Advancing Diabetic Retinopathy Research with a Dataset from Aizawl region of Mizoram

This study presents a high-resolution Diabetic Retinopathy dataset collected from Eye Care Hospital in Aizawl, Mizoram, highlighting its importance for advancing DR detection in underrepresented populations. The dataset includes 495 retinal images captured using an OCT RS 330 device, with quality maintained through a preprocessing pipeline. Images were filtered for clarity using blur metrics and enhanced with adaptive histogram equalization to emphasize diagnostically relevant features. A baseline deep learning pipeline for multiclass DR classification achieved training, validation, and test accuracies of 50%, 49%, and 56%, respectively, reflecting the dataset's complexity. Advanced augmentations like MixUp and CutMix were applied, setting a benchmark for future research. This work underscores the dataset's unique contributions by providing a foundation for robust diagnostic models. This study supports inclusive and practical AI applications in ophthalmology.