Pre-diagnosis And Risk Prediction Based On Sentence2vec

In this era of informationization,with the rapid development of medical technology and science and technology,new concepts such as smart medical care and Internet medical care have been proposed,and data-driven analysis methods for medical data have emerged.Therefore,it is increasingly important to establish a complete Electronic Healthcare Records system.By reasonably analyzing the acquired medical data,predictive modeling can be performed to achieve early diagnosis of the disease.Or by analyzing the patient's health status by analyzing the patient's EHR data to predict the patient's risk of having a disease in the future.The Electronic Health Record(EHR)system,like other health digital systems,can make healthcare more efficient,secure,and intelligent.The combination of electronic health records and natural language processing can reduce the need for humans to do routine,time-consuming,and repetitive tasks.The vacated staff can be redeployed to support higher-end work and greatly promote health care.development of.But there are also many reasons why we still face many challenges in disease risk prediction on the EHR dataset.For example,the absence of related fields in the EHR data set may lead to incomplete extraction of features.In addition,due to manual errors,mis-recording,missing notes,and even more records may cause us to analyze the data causing noise,which affects our extraction of features.Therefore,the ability to efficiently analyze e-health health records is critical to improving health care.In this paper,we innovatively use the combination of sentence2 vec and CNN,first use sendenc2 vec to vectorize the patient's electronic health record indicators,and then put the corresponding generated vector into the trained CNN classification network,and then The results obtained are analogized to obtain the results of early diagnosis and risk prediction for the disease.Compared with the traditional EHR dataset processing method,the proposed method not only ignores the difference in the length of the patient's electronic health record,that is,the impact of the high-dimensional sparsity of the EHR dataset on the experimental results,and more completely preserves the patient's relevant information.At the same time,it also takes into account the connection between different physical indicators of the patient,so that the characterization information can be extracted more effectively.The experimental results show that for congestive heart failure and diabetes,the method proposed in this paper can predict more accurate results 180 days in advance,and the accuracy reaches more than 99% 90 days in advance.Knowing the risk of a disease in advance can lead to more time for doctors to develop more accurate treatments for patients,which is a very meaningful thing in the history of health care.