Macretina: a dataset, to support deep learning assisted retinopathy of prematurity diagnosis

Abstract

Retinopathy of Prematurity (ROP) is a vision-threatening retinal disease found in premature babies, where early diagnosis is very important to prevent irreversible vision loss. In recent years, several studies have been conducted on the development of reliable AI-based screening systems. However, due to the lack of well-annotated public datasets most of them have been limited to experimental research, using single-central datasets. In this study, we introduce Macretina, a comprehensive and expert-annotated dataset curated from 1432 retinal fundus images of 112 premature babies collected at Macretina Hospital, Indore, India. These images were captured using the 3nethra Neo wide-field retinal imaging system, commonly used for retinopathy of prematurity (ROP) screening. The dataset is specially designed to support AI-based automated ROP diagnosis and is organized into three subsets, each addressing a distinct pathologically relevant retinal feature for ROP screening. The three subsets are: Macretina-Ridge which supports binary classification for ridge/demarcation line detection, Macretina-OD which supports object detection for optic disc localization, and Macretina-BV which supports semantic segmentation for blood vessel analysis. We also evaluated the utility of each subset using standard Deep Convolutional Neural Networks (DCNNs), and the experiments achieved promising results across Classification, Object Detection, and Segmentation tasks. Our dataset captures a wide range of disease severity and imaging variations, making it well-suited for developing clinically relevant and generalizable AI models. © The Author(s) 2025.