Research On Key Technologies Of Disease Prediction Based On Attention Mechanism

With the development of information technology in the medical industry,more and more medical and health data have been generated.The mining and application of health big data has also become an inevitable trend.Both doctors' diagnosis and treatment plan need the support of big data technology.The improvement of people's living standards and health consciousness has made people put forward higher requirements for the treatment and prevention of diseases.And the country have put forward that diseases must be given priority to prevention,emphasizing the treatment before onset.The electronic health record contains the patient's disease,treatment information,and demographic information.Through the mining and analysis of these data,disease prediction tasks can be achieved,thereby helping doctors to prevent diseases and improve the quality of medical services.There are some issues that need to be addressed when mining electronic health records.The first is the inconsistency of data in the electronic health record.The diagnostic information in the electronic health record is recorded by medical codes In the process of medical code assignment,many reasons such as social discrimination and non-standard coding can lead to incorrect medical codes.For example,many epilepsy data are recorded as high fever convulsions.This coding error will cause inconsistency between diagnostic information and drug information data,affecting the final prediction task.In addition,the patient's historical medical information itself has a large amount of data and is high-dimensional time-series data.These characteristics make traditional machine learning algorithms poorly model the data.Although researchers have applied deep learning models such as recurrent neural networks to disease prediction tasks,there are still some problems that lead to poor prediction results.And for the medical field,the interpretability of the model is very important,and how to provide interpretability to the constructed prediction model is also an important issue.This paper designs a medical code correction model to correct wrong medical codes to solve data inconsistency.At the same time,a model based on bi-directional recurrent neural network and crossover-attention mechanism is designed to complete the disease prediction task.The specific work is as follows:1.This paper proposes a medical code correction model based on knowledge embedding and self-attention mechanism.Because the patient's medicine corresponds to the patient itself,there will not be problems such as covering the disease,so the model uses the drug information at the time of medical treatment to analyze the patient's disease information.The model uses drug-related medical knowledge to help improve the embedded representation of the drug,and then uses a self-attention mechanism to generate a hidden representation of the drug combination.Attention mechanisms help discover drugs that are important for predicting outcomes and make the model more effective.Finally,experiments on a real dataset show that the model proposed in this paper has the best effect,and at the same time,the role of attention mechanism in the model is analyzed in detail.2.This paper designs a disease prediction model based on the crossover-attention mechanism.The model uses the patient's historical medical records to predict the next medical visit.Unlike previous methods,the model treats the patient's diagnostic information and drug information separately,and then combines them through a crossover-attention mechanism.The crossover-attention mechanism helps the model to find diseases and medications that are important to the prediction result,which not only improves the accuracy of the model prediction,but also provides interpretability for the prediction result.The experiments on two real datasets show that the model has the best prediction effect,and the cross-attention mechanism provides a good interpretability for the model.