Enhancing Ocular Disease classification with ConvNeXt-Tiny

Early detection of eye diseases is critical to the prevention of vision loss and timely clinical intervention to improve clinical outcomes. In this study, we are proposing a new deep learning framework for multi-class retinal image classification that is based on the ConvNeXt-Tiny architecture. The proposed approach was evaluated using two publicly available datasets namely, ODIR-5K (8 classes) and Eye Disease (4 classes). We trained and compared two model configurations. First, a baseline ConvNeXt-Tiny and its' enhanced variant which was the augmented with Squeeze-and-Excitation (SE) attention blocks. The baseline model achieved 92% accuracy on ODIR-5K and 91% on Eye Disease datasets which are equal or better than present published results. With the addition of SE attention resulted in improved performance with 96% accuracy on ODIR-5K and 95% on Eye Disease. F1 scores and recall also improved with difficult classes such as Normal and Glaucoma. These results demonstrate that by integrating channel-wise attention into a lightweight CNN–Transformer hybrid backbone improves feature discrimination and generalization across datasets. Our approach offers a robust and computationally efficient solution for creating automated retinal disease screening.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11318600/v1
Primary Topic
Retinal Imaging and Analysis
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Enhancing Ocular Disease classification with ConvNeXt-Tiny

Sajid Rahim, Nabila Zrira, Zineb Farahat, Ibtissam Benmiloud et al.
Research Square
Retinal Imaging and Analysis
preprint

Enhancing Ocular Disease classification with ConvNeXt-Tiny

Sajid Rahim, Nabila Zrira, Zineb Farahat, Ibtissam Benmiloud, Fatimazahra IDOUZGARH
preprint en

Abstract

Early detection of eye diseases is critical to the prevention of vision loss and timely clinical intervention to improve clinical outcomes. In this study, we are proposing a new deep learning framework for multi-class retinal image classification that is based on the ConvNeXt-Tiny architecture. The proposed approach was evaluated using two publicly available datasets namely, ODIR-5K (8 classes) and Eye Disease (4 classes). We trained and compared two model configurations. First, a baseline ConvNeXt-Tiny and its' enhanced variant which was the augmented with Squeeze-and-Excitation (SE) attention blocks. The baseline model achieved 92% accuracy on ODIR-5K and 91% on Eye Disease datasets which are equal or better than present published results. With the addition of SE attention resulted in improved performance with 96% accuracy on ODIR-5K and 95% on Eye Disease. F1 scores and recall also improved with difficult classes such as Normal and Glaucoma. These results demonstrate that by integrating channel-wise attention into a lightweight CNN–Transformer hybrid backbone improves feature discrimination and generalization across datasets. Our approach offers a robust and computationally efficient solution for creating automated retinal disease screening.

Research Square
National School of Mineral Industry (MA), Université Internationale Abulcasis des Sciences de la Santé (MA), McMaster University (CA)
Retinal Imaging and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon