Disease Assistant Diagnosis And Drug Recommendation Based On Deep Learning

Based on a large number of electronic medical records,intelligent medicine uses modern computer technology combined with existing medical data to assist doctors in diagnosis and treatment.In this paper,the deep learning model is constructed by combining the electronic medical record data to help the doctor to complete the diagnosis of the disease and subsequent drug recommendation,and thus improve the treatment efficiency of the doctor.The following work was mainly completed:Firstly,the 1D-DLSTM disease diagnosis model is constructed.First of all,by combining with ICD-10 to build a professional thesaurus in the field of medicine,and use Jieba segmentation method joint medical lexicon to segment medical records.Secondly,the Skip-gram model is used to train medical word vector in the BERT model,so that the word vector covers a large number of medical record characteristics.Finally,the 1DDLSTM disease diagnosis model is proposed,with the word vector matrix described in the medical record text as the input,the disease-diagnosed disease as the model output.The convolution layer scans the word vector,and further utilizes the LSTM kernel to extract the word vector characteristics.After the parameter adjustment of the reverse propagation and the normalized output of the results,and finally train to complete the disease diagnosis model.By entering the patient’s symptoms,the patient’s disease can be diagnosed.Experiments show that the final diagnostic accuracy of 1D-DLSTM disease diagnosis model is 90%,which is 3% higher than that of CNN.Secondly,recommend medicines based on medical knowledge map.Use the Request method is used to crawl through medical websites to obtain the description of the disease and its properties.The crawled entity established according to the medical terminology specification in ICD-10,and the relationship between the entity and its attributes is established according to the description on the website.Finally,according to the Chinese grammar principle,the subject predicate object is mapped to the entity and relationship to form the medical knowledge triplet,storing the medical knowledge triple in Neo4 j,and retrieving the corresponding drugs for disease based on the database.Finally,build a BP-ADRP model to screen medicines.First of all,using the sliding window traversal method to extract the medical examination data in the electronic medical record,the pre-drug physical examination as the characteristic data,the post-drug physical examination as the label data.Based on the BP neural network in deep learning,the BP-ADRP model was constructed,and the adverse drug reactions were predicted by BP-ADRP.In this paper,the characteristic data of clinical laboratory is the model input and the label data is the model output.The model can output the adverse drug reaction prediction results by inputting the physical examination data of patients before using drugs.The final prediction accuracy of BP-ADRP model is 88%,12% higher than LSTM algorithm,14% higher than that of CNN algorithm,and 7% higher than that of SVM algorithm,which can accurately predict adverse drug reactions in clinical areas.Retrieving the drugs obtained from the knowledge map,screening the drugs that patients will have adverse reactions,and provide the final recommended drugs,which can effectively assist the doctor in the medication of patients.