Integration of AI and SNOMED CT in Chest X-Ray Diagnosis Software System

In the medical field, the application of artificial intelligence (AI), especially deep learning models of convolutional neural networks (CNN) to disease diagnosis is becoming more and more popular. However, the research and application of AI in disease diagnosis in Vietnam is still limited. Disease prediction is still fragmentary, there is no link between clinical and laboratory results, as well as the problem of disease identification is not synchronized and consistent among studies because labels are described in text that can be confusing, especially in the medical field. Therefore, this research paper applies Systematized Nomenclature of Medicine-Clinical Terms (SNOMED CT) as a set of electronic clinical terminology in labeling input data for machine learning in the field of medical pathology diagnosis as well as making suggestion of the templates for each specific disease. Specifically, the article evaluates the results of a method of diagnosing abnormalities in chest X-ray (XR) images based on the Densenet-121 network architecture and normalizing the forms based on SNOMED CT codes. DICOM medical images are obtained directly and automatically from the X-ray modality through the Picture Archiving and Communication System (PACS) at Telemedicine and Mobile Healthcare lab, Ho Chi Minh City University of Technology. Chest abnormalities are detected, classified and returned to the SNOMECD CT codes corresponding to the major problems in the sample template on workstation using the AI TOOL ASSISTANT software system. A survey of 222 actual abnormally diagnosed chest X-rays images at the international clinic participating in the trial showed that the accurate of the prediction results was 82% when compared with the doctor's conclusions according to these samples.