An Integrated Deep Learning Framework for Diabetic Retinopathy: Channel and Spatial Attention U-Net for Lesion Segmentation and CNN-Based Fundus Image Denoising with Ensemble Feature Classification

Dr. Anand M ¹, Dr. Chaya P ², Dr. Vishwesh J ³, Dr. Neethi M V ⁴, Dr. Ayesha Taranum ⁵, Dr. Pradeep Kumar R ^6*

¹Associate Professor, Department of Information Science and Engineering, GSSS Institute of Engineering and Technology for Women, Mysuru, Karnataka, India 570016.
Email: anandm@gsss.edu.in
ORCID: 0009-0002-4787-7347

²Associate Professor, Department of Information Science and Engineering, GSSS Institute of Engineering and Technology for Women, Mysuru, Karnataka, India 570016.
Email: chayaneetha@gmail.com
ORCID: 0009-0000-1107-1902

³Associate Professor, Department of Computer Science and Engineering, GSSS Institute of Engineering and Technology for Women, Mysuru, Karnataka, India 570016.
Email: vishweshj@gsss.edu.in
ORCID: 0000-0003-2112-3512

⁴Assistant Professor, Department of Computer Science and Engineering-Data Science, ATME College of Engineering, Mysuru, Karnataka, India 570028.
Email: neethi.mv@gmail.com
ORCID: 0000-0003-1570-3094

⁵Associate Professor and Deputy HOD, Department of Computer Science and Engineering, Vidyavardhaka College of Engineering, Mysuru, Karnataka, India 570015.
Email: ayesha.cs@vvce.ac.in
ORCID: 0000-0002-3171-6656

^6*Assistant Professor, Department of Mathematics, GSSS Institute of Engineering and Technology for Women, Mysuru, Karnataka, India 570016.
Email: pradeepr.mysore@gmail.com
ORCID: 0000-0001-7152-0298

*Corresponding Author: Dr. Pradeep Kumar R, Assistant Professor, Department of Mathematics, GSSS Institute of Engineering and Technology for Women, Mysuru, Karnataka, India 570016. Email: pradeepr.mysore@gmail.com

---

ABSTRACT

Diabetic retinopathy is thought to be the primary cause of visual loss. It is a microvascular illness that specifically affects the retina, causing vessel obstruction that deprives the retinal tissues of nourishment. Early detection is key to effective treatment, since advanced stages might result in irreversible blindness or loss of vision. Therefore, efforts are made to build automatic detection systems that would both speed up and lower the cost of the identification process. Our study presents a deep learning architecture based on UNet for the segmentation of blood vessels, exudates, and microaneurysms in diabetic retinopathy. However, because to the small number of credible datasets, the accuracy of the current prediction algorithms is not yet good enough for eye specialists to rely on them as trustworthy diagnosis tools. In addition, a variety of noise kinds are included in the recorded data. Eliminating the noise thus becomes a crucial undertaking for this study. For filtering and classification, we thus looked into an approach that coupled denoising with ensemble-based learning. Noise is identified by residual noise mapping, a feature of CNN-based architecture used for filtering. Using an ensemble classifier, we categorize the features acquired in the following stage, which presents a CNN-based feature extraction model.

Keywords: Diabetic retinopathy, Segmentation, Denoising, Filtering, Classification.

How to cite this article: Anand M, Chaya P, Vishwesh J, Neethi MV, Taranum A, Pradeep Kumar R. An integrated deep learning framework for diabetic retinopathy: channel and spatial attention U-Net for lesion segmentation and CNN-based fundus image denoising with ensemble feature classification. Int J Drug Deliv Technol. 2026;16(7s): 156-165; DOI: 10.25258/ijddt.16.7s.19

Source of support: Nil.

Conflict of interest: None