ScholarDeck

StudyThe British journal of ophthalmology · 2026

DeepAdapter: a generalisable algorithm integrating self-supervised learning and unsupervised domain adaptation for robust retinopathy of prematurity screening.

Jiaman Zhao, Longhui Li, Zhenzhe Lin … Haotian Lin

OphthalmologyStudy

DeepAdapter algorithm gave more accurate retinopathy of prematurity detection on outside data than standard AI.

DeepAdapter: a generalisable algorithm integrating self-supervised learning and unsupervised domain adaptation for robust retinopathy of prematurity screening.

Jiaman Zhao … Haotian Lin

The British journal of ophthalmology · 2026

Purpose

To develop and validate DeepAdapter, a novel deep learning algorithm that integrates self-supervised learning (self-supervised learning) and unsupervised domain adaptation (unsupervised domain adaptation) to enhance model generalisability for retinopathy of prematurity (retinopathy of prematurity) screening.

Methods

First, self-supervised learning was applied to 500 000 unlabelled infantile colour fundus photographs (colour fundus photographs) retrospectively collected from four Chinese clinical centres to learn general fundus representations.

Results

Diagnostic accuracy on external hospital data: 0.828 with DeepAdapter vs 0.739 with standard method.

DeepAdapter (external)

0.828accura

Supervised (external)

0.739accura

More results

Compared with the supervised method, DeepAdapter decreased Correlation Alignment distance from 0.297 to 3.550×10 -6 .

A web-based system integrating quality control and retinopathy of prematurity diagnosis was developed for large-scale, multicentre clinical screening.

“

Conclusion

DeepAdapter effectively mitigated domain shift and significantly improved model generalisability for retinopathy of prematurity screening, providing a valuable reference for developing generalisable models in other medical specialties.

Post

DeepAdapter: a generalisable algorithm integrating self-supervised learning and unsupervised domain adaptation for robust retinopathy of prematurity screening.

DeepAdapter algorithm gave more accurate retinopathy of prematurity detection on outside data than standard AI.

Related papers