An Effective Heuristic Optimizer with Deep Learning-assisted Diabetic Retinopathy Diagnosis on Retinal Fundus Images
Diabetic Retinopathy (DR), a common diabetes complication affecting retinal blood vessels, may result in vision damage if not addressed promptly. Early and accurate detection is crucial for effective management, and Deep Learning (DL) techniques offer promising tools for the automated screening of R...
| Published in: | Engineering, Technology & Applied Science Research |
|---|---|
| Main Authors: | , |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
D. G. Pylarinos
2024
|
| Subjects: | |
| Online Access: | https://doi.org/10.48084/etasr.7004 https://doaj.org/article/916757c5895048f4b2ff1ac821ede453 |
| Summary: | Diabetic Retinopathy (DR), a common diabetes complication affecting retinal blood vessels, may result in vision damage if not addressed promptly. Early and accurate detection is crucial for effective management, and Deep Learning (DL) techniques offer promising tools for the automated screening of Retinal Fundus Images (RFIs). This approach enhances objectivity, reduces inter-observer variability, and has the potential to extend the DR diagnoses to regions with limited access to specialized medical professionals. This manuscript presents the design of the Beluga Whale Optimizer (BWO) with Deep Learning (DL)-assisted DR Diagnosis on RFIs (BWODL-DRDRFI) technique in the Internet of Things (IoT) platform. The proposed technique automatically examines the RFIs for identifying and classifying DR. During the IoT-based data-gathering procedure the patient utilizes a head-mounted camera for capturing the RFI and sends it to a cloud server. Median Filtering (MF)-based image preprocessing is performed to eradicate noise. Next, the BWODL-DRDRFI technique exploits the ShuffleNet-v2 approach to derive feature vectors. For DR recognition, the BWODL-DRDRFI technique applies a deep Stacked AutoEncoder (SAE) model. Finally, the BWO model optimally adjusts the hyperparameter values of the DSAE model for greater classification performance. The simulation output of the BWODL-DRDRFI approach can be examined on a standard image dataset and the outputs are computed on discrete measures. The simulation result highlighted the enhanced performance of the BWODL-DRDRFI approach in the DR diagnosis process. |
|---|