Research And Implementation Of Prescription Drug Recommendation Algorithm For Patients With Dementia Based On Deep Learning

Dementia is a neurological disease with stupidity as the main clinical manifestation.It mainly occurs in the elder people.The causes of dementia are complex and it is difficult to be cured,patients usually need long-term medical care.However,at this period,medical resources in China are relatively scarce.Most dementia patients can only be cared by their families or nursing staff at home.Therefore,the research and design of a telemedicine consultation system for dementia patients can improve the utilization of limited medical resources.At the same time,providing drug recommendation for doctors when they are prescribing drugs can greatly improve doctors’ work efficiency.The main contents of this thesis are as follows:(1)Uses Python crawler to collect information of dementia-related drugs,and then,based on the diagnosis and treatment knowledge of dementia,the collected drug information and the definitions of various ontologies,the drug-centered knowledge graph is designed,the relationships and attributes in the drug knowledge graph are clarified.In order to identify the entities in drug text descriptions such as adaptive diseases and adaptive symptoms and aid to construct the drug knowledge graph,the BERT-Bi LSTM&IDCNN-CRF algorithm are designed to combine the text sequence features extracted in different ways.Finally,the Neo4 J database is used for storing the drug knowledge graphs.(2)The probability of medication of being recommended is calculated by neural network,then the drugs are ranked according to the calculated probability.Finally the first five drugs are taken as the recommended drugs for patients.Meanwhile,considering dementia is a progressive disease,this thesis proposes an LSTM-Deep FM network model to calculate medication’s probability of being recommended.In this model,the LSTM network is used for feature extraction of patient medication sequences,and then the drug recommendation probability is calculated by combining the patient’s current consultation details and information on medication use.The F1 score of the final recommendation algorithm is 0.418,which is 0.043 higher than the Deep FM-based recommendation method,providing an effective reference for doctors’ medication use.(3)In response to the needs of the cooperative medical institutions and the characteristics of the dementia patients’ treatment,the subsystem of the telemedicine consultation system is designed and implemented.Using this system,doctors and nurses can realize remote diagnosis,care and consultation for patients.The system also integrates drug knowledge graph and intelligent medication recommendation algorithms to facilitate doctors to view drug information and diagnose medication selection.