Uncertainty-Aware Multi-Source Retinal Fluid Segmentation in OCT

2026-07-09

AI Summary

This study addresses the challenges of time-consuming manual annotation of fluid in retinal OCT images and the poor generalization of single-source segmentation models across diverse imaging devices. To overcome these limitations, the authors propose an attention-guided TransUNet that integrates domain-adaptive normalization with Bayesian pixel-wise uncertainty estimation, marking the first application of uncertainty modeling to multi-source OCT fluid segmentation. The method achieves robust segmentation of three types of retinal fluid across four independent OCT devices, attaining an average Dice coefficient of 0.78. Notably, predicted uncertainty is 1.34 times higher (p < 10⁻⁴) in regions where expert annotations disagree, demonstrating its potential as a reliable signal for clinical triage and significantly enhancing the model’s practical utility.

Abstract

Measuring retinal fluid from optical coherence tomography (OCT) drives treatment decisions in macular disease, but manual annotation is slow and segmentation models trained on one scanner degrade on another. We present an attention-guided TransUNet that segments three fluid types across four independent OCT sources, combining a domain-adaptive normalisation scheme with an uncertainty estimate that flags unreliable pixels. The model reaches a mean fluid Dice of 0.78, and -- most usefully for clinicians -- its uncertainty is 1.34x higher exactly where expert graders disagree (p<10^-4), turning a raw segmentation map into an actionable clinical triage signal.

Problem

• retinal fluid segmentation
• multi-source OCT
• domain generalization
• annotation variability
• clinical reliability

Innovation

• Uncertainty-aware segmentation
• Multi-source OCT
• Domain-adaptive normalization
• Attention-guided TransUNet
• Retinal fluid segmentation