Classification of Retinal OCT Images Using Deep Learning
Retinal disorders develop slowly and without obvious signs. By the time symptoms arrive, damage is done. Optical coherence tomography sees earlier — if something is watching the scans.
Venue
IEEE ICCCI 2022
Year
2022
Role
Co-author · undergrad research
DOI
10.1109/ICCCI54379.2022.9740985
01 · Research problem
IndexChapter 01
The research problem
Retinal disorders have become a serious public health concern, and they are unusually cruel: they progress quietly, without obvious signs, until vision is already lost. Optical coherence tomography produces cross-sectional scans of the retina that reveal those changes early — but reading them at population scale requires more ophthalmologists than exist.
Chapter 02
Method
The paper applies deep learning to multi-class classification of retinal OCT images, distinguishing healthy retinas from a set of distinct retinal conditions. It was presented at the 2022 International Conference on Computer Communication and Informatics.
Stage 01 — OCT scans
Optical coherence tomography produces cross-sectional scans of the retina that reveal disease changes earlier than symptoms appear.
Illustrative pipeline — the DOI resolves to the authoritative IEEE Xplore record.
The companion artifact is the Retinal OCT-C8 dataset — 24,000 images across 8 conditions with balanced folds, now indexed in the openmedlab Awesome-Medical-Dataset collection.
Chapter 03
Results scope
Chapter 04
Scope and authorship
Cite this
Subramanian, M., Shanmugavadivel, K., Naren, O. S., Premkumar, K., & Rankish, K. (2022). Classification of Retinal OCT Images Using Deep Learning. 2022 International Conference on Computer Communication and Informatics (ICCCI).
doi.org/10.1109/ICCCI54379.2022.9740985 ↗Lessons
What it taught me
- This was the first artifact of mine that other people built on. Watching strangers train models on the OCT-C8 dataset taught me that infrastructure outlives results.
- Balanced folds are a kindness to whoever comes next. Most of the effort was not modelling — it was making the data trustworthy.
- Eight classes of retinal disease is a taxonomy, and taxonomies are interface design. Naming things well is the same skill in a hospital and in a component library.