Unsupervised learning improves retinal scan analysis across devices
International Multi-Centre Validation of Unsupervised Domain Adaptation for Precise Discrimination between Normal and Abnormal Retinal Development
Computer Vision and Pattern RecognitionMachine Learning
Summary
Retinal scans help detect eye problems in children, but differences in scanning machines make it hard to compare results across hospitals. The authors developed a computer program that learns to tell normal from abnormal retinal development even when the scans come from different machines, without needing expert labels on every machine. They tested this approach using scans from five centers in four countries and found it worked well, almost as good as human experts. This method can help doctors consistently assess retinal health and use large collections of eye scans from different devices.
What this means in practice
- •For hospital imaging teams: Enable consistent classification of retinal scans from different OCT devices without the need for labelled data on each device.
- •For medical ai software developers: Create adaptive retinal image analysis tools that maintain accuracy across various OCT scanners for broader clinical deployment.$Commercial implications: This paper enables software products that provide reliable cross-device retinal scan assessments, removing the need for costly per-device labelling.
Authors
Zhanhan Tu, Nikhil Reddy, Helen J. Kuht, Ha-Jun Yoon, Gail D. E. Maconachie, Ravi Purohit, Prateek Pani, Line Kessel, Jinu Han, Jing Jin, Shui-Hua Wang, Karthik Bharadwaj, Viral Sheth, Michael Hisaund, Rebecca J. McLean, George Garratt, Chetan Kantibhai Patel, Kanmin Xue, Callum Hunt, Zisen Zhou, Ian De Silva, Seema Teli, Sonal Shah, Vanessa Rodwell, Aarti Patel, Irene Gottlob, Yu-Dong Zhang, Zedong Zheng, Girish Varma, Mervyn G. Thomas
Abstract
Arrested retinal development can cause childhood visual impairment, but grading optical coherence tomography scans is difficult to standardise across devices and centres. We developed and evaluated an unsupervised domain adaptation model to classify normal and abnormal retinal development across different OCT systems. This retrospective multicentre study used OCT scans from five centres in four countries and the UK Biobank. Data covered four OCT devices, including table mounted and handheld systems. A ResNet50 based Domain Adversarial Neural Network was trained using labelled images from a source device and unlabelled images from a target device. We evaluated six source to target pairings and compared model performance with seven clinician graders. A separate Heidelberg Spectralis dataset was reserved for external evaluation. Across validation sets, accuracy ranged from 87.75% to 96.06%, sensitivity from 88.68% to 98.38%, and specificity from 78.00% to 98.02%. The best performing model, adapted from TM OCT1 to HH OCT1, achieved 96.06% accuracy and an area under the receiver operating characteristic curve of 0.994, with a 95% confidence interval from 0.981 to 0.999. Several models also maintained discrimination on the unseen Heidelberg device. Unsupervised domain adaptation enabled OCT classification across devices without labelled target device data. The approach may support more consistent assessment of retinal development and large scale analysis of existing imaging datasets.