RNFL maps
Peripapillary retinal nerve fibre layer thickness maps.
Ophthalmic AI
Interpretable representation learning for glaucoma progression.
Modelling retinal structure using OCT-derived RNFL and GCIPL thickness maps, with a focus on continuum-aware latent spaces.
Problem
Glaucoma is not only a classification problem.
In many medical AI settings, the goal is often framed as healthy versus disease classification. But glaucoma-related structural change can be better understood as a continuum involving normal, suspect, and glaucomatous patterns.
The research objective is to learn representations that can support classification while also revealing how retinal thickness structure changes across latent directions.
Data
OCT-derived retinal thickness maps.
GCIPL maps
Macular ganglion cell-inner plexiform layer thickness maps.
Eye orientation
Right-eye maps can be horizontally flipped for anatomical alignment.
Resized maps
High-resolution maps can be resized for efficient model training.
Pipeline
From OCT scan to interpretable latent representation.
OCT Scan
→
RNFL / GCIPL Map
→
Preprocessing
→
MixtureBetaVAE
→
Latent Traversal
Placeholder results
Result cards ready for real findings.
These cards are currently placeholders and can later be updated with final validated values, figures, and links to papers or thesis chapters.
Healthy AUC
Placeholder/result snapshot for healthy class discrimination.
Glaucoma AUC
Placeholder/result snapshot for glaucoma class discrimination.
Latent traversal
Space for clinically meaningful traversal examples and reconstructions.
Reconstruction
Space for reconstruction fidelity plots and error summaries.
Interpretability
What should the model help us ask?
What changes from healthy to suspect?
Latent traversals can be used to inspect structural transitions.
Which regions drive reconstruction error?
Domain-aware loss maps can highlight clinically relevant areas.
Are generated maps plausible?
Interpretability depends on traversals being coherent and realistic.
Next
Explore the shared method behind this domain.
The ophthalmic AI page connects to the broader MixtureBetaVAE framework used across research domains.
Open MixtureBetaVAE