Web-based Training Application to Support Manual Annotation of Diabetic Retinopathy Features on Retinal Fundus Images

Diabetic Retinopathy (DR) is a chronic eye condition which occurs as the most common consequence of Diabetes Mellitus (DM). Without continuous monitoring by competent ophthalmologists, DR may lead to worse complications, including loss of vision. With the increasing number of DM population in Indonesia, there is also an increasing needs for more ophthalmologists to detect and manage DR patients. Therefore, ophthalmologists need an accessible learning platform to improve their ability to establish a fast and accurate DR diagnosis. We propose web-based training application called Optan, which accommodates repeated manual annotation of DR features on digital fundus images and provides a quantitative evaluation of its accuracy. Such functionalities will provide ophthalmologists in training with a more precise feedback on their diagnostic performance. Optan web interface is designed with a simple layout to make it easily understandable by users with a choice of various shapes to annotate different DR features. In addition, it is equipped with a function to calculate the Jaccard index similarity of annotations made by a resident ophthalmologist compared to a reference specialist. We evaluated Optan user satisfaction and quantitative evaluation feature on respondents including ophthalmologists and academics working with fundus images. Optan scores an average of 7.3 from the scale of 9 in the standard QUIS user satisfaction survey. The highest annotation consistency of six ophthalmologists is achieved on optic disk circular annotation as opposed to microaneurysm point and blood vessel line annotations.