Unsupervised Rank-Based Fusion with Rough Fuzzy C-Means (URRFCM) for Diabetic Retinopathy Detection

Resumen

Diabetic retinopathy (DR) is the most prevalent type of diabetic eye disease and can affect all individuals at any stage of diabetes. If not detected and treated at an early stage, DR may progress to severe vision impairment or even permanent visual loss. Early diagnosis is therefore crucial for improving treatment outcomes. To facilitate timely detection, the development of automated, computer-aided diagnostic systems offers a fast, cost-effective, and widely accessible solution. This study presents Unsupervised Rank Based Fusion with Rough Fuzzy C-Means (URRFCM) framework for DR detection. The proposed method employs multiple pretrained convolutional neural networks (CNNs) for feature extraction, Rank Based Fusion (RBF) technique for feature fusion and intigrate Rough Fuzzy C-Means approach (RFCM) for DR detection. In RFCM, rough set theory is applied to enhance the extracted features, finally, fuzzy clustering is applied to refine the deep clustering outputs. Extensive experiments conducted on three diabetic retinopathy datasets, that the proposed framework significantly outperforms five state-of-the-art deep clustering approaches in DR detection, as validated through comprehensive analysis.