Research On The Radiologic Interpretation Knowledge Graph Based On Chest X-Ray Three-Layer Structure Model For Intelligent Diagnosis Of Radiology

1 ObjectiveIn order to improve the diagnostic efficiency and report standardization of medical images,this study proposes and constructs a knowledge graph of radiologic interpretation based on a three-layer structure model by simulating doctors’ diagnostic thinking,and applies it to intelligent diagnosis of medical images to improve the writing efficiency,standardization and interpretability of diagnostic reports.This method can reduce the burden of doctors and put forward new ideas for the bottleneck problem of intelligent diagnosis of medical images.2 Methods(1)Proposition and construction of knowledge graph of three-layer structure model Radiologic Interpretation Knowledge Graph.In order to improve the accuracy and interpretability of intelligent diagnosis,this paper draws on the experience of knowledge graph to participate in the generation of diagnostic reports.In this paper,the lesions location was added and the lesions characteristics were refined.Thus a new knowledge expression method was proposed,namely the Radiologic Interpretation Knowledge Graph of the three-layer structural model of "lesion area-lesion feature-detail feature".This method reorganized and refined the professional medical knowledge,making its expression clearer and more comprehensive.(2)X-ray report generationFor the standardized generation of findings,we analyze medical images through algorithms and get structured labels by combining the contents of the knowledge graph,use the designed rules to call the interface,and interact the structured data with the graph nodes in the secondary graph database to get the standardized findings.In impression generation,we take the standardized findings and original impressions as training data,divide the data set according to the ratio of 7: 2: 1,then predict the generated results by training the impression generation model.A standardized examination report can be obtained by combining the standardized findings and the impressions generated.3 ResultsIn the results of standardized findings: 1200 pieces of data randomly selected were evaluated by professional doctors for their accuracy and fluency.In the accuracy evaluation,“qualified” data accounts for 100%;In fluency evaluation,"qualified" data,accounts for 99%,and "unqualified" data accounts for 1%.In the results of impression generation,300 items were randomly selected and their accuracy was evaluated by doctors.Among them,“qualified” data accounts for61.34%,and “unqualified” data accounts for 38.66%.4 ConclusionIn this paper,a knowledge graph of radiologic interpretation based on three-layer structure model is proposed,which is applied to the intelligent diagnosis task of medical images,and the results are evaluated by experts in practical application.The results show that the intelligent diagnosis research driven by knowledge graph improves the professionalism,accuracy and interpretability of diagnosis results,and can provide more professional decision-making assistance for clinicians to a certain extent,and effectively explore its application.