[Paper Review] retina-VAE: Variationally Decoding the Spectrum of Macular Disease

Stephen G. Odaibo|arXiv (Cornell University)|Jul 11, 2019

Retinal Imaging and Analysis39 references4 citations

TL;DR

retina-VAE introduces a variational autoencoder that learns a clinically relevant 3D latent space from patient profile vectors (pVecs) containing demographic and clinical data, spontaneously clustering into 14 biologically plausible subtypes of macular disease; these clusters reveal potential disease subtypes with distinct sex, race, age, and polyp status distributions, suggesting personalized treatment responses to anti-VEGF therapies.

ABSTRACT

In this paper, we seek a clinically-relevant latent code for representing the spectrum of macular disease. Towards this end, we construct retina-VAE, a variational autoencoder-based model that accepts a patient profile vector (pVec) as input. The pVec components include clinical exam findings and demographic information. We evaluate the model on a subspectrum of the retinal maculopathies, in particular, exudative age-related macular degeneration, central serous chorioretinopathy, and polypoidal choroidal vasculopathy. For these three maculopathies, a database of 3000 6-dimensional pVecs (1000 each) was synthetically generated based on known disease statistics in the literature. The database was then used to train the VAE and generate latent vector representations. We found training performance to be best for a 3-dimensional latent vector architecture compared to 2 or 4 dimensional latents. Additionally, for the 3D latent architecture, we discovered that the resulting latent vectors were strongly clustered spontaneously into one of 14 clusters. Kmeans was then used only to identify members of each cluster and to inspect cluster properties. These clusters suggest underlying disease subtypes which may potentially respond better or worse to particular pharmaceutical treatments such as anti-vascular endothelial growth factor variants. The retina-VAE framework will potentially yield new fundamental insights into the mechanisms and manifestations of disease. And will potentially facilitate the development of personalized pharmaceuticals and gene therapies.

Motivation & Objective

• To develop a clinically relevant latent representation for the spectrum of macular disease, particularly exudative forms requiring anti-VEGF therapy.
• To move beyond rigid, mutually exclusive disease classifications toward a continuous, personalized disease spectrum model.
• To identify latent subtypes with potential biological and therapeutic relevance using patient profile vectors (pVecs).
• To explore whether variational autoencoders can uncover biologically meaningful clusters from synthetic patient data reflecting real-world disease statistics.
• To enable future development of personalized pharmaceuticals and gene therapies by identifying patient subgroups with distinct treatment responses.

Proposed method

• A variational autoencoder (VAE) is trained on 3,000 synthetic 6-dimensional patient profile vectors (pVecs) representing exudative AMD, CSCR, and PCV.
• The pVecs include clinical findings and demographic variables: age, sex, race, presence of polyps, subretinal hemorrhage, and disease type.
• The model learns a 3D latent space that best reconstructs the input pVecs, with architecture selection based on reconstruction loss and training stability.
• Spontaneous clustering of latent vectors into 14 groups is observed, with K-means used only to identify and inspect cluster membership.
• Cluster properties are analyzed for patterns in sex, race, age, polyp status, and subretinal hemorrhage to assess biological plausibility.
• The framework leverages variational inference to generate a continuous, interpretable latent space that encodes clinically relevant disease subtypes.

Experimental results

Research questions

• RQ1 Can a variational autoencoder learn a clinically meaningful 3D latent space from patient profile vectors representing macular disease spectrum?
• RQ2 Do the latent vectors spontaneously form biologically plausible clusters that reflect known disease subtypes or demographic patterns?
• RQ3 Are there consistent associations between cluster membership and clinical variables such as sex, race, age, and polyp status?
• RQ4 Can the latent space capture shared pathophysiological mechanisms across exudative AMD, CSCR, and PCV, particularly those involving choroidal neovascularization?
• RQ5 Do the identified clusters suggest differential responses to anti-VEGF therapy, enabling personalized treatment prediction?

Key findings

• The 3-dimensional latent space architecture achieved the best training performance compared to 2D or 4D configurations, indicating optimal balance between expressiveness and generalization.
• The latent vectors spontaneously formed 14 distinct clusters without prior clustering constraints, suggesting inherent structure in the disease spectrum.
• Each cluster exhibited strong internal consistency in key demographic and clinical variables: for example, clusters were either all-male or all-female, and some were exclusively Asian with no other racial groups.
• Clusters 1, 4, 6, 8, 11, and 12 were composed solely of Asian patients, indicating potential race-specific disease subtypes.
• Clusters 2, 12, and 14 had the youngest median ages (39), while clusters 3, 4, 9, and 8 had the oldest median ages (78–79), suggesting age-related stratification.
• Subretinal hemorrhage was the only categorical variable distributed across all clusters, indicating it is not a defining feature of any single subtype.