Advanced Segmentation of Diabetic Retinopathy Lesions Using DeepLabv3+
Meher Boulaabi, Takwa Ben Aïcha Gader, Afef Kacem Echi, Sameh Mbarek
TL;DR
This work tackles Diabetic Retinopathy lesion segmentation by training four lesion-type–specific binary DeepLabV3+ models (MA, HE, EX, SE) and fusing their outputs for a unified segmentation map. The pipeline leverages careful preprocessing (cropping and CLAHE on the L channel of LAB), modest data augmentation, and a constrained training regime to address limited annotations in IDRiD. Empirical results on IDRiD show near-perfect per-class accuracy and high IoU, with MAE/MSE metrics supporting precise pixel-level predictions and ROC analyses confirming robust class discrimination. The approach achieves state-of-the-art performance on a challenging, multi-label DR dataset, offering a practical path toward reliable automated DR lesion analysis in clinical settings.
Abstract
To improve the segmentation of diabetic retinopathy lesions (microaneurysms, hemorrhages, exudates, and soft exudates), we implemented a binary segmentation method specific to each type of lesion. As post-segmentation, we combined the individual model outputs into a single image to better analyze the lesion types. This approach facilitated parameter optimization and improved accuracy, effectively overcoming challenges related to dataset limitations and annotation complexity. Specific preprocessing steps included cropping and applying contrast-limited adaptive histogram equalization to the L channel of the LAB image. Additionally, we employed targeted data augmentation techniques to further refine the model's efficacy. Our methodology utilized the DeepLabv3+ model, achieving a segmentation accuracy of 99%. These findings highlight the efficacy of innovative strategies in advancing medical image analysis, particularly in the precise segmentation of diabetic retinopathy lesions. The IDRID dataset was utilized to validate and demonstrate the robustness of our approach.