Recognizing PICO Elements In Medical Text Based On Deep Learning

Evidence Based Medicine is an emerging discipline that developed rapidly in the field of clinical medicine in the 1990 s.It is a medicine that follows scientific evidence.It can fully apply the best research evidence currently available and combine the personal clinical experience of clinicians,formulate the best treatment measures for patients.Evidence Based Medicine solves actual clinical problems through the use of a general specification,the PICO principle.Its main content includes patients/participants,intervention methods,comparison measures,and the outcome indicators of concern.The analysis of Evidence Based Medicine first selects a group of potentially relevant medical texts to form the evidence base on which the answer to a specific question depends.In order to facilitate this selection process,all medical texts can be organized in accordance with the PICO principle,but a large part of the medical texts contain unstructured text data without clear PICO elements.Therefore,how to quickly and efficiently identify PICO elements from medical texts has attracted more and more attention.The research content of this article includes the following three parts:(1)PICO elements recognition based on Bidirectional Gated Recurrent Unit model combined with Conditional Random Field.Traditional machine learning methods have the problem of insufficient feature extraction,and it needs to train multiple models to recognize PICO elements,which not only consumes a large amount of computing resources,but also has low efficiency.In order to solve the above problems,this paper proposes a method of PICO elements identification based on Bidirectional Gated Recurrent Unit model and Conditional Random Field.After training,testing and evaluation,the F1 value on the P element is 88.24%,and the F1 value on the I/C element is 80.49%,and the F1 value on the O element is 86.62%.Experimental results show that this method can not only improve the problem of insufficient feature extraction in traditional machine learning models,but also extract multiple elements at the same time,avoid the waste of resources caused by creating multiple models,and it is better than the traditional machine learning model in the recognition effect.(2)PICO elements recognition based on BERT model combined with Bidirectional Gated Recurrent Unit and Conditional Random Field.Traditional word vectors cannot solve the problem of polysemous words,and their ability to predict contextual semantic information is limited.this paper proposes a PICO elements identification method based on the BERT model combined with Bidirectional Gated Recurrent Unit and Conditional Random Field.After model training,testing and evaluation,the F1 value on the P element is 91.22%,and the F1 on the I/C element is 85.98%,and the F1 value on the O element is 89.87%.Experimental results show that this method provides more accurate and efficient training through the word vector representation of the lower layer and the extraction of contextual semantic information of the upper layer,so as to obtain better recognition results.(3)Design and implementation of PICO elements recognition system.The BERT combined Bidirectional Gated Recurrent Unit and Conditional Random Field method are used to train the model and write the model interface to realize the PICO elements recognition system.The PICO elements identification method proposed in this paper has achieved good recognition results,and can quickly and efficiently identify PICO elements from medical texts,in the hope that PICO elements identification can solve specific clinical problems with high quality.