3.1. Data Acquisition

In this study, the author utilizes the APTOS and EyePacs datasets. The APTOS dataset is available in the repository [42]. It was generated by Aravind Eye Hospital, India. The clinicians captured the fundus images across India. The images were captured using multiple cameras. Thus, the images contain noise and artifacts. The EyePacs dataset is publicly available in the Kaggle repository [43]. It covers a larger number of fundus images. The images were collected from primary care centers across the USA. The dataset provider resized the images into 1024 × 1024 pixels and cropped the black spaces. Based on the severity, the clinicians rated each image as 0 (no DR), 1 (mild), 2 (moderate), 3 (severe), and 4 (proliferative DR). Table 1 presents the properties of the datasets and the definition of the notations used in this section is presented in Table 2. Figure 2 shows the sample images of the datasets.

Sample images.

Dataset characteristics.

Notation and definition.