Related Experiment Video
Updated: Feb 27, 2026
10:14
Author Spotlight: Ex Vivo OCT-Based Multimodal Imaging of Human Donor Eyes for Research into Age-Related Macular Degeneration
Published on: May 26, 2023
4.3K
Domain-Shift AI Technology for Vendor-Agnostic Multiple Macular Disease Detection From 3D OCT Scans
Zi Qi Tang1, Yu Han Zhang2, An Ran Ran1
1Department of Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong Special Administrative Region, China.
JAMA Ophthalmology
|February 26, 2026
Summary
A new deep learning (DL) model analyzes optical coherence tomography (OCT) scans from different vendors, achieving high accuracy for disease detection. This vendor-agnostic approach enhances diagnostic capabilities in diverse ophthalmic settings.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Deep learning (DL) models require diverse datasets for robust disease detection.
- Optical coherence tomography (OCT) scans are crucial for diagnosing macular conditions.
- Analyzing OCT scans from various vendors presents a significant challenge due to data heterogeneity.
Purpose of the Study:
- To develop a vendor-agnostic DL model for multidisease classification using 3D OCT scans from multiple vendors.
- To evaluate the model's performance across different OCT vendors and scan types (2D/3D).
- To incorporate an 'uncertain' category for novel macular conditions and a triage module.
Main Methods:
- A Residual Neural Network (ResNet) 3D model was trained on 3D OCT scans from Vendor 1.
- An unsupervised domain adaptation technique (Test Entropy) was employed to address vendor discrepancies.
- The model was tested on nine external datasets from Vendor 1 and Vendor 2, including 2D scans.
Main Results:
- The vendor-agnostic DL model demonstrated high diagnostic performance across different vendors, with Area Under the Receiver Operating Characteristic Curve (AUROC) ranging from 0.754 to 0.999.
- Micro-average Negative Predictive Values (NPVs) consistently exceeded 97.5%.
- The 'uncertain' category showed high specificity and accuracy but variable sensitivity, with clinically important miss rates below 9% for urgent and semi-urgent cases.
Conclusions:
- The developed vendor-agnostic DL model shows significant potential for broad deployment in eye care settings.
- This technology can streamline disease detection and triage in modern ophthalmic practices.
- The model's ability to handle diverse OCT data enhances its clinical utility and applicability.
