End-to-end Two-Branch Classifier for Retinal Imaging Analysis

Retinal image analysis is essential in the identification and classification of various retinal diseases. Automatic identification of retinal disease is a big step towards early diagnosis and prevention of disease exacerbation. However, the existing retinal image open datasets have limitations because they do not consist of various diseases. To solve this problem, we participated in the IEEE International Symposium on Biomedical Imaging Challenge 2021 - Retinal Image Analysis for multi-Disease Detection (RIADD) Challenge and were placed in 4th place. This study shows that we extend our proposed Two-Branch Classifier model to solve the problem and achieve better performance (overall integration score increased by 2.23%). The model is trained end-to-end on 1,920 cases and tested on 620 cases.