The development and implementation of deep learning assisted interoperable retinal image structured report module in PACS

With the increasing incidence of retinal fundus disease, more attention has been paid to the screening and prevention of fundus diseases. However, the low efficiency and interoperability issues hinder the application of retinal disease reporting. In this study, we developed an intelligent classification module for retina based on inception V3 pre-trained model. The top rank of eight classification scores was output as recommendation for image reader reviewing, and the other structured report (SR) module was designed to generate FHIR-compliant retinal diagnostic SRs by customizing existing FHIR DiagnosticReport and Observation resources. The results demonstrated the artificial intelligence (AI) classification pop-up window and the validation of produced SR via a public HAPI FHIR server. The results suggested that the SR was easily shared and could be accurately and integrated with other hospitals or clinical research institutions. In summary, our developed AI assisted SRs module is clinical workflow friendly, and the produced SRs could be easily shared and integrated, which can provide a meaningful use way for further analytic research.