Diabetic Retinopathy (DR) is one of the eye diseases suffered by diabetes patients that will cause blindness if it does not get effectively treated for a certain period of time. Early detection is needed to help patients get effective treatment based on their severity. Researchers have done copious amounts of research regarding the methods for DR detection using shallow learning and deep learning approaches. The proposed method in this paper is a combination of two deep learning architectures, namely Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). CNN is used to detect lesions on retinal fundus images, and LSTM is used for generating description sentences based on those lesions. In the training and testing process, the CNN output will be used for the input of LSTM. The training process's target is to produce a model that can map retinal fundus images into a sentence. The results of this experiment using the MESSIDOR data set has an accuracy of around 90%.
|Journal||Journal of Physics: Conference Series|
|Publication status||Published - 7 Jan 2021|
|Event||10th International Conference and Workshop on High Dimensional Data Analysis, ICW-HDDA 2020 - Sanur-Bali, Indonesia|
Duration: 12 Oct 2020 → 15 Oct 2020