The Design And Implementation Of Medical Record Entity Recognition Algorithm Based On Multi-Layers Learning

With the rapid development of medical information, medical institutions generate a large number of original medical record data during the clinical diagnostic process. Since most electronic medical records(EMR) are unstructured and narrative texts and the records can’t store, organize and manage the clinical information of the records very well, it is hard to use the information of EMR sufficiently. In the field of researches, the precision and robustness of medical record entity recognition and the sharing and standardization of organization model of them have been the important process of the information extraction of the medical records. Artificial clinical system requires the medical record entity model not only describe the entity, but also describe semantic relations between entities. Recognizing medical record entities and relations can structure the medical records to support the demand of modern clinical system.Named entity recognition is a basic research task of text information extraction and it is always used to recognize the descriptive information in text as named entities, and it marks the named entities with defined symbols. In this thesis, named entity recognition technology is applied into EMR, the record named entity is defined as record entity, and medical record entity recognition algorithm based on multi-layers learning is designed and implemented.Firstly, the features of electronic medical records are analyzed in this thesis, and the record entities which are needed to be recognized are defined. The relations among the medical record entities are analyzed in the thesis, and a clinical medical record ontology for entity, description and relation is designed. Secondly, the medical record entity recognition algorithm based on multi-layers learning is designed. it is formed by three layers:medical record entity recognition based on CRF completes the initial recognition by using the marked texts to train the CRF recognition model; medical record entity recognition based on decision tree regards the recognition as an classification task and uses the decision tree algorithm to train the classifier to modify the initial recognition result; medical record entity recognition based on prior rules integrates the result of CRF and decision tree by defining the construction rules of the complex medical record entities. Finally, after the tests of medical record entity recognition based on multi-layers, it is verified that the algorithm has good performance on precision and ROC of recognition, and it has great property of robustness.