Research On Medical Pre-diagnosis Method Based On Compound Features And Gene Expression Programming

In the medical domain,online consultation platforms and Internet hospitals have become important media for communication between patients and doctors.However,because most patients lack a comprehensive understanding of their own conditions and medical professional knowledge,it is not uncommon for patients to find the wrong doctor on the consultation platform,register the wrong department on the online registration system,and find the wrong doctor.Therefore,the intelligent medical prediagnosis system came into being.The intelligent medical pre-diagnosis system is a human-machine dialogue system.If the patient provides the inquiry and their existing symptoms,the system can easily triage the patient to the correct department and doctor,to carry out the follow-up process.Thus,the intelligent medical pre-diagnosis system is essentially a text classifier.That is,according to the patient's main complaints and symptoms,it determines which department the patient should be classified to.Text classification tasks are usually solved by machine learning methods.Among them,neural networks have outstanding performance in various types of text classification tasks due to their excellent coding capabilities and feature extraction capabilities.There are two commonly used neural network models,convolutional neural networks and recurrent neural networks respectively,both of which have certain limitations when dealing with text classification tasks.Convolutional neural network extracts local features in sentences through two operations,convolution and pooling.Thus,the features of keywords in sentences can be well extracted.However,due to the size limitation of the convolution kernel,the local features are isolated from each other,so that the convolutional neural network cannot extract the contextual relationship of the local features.While the recurrent neural network is to use the special structure of the feedback connection between each state to memorize the context information and dependencies in the sequence to extract the global features of the sequence.However,because of this unique structure,the recurrent neural network cannot extract the local information in the sentence very well.To solve this problem,this paper optimizes the structure of convolutional neural network,and combines it with bidirectional long short-term memory to propose a hybrid network combining local composite features and global features which can achieve better performance of text classification tasks.The experimental results show that,with the help of hybrid network,The accuracy of the pre-diagnosis system has been significantly improved.On the other hand,to conduct triage more accurately,the pre-diagnosis system also needs to make comprehensive judgments based on the existing symptoms of patients.Some of the existing pre-diagnosis systems use a one-step approach,directly shunting patients to the corresponding departments from the patient's main complaint,while others establish a function to describe the relationship between the patient's selected symptoms and corresponding diseases.The pre-diagnosis system can find out which disease the patient has based on this function.Undoubtedly,these two methods are unreliable and will have a great impact on the accuracy of the system.In essence,determining the corresponding disease based on patient symptoms is an optimization problem.Gene expression programming,as a globally optimal search algorithm,solves the optimization problem by iterating the population to find the optimal individual.Therefore,gene expression programming is for improving the overall accuracy of the system.On the other hand,the complexity of the medical domain makes the function model need to be updated frequently,and the standard GEP algorithm runs under a single-threaded,which cannot meet this requirement.Based on the standard GEP algorithm,we propose a multi-threading evaluation and gene-reuse strategy to improve the running performance of the GEP algorithm to adapt to this application.Finally,this paper proposes a new structure of intelligent medical pre-diagnosis system.We use a hybrid network combining local composite features and global features as the "upper half",while proposed gene expression programming is the "bottom half".Experiments showed that the intelligent medical pre-diagnosis system with this architecture has achieved a certain improvement in the classification accuracy compared with the ordinary pre-diagnosis system.