Stage Aware Diagnosis of Diabetic Retinopathy via Ordinal Regression

TL;DR

The paper addresses automatic staging of Diabetic Retinopathy into five ordinal grades (0–4) using fundus imagery from the APTOS-2019 dataset. It proposes an Ordinal Regression framework based on a ResNet50 backbone with a standardized preprocessing pipeline (Green Channel isolation, CLAHE, median filtering) and a 3-channel 224×224 input, trained with Mean Squared Error to produce a continuous severity score clamped to $[0,4]$. Evaluation with Quadratic Weighted Kappa ($QWK$) yields a peak of $0.8992$ on the validation set, indicating strong agreement with clinical grading and surpassing several contemporaries. While achieving state-of-the-art performance, the approach faces practical overhead from preprocessing and calls for enhanced interpretability and exploration of alternative architectures to improve robustness in clinical deployments.

Abstract

Diabetic Retinopathy (DR) has emerged as a major cause of preventable blindness in recent times. With timely screening and intervention, the condition can be prevented from causing irreversible damage. The work introduces a state-of-the-art Ordinal Regression-based DR Detection framework that uses the APTOS-2019 fundus image dataset. A widely accepted combination of preprocessing methods: Green Channel (GC) Extraction, Noise Masking, and CLAHE, was used to isolate the most relevant features for DR classification. Model performance was evaluated using the Quadratic Weighted Kappa, with a focus on agreement between results and clinical grading. Our Ordinal Regression approach attained a QWK score of 0.8992, setting a new benchmark on the APTOS dataset.

Figures (4)

• Figure 1: Stages of Diabetic Retinopathy (per ICDRSS)
• Figure 2: Preprocessing Pipeline
• Figure 3: Ordinal Regression with ResNet50 Model Architecture
• Figure 4: Ordinal Regression Test Set Confusion Matrix