U-net segmentation and ensemble learning for automated diabetic retinopathy stage identification
Author Identifier (ORCID)
Leisa Armstrong: https://orcid.org/0000-0002-9634-016X
Abstract
Diabetic retinopathy (DR) is considered as one of main causes of vision loss, progressing through various stages based on vascular changes. Early detection is crucial, but current screening methods often miss DR until irreversible damage occurs. Manual diagnosis using retinal fundus images is laborintensive, prone to human error, and requires skilled professionals. To address these challenges, automated image processing and machine learning (ML) techniques are increasingly vital. In this study, a comprehensive image preprocessing pipeline including enhancement, filtering, and U-Netbased segmentation was implemented prior to model evaluation. This paper examines the evolution of eight traditional machine learning (ML) algorithms and five pre-trained deep learning (DL) models on a 5500 DR fundus image dataset. To further improve classification accuracy, a stacked ensemble framework was developed, integrating deep feature extraction from the topperforming DL models, ResNet-101 and GoogLeNet. By integrating the strengths of top-performing models, the proposed stacked classifier demonstrated superior performance, achieving an F1-score of 0.9719, a recall of 0.9677, a precision of 0.9811, and an accuracy of 0.9722. These results surpass the performance of individual machine learning and deep learning methods, as well as existing studies in the field.
Document Type
Conference Proceeding
Date of Publication
1-1-2025
Publication Title
2025 International Conference on Quantum Photonics, Artificial Intelligence, and Networking (QPAIN)
Publisher
IEEE
School
School of Science
RAS ID
84808
Funders
Military Institute of Science and Technology
Copyright
subscription content
Comments
Ghose, R., Armstrong, L., & Sazzad, T. S. (2025). U-net segmentation and ensemble learning for automated diabetic retinopathy stage identification. In 2025 International Conference on Quantum Photonics, Artificial Intelligence, and Networking (QPAIN) (pp. 1-6). IEEE. https://doi.org/10.1109/QPAIN66474.2025.11171824