A crowdsourcing framework for retinal image semantic annotation and report documentation with deep learning enhancement

To propose and implement a crowdsourcing framework for retinal image annotations to improve the annotation efficiency. In this study, open-source Bluelight was taken as backbone of front end online manual semantic report documents, based on that intelligent classification with deep learning (DL) supplemented. For DL modules, we trained Mask-RCNN model explicitly label area optic disc macula. Furthermore, Inception V3 classify diabetic retinopathy (DR) normal retina. Then, used Flask backend serving models. Finally, implementation interoperable reports documentation retrieval were conducted Lucene. The specially designed professional doctors computer researchers who have ability annotate. It efficiently quickly completed macular area, at same time classified DR. Under Browser/Server architecture, tool achieved good cross-platform performance. particular, could provide documents facilitate optimization subsequent Such retina effect worth further improvement clinical validation.