Deep Imbalanced Regression Model for Predicting Refractive Error from Retinal Photos

Deep Imbalanced Regression Model for Predicting Refractive Error from Retinal Photos

Ophthalmology Science

10.1016/j.xops.2024.100659

https://doi.org/10.1016/j.xops.2024.100659

Samantha Min Er Yew, Xiaofeng Lei, Yibing Chen, Jocelyn Hui Lin Goh, Krithi Pushpanathan, Can Can Xue, Ya Xing Wang, Jost B. Jonas, Charumathi Sabanayagam, Victor Teck Chang Koh, Xinxing Xu, Yong Liu, Ching-Yu Cheng, Yih-Chung Tham

28 November 2024

Yew, S. M. E., Lei, X., Chen, Y., Goh, J. H. L., Pushpanathan, K., Xue, C. C., Wang, Y. X., Jonas, J. B., Sabanayagam, C., Koh, V. T. C., Xu, X., Liu, Y., Cheng, C.-Y., & Tham, Y.-C. (2025). Deep Imbalanced Regression Model for Predicting Refractive Error from Retinal Photos. Ophthalmology Science, 5(2), 100659. https://doi.org/10.1016/j.xops.2024.100659

Purpose: Recent studies utilized ocular images and deep learning (DL) to predict refractive error and yielded notable results. However, most studies did not address biases from imbalanced datasets or conduct external validations. To address these gaps, this study aimed to integrate the deep imbalanced regression (DIR) technique into ResNet and Vision Transformer models to predict refractive error from retinal photographs. Design: Retrospective study. Subjects: We developed the DL models using up to 103 865 images from the Singapore Epidemiology of Eye Diseases Study and the United Kingdom Biobank, with internal testing on up to 8067 images. External testing was conducted on 7043 images from the Singapore Prospective Study and 5539 images from the Beijing Eye Study. Retinal images and corresponding refractive error data were extracted. Methods: This retrospective study developed regression-based models, including ResNet34 with DIR, and SwinV2 (Swin Transformer) with DIR, incorporating Label Distribution Smoothing and Feature Distribution Smoothing. These models were compared against their baseline versions, ResNet34 and SwinV2, in predicting spherical and spherical equivalent (SE) power. Main Outcome Measures: Mean absolute error (MAE) and coefficient of determination were used to evaluate the models’ performances. The Wilcoxon signed-rank test was performed to assess statistical significance between DIR-integrated models and their baseline versions.

Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)

This research / project is supported by the A*STAR - Industry Alignment Fund - Pre-positioning
Grant Reference no. : H20c6a0031

This research / project is supported by the National Medical Research Council of Singapore - HPHSR Clinician Scientist Award - Investigator level
Grant Reference no. : NMRC/MOH/ HCSAINV21nov-0001

https://oar.a-star.edu.sg/communities-collections/articles/23351

2666-9145

Institute of High Performance Computing 

https://doi.org/10.1016/j.xops.2024.100659