Enhanced detection and segmentation of retinal hard exudates in diabetic retinopathy using an FPN with CNN efficientnet-B0 encoder

This paper explores the application of a Feature Pyramid Network (FPN) with an EfficientNet-B0 Encoder for detecting and segmenting exudates in retinal images, a critical task in Diabetic Retinopathy assessment. Accurate exudate evaluation is vital as it provides insights into retinopathy progression. Our two-stage approach first detects and removes the optic disc to reduce false positives, followed by exudate detection and segmentation. Leveraging EfficientNet-B0 (LENB0) within the FPN architecture enables multi-scale feature representation, which is crucial for handling exudates of varying sizes and shapes. Evaluated on standard datasets, the proposed model achieves superior pixel accuracy and mIoU performance, demonstrating its potential as a reliable clinical tool.