PROJECT TITLE :

CANet Cross-Disease Attention Network for Joint Diabetic Retinopathy and Diabetic Macular Edema Grading

ABSTRACT:

One in three people who are working-age and have diabetes will go blind due to diabetic retinopathy or diabetic macular edoema. When it comes to designing personalised treatment plans for patients, the use of automatic grading of DR and DME can be crucial. But previous works either grade DR or DME and overlook the association between DR and its complication, i.e. DME. Furthermore, the geographical information, such as the macula and soft hard exhaust remarks, are frequently employed as a prior for grading. Obtaining these annotations is time-consuming and costly; as a result, approaches that just require image-level monitoring should be developed. Here, we describe an unique cross-disease attention network (CANet) for grading DR and DME in tandem by examining the internal link between the two conditions with only image-level oversight. The disease-specific attention module and the disease-dependent attention module are two of our most important contributions, as they allow us to acquire unique features for specific diseases while also capturing the intrinsic relationship between the two disorders. In order to develop disease-specific and disease-dependent features and to maximise overall performance jointly for grading DR and DME, we merge these two attention modules in a deep network. The ISBI 2018 IDRiD challenge dataset and the Messidor dataset are used to assess the performance of our network. Our solution outperforms previous methods on the ISBI 2018 IDRiD challenge dataset and the Messidor dataset.