode of severe patients is constructed by using the time series data construction method,which laies a foundation for the subsequent construction of albumin prediction model.Secondly,a sequence prediction model based on two-channel neural network fusion is proposed.The research on the task of predicting albumin level in severe patients is still in the preliminary exploration stage.In this thesis,the medical problem is transformed into a time series regression prediction problem.And the time series data of nutritional support is modeled.The model extracted the local correlation feature and macro time feature of patient variables respectively to predict.Compared with the traditional prediction model,the prediction results have been effectively improved.Then,a model CG＿GCN for predicting albumin level based on similar patient network graph embedding is proposed.Due to the large number of clinical patients and complex nutritional support methods,it is difficult to analyze the correlation between patient populations only by extracting temporal features,which affects the prediction of albumin level.To solve this problem,the model introduced similar patient information to enhance the accuracy of prediction.In this method,the population of severe patients was represented as a graph,in which the hidden local correlation features and temporal features between variables are represented as the graph node features.The potential relationship information between the albumin change trends of patients with the same disease is used as the edge to integrate,so as to better extract the association information between the patient population.Experimental results show that the accuracy of the proposed method is greatly improved in the task of predicting albumin level.Finally,a visualization system for ICU nutrient level analysis is developed.To further support the clinical doctor nutrient level in severe patients with analysis,in this thesis,by constructing the nutritional support of standard data sets,designs and realizes the system data import,statistics,albumin and liver result analysis,details and albumin level in patients with prediction,in order to help the doctor to analyze various nutrient level in severe patients,and formulate relevant nutritional treatment quickly.