Structure Method And Application Of Medical Imaging Diagnosis Report

The medical imaging diagnosis report records the imaging signs and diagnosis conclusions in the medical imaging service in detail and is an important data resource for clinical disease prediction,decision support,and drug pattern mining.Because the medical imaging diagnosis report is written in natural language,it is difficult to be processed and analyzed by computer directly,which leads to the low utilization of resources.At the same time,as an important part of medical materials,the correct and standardized writing of medical imaging diagnosis reports is a basic skill that imaging students must learn and master.However,the current teaching focuses more on the identification of different imaging signs,and lacks the training environment of imaging diagnosis report writing.This thesis takes the medical imaging diagnosis report as the research object.The main research contents and contributions are as follows:In order to solve the problem of low precision of the existing structured methods of medical image diagnosis report based on dependency analysis,a structured method of medical image diagnosis report based on entity recognition and rule extraction is proposed in this thesis.First,a medical image entity recognition model based on domain dictionary features is constructed,and the entity recognition precision rate on the PET-CT dataset reaches 0.89.On this basis,this thesis designs and implements a medical imaging diagnostic report rule extraction algorithm,which converts the entity labeling sequence into structured data in the form of <inspection item,inspection result>.Compared with the structured method of dependency analysis,the structured method of medical imaging diagnosis report proposed in this thesis improves the recognition precision of inspection items and inspection results by 5.62% and 2.31%.Because of the high degree of language overlap in medical imaging diagnostic reports,the existing automatic scoring methods are low in precision when applied to medical imaging diagnostic reports.In this thesis,an automatic scoring method based on structured medical imaging diagnostic reports for reports written by students is designed and implemented,covering both correctness and standardization.This method firstly constructs the disease prediction model and the semantic similarity model to calculate the correctness of the reports written by students,and then calculates the standardization of medical image diagnosis reports from three aspects: the use of terminology,the order of description,and the completeness of description.On the data set of PET-CT diagnostic reports,the precision and recall rate of the automatic scoring method of medical image diagnostic reports proposed in this thesis both reach 0.96.Compared with co-occurrence and Siamese-BERT,the precision was improved by26% and 27%.