Anatomy-Aware Unsupervised Detection and Localization of Retinal Abnormalities in Optical Coherence Tomography

TL;DR

This paper introduces an unsupervised framework for detecting and localizing retinal abnormalities in OCT images by modeling healthy anatomy, avoiding the need for lesion annotations, and demonstrating strong cross-dataset generalization.

Contribution

It presents a novel unsupervised anomaly detection method leveraging a discrete latent model with retinal layer-aware supervision, improving robustness and generalization in retinal OCT analysis.

Findings

Achieved AUROC 0.799 on Kermany dataset, outperforming baselines.

Demonstrated AUROC 0.884 on Srinivasan dataset, indicating strong domain adaptation.

Achieved Dice 0.200 and mIoU 0.117 on RETOUCH benchmark for anomaly segmentation.

Abstract

Reliable automated analysis of Optical Coherence Tomography (OCT) imaging is crucial for diagnosing retinal disorders but faces a critical barrier: the need for expensive, labor-intensive expert annotations. Supervised deep learning models struggle to generalize across diverse pathologies, imaging devices, and patient populations due to their restricted vocabulary of annotated abnormalities. We propose an unsupervised anomaly detection framework that learns the normative distribution of healthy retinal anatomy without lesion annotations, directly addressing annotation efficiency challenges in clinical deployment. Our approach leverages a discrete latent model trained on normal B-scans to capture OCT-specific structural patterns. To enhance clinical robustness, we incorporate retinal layer-aware supervision and structured triplet learning to separate healthy from pathological representations, improving model reliability across varied imaging conditions. During inference, anomalies are detected and localized via reconstruction discrepancies, enabling both image and pixel-level identification without requiring disease-specific labels. On the Kermany dataset (AUROC: 0.799), our method substantially outperforms VAE, VQVAE, VQGAN, and f-AnoGAN baselines. Critically, cross-dataset evaluation on Srinivasan achieves AUROC 0.884 with superior generalization, demonstrating robust domain adaptation. On the external RETOUCH benchmark, unsupervised anomaly segmentation achieves competitive Dice (0.200) and mIoU (0.117) scores, validating reproducibility across institutions.