Diabetic Retinopathy (DR) affects people who have diabetes mellitus for a long period (20 years). It is one of the most common causes of preventable blindness in the world. If not detected early, this may cause irreversible damage to the patient's vision. One of the signs and serious DR anomalies are exudates, so these lesions must be properly detected and treated as soon as possible. To address this problem, the authors propose a novel method that focuses on the detection and classification of Exudateas Hard and soft in retinal fundus images using deep learning. Initially, the authors collected the retinal fundus images from the IDRID dataset, and after labeling the exudate with the annotation tool, the YOLOV3 is trained with specific parameters according to the classes. Then the custom detector detects the exudate and classifies it into hard and soft exudate.
Top2. Literature Survey
In the paper (Prem, S. S., & Umesh, A. C., 2020) an algorithm focuses on DR detection based on features such as local binary pattern (LBP) and wavelet transform approximation coefficient matrix. Images are classified as exudate and nonexudative by using machine learning classification algorithms such as support vector machine (SVM), k-nearest neighbor (KNN), decision tree, random forest (RF), and artificial neural network (ANN). The exudate classification includes three main stages there are Image Enhancement and Segmentation (Pandey, B. K. et.al., 2021), Feature Extraction, and Image Classification. For implementing this process DIARETDB1 dataset is used. It consists of 89 images of size 1500 x 1152. In the article (Mateen, M. et. al., 2020), the pre-trained convolutional neural network (CNN) has been used for exudates detection. The retinal datasets are used for experiments: (i) e-Ophtha and (ii) DIARETDB1. E-Ophtha dataset contains 47 retinal fundus images. DIARETDB1 dataset contains 89 retinal fundus photographs with a resolution of 1500 ×1 152. The data preprocessing is performed for the standardization of exudate patches. Furthermore, region of interest (ROI) localization is used to localize the features of exudates, and then transfer learning is performed for feature extraction using pre-trained CNN (Pandey, D. et.al., 2021) models (Inception-v3, Residual Network-50, and Visual Geometry Group Network-19).
In the article (Lin, L. et. al., 2020), a total of 603 fundus images from DR patients and 631 fundus images from healthy people were collected from the Department of Ophthalmology, Gaoyao People’s Hospital, and Zhongshan Ophthalmic Center, Sun Yat-sen University. In addition to exudate annotations, they also provide four additional labels for each image: left versus-right eye label, DR grade (severity scale) from three different grading protocols, and the bounding box of the optic disc (OD), and fovea location. In the article (Center, V. T. U. P. G., 2020) a supervised learning technique called linear regression is used. The database used are DiaRetDB0, DiaRetDB1, MESSIDOR and IDRID. A total of 60 images with both healthy and unhealthy retinal images are considered. Image processing and linear regression, a machine learning technique is involved which is used to train machines to differentiate between optic disk and exudates. In this paper (Borsos, B. et.al., 2019), the database used here is Indian Diabetic Retinopathy Image Dataset (IDRID). The preprocessing is done using three phases such as background and foreground pixel extraction and a data normalization operator which is similar to Z-transform. The modified algorithm SLIC is used for providing homogeneous super pixels in the images. Then the ANN based classification of pixels using fifteen features are extracted.