Cloud-based optimized deep learning framework for automated glaucoma detection using stationary wavelet transform and improved grey-wolf-optimization with ELM approach
Glaucoma, a progressive eye disease, can cause irreversible vision loss if not detected early. Timely diagnosis is crucial, especially in underserved areas where machine learning and cloud technology can offer a viable solution for remote glaucoma screening. This study presents an automated eHealth...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | Results in Engineering |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2590123025007595 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Glaucoma, a progressive eye disease, can cause irreversible vision loss if not detected early. Timely diagnosis is crucial, especially in underserved areas where machine learning and cloud technology can offer a viable solution for remote glaucoma screening. This study presents an automated eHealth system designed to enhance early glaucoma detection and mitigate its impact. The proposed model follows a multi-stage approach, beginning with the application of a stationary wavelet transform (SWT) for image preprocessing and augmentation. The augmented fundus images are subsequently processed through four convolutional neural network (CNN) models—ResNet50, InceptionV3, VGG16, and Xception—to extract deep features, which are then combined into a final feature matrix. To optimize the data for the next stage, principal component analysis (PCA) is applied to reduce the feature dimensions. Finally, an improved gray wolf optimization algorithm integrated with an extreme learning machine (IMGWO-ELM) classifies the images as either healthy or glaucomatous. This optimization enhances generalization and reduces overfitting, making the model a promising tool for advancing glaucoma diagnosis. The model's performance was evaluated in a cloud-based environment using two datasets: ORIGA and G1020. These datasets contain fundus images of individuals with and without glaucoma. We compared a stand-alone system with a cloud-based setup utilizing three virtual machines (4 vCPU–16 GB RAM, 8 vCPU–32 GB RAM, and 16 vCPU–64 GB RAM). A five-fold, ten-run cross-validation was employed across both configurations. The cloud setup with 16 vCPU–64 GB RAM achieved superior classification accuracies of 93.8% on the G1020 dataset and 96.74% on ORIGA. The deep CNN classifiers demonstrated exceptional performance, achieving a recall of 0.99 and an ROC score of 1.00, indicating perfect classification metrics. This study advances glaucoma detection by showcasing the efficacy of the ELM and CNN models, offering a promising direction for future research and improved patient outcomes. |
|---|---|
| ISSN: | 2590-1230 |