Ophthalmology science · 2026
Predicting Visual Field Loss in Glaucoma Using OCT and Deep Learning: A Comparative Study of U-Net Variants.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Ohn Kyoung, Hwang Jiwook, Jung Jiwon, Whang Woong-Joo
Аннотация
PURPOSE: Glaucoma is a chronic eye disease that progressively damages the optic nerve, leading to irreversible visual field (VF) loss. OCT and VF tests are essential for monitoring structural and functional changes in glaucoma. This study applies 3 deep learning models-R2 U-Net, Dense U-Net, and Nested U-Net (UNet++)-to predict VF outcomes using retinal nerve fiber layer (RNFL) thickness maps from OCT images.
DESIGN: A retrospective cross-sectional study.
SUBJECTS: A total of 1640 patients with glaucoma diagnosed at a tertiary referral center were included. Only 1 eye (left eye) per patient was analyzed to avoid intereye correlation. Eyes included patients with early, moderate, and advanced glaucoma.
METHODS: We used a dataset of OCT and VF data from 1640 glaucoma patients, divided into training, validation, and test sets. The 3 deep learning models were trained and evaluated using 5 performance metrics: mean squared error (MSE), mean absolute error (MAE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and Fréchet inception distance (FID). The goal was to predict VF outcomes based on OCT-derived RNFL thickness maps.
MAIN OUTCOME MEASURES: Accuracy and image quality of predicted VF maps compared with ground truth VF maps, assessed by MSE, MAE, PSNR, SSIM, and FID.
RESULTS: R2 U-Net outperformed all other architectures, with the lowest MSE and MAE values and the highest SSIM and PSNR scores, indicating superior accuracy and image quality. Nested U-Net and Dense U-Net lagged, with Dense U-Net showing the lowest predictive accuracy.
CONCLUSIONS: This study is the first to apply generative artificial intelligence (AI) models, such as R2 U-Net, to predict VF loss based on OCT data, with both models demonstrating exceptional performance. These findings highlight the potential of generative AI to enhance glaucoma diagnosis and facilitate personalized treatment planning.
FINANCIAL DISCLOSURES: The authors have no proprietary or commercial interest in any materials discussed in this article.
DESIGN: A retrospective cross-sectional study.
SUBJECTS: A total of 1640 patients with glaucoma diagnosed at a tertiary referral center were included. Only 1 eye (left eye) per patient was analyzed to avoid intereye correlation. Eyes included patients with early, moderate, and advanced glaucoma.
METHODS: We used a dataset of OCT and VF data from 1640 glaucoma patients, divided into training, validation, and test sets. The 3 deep learning models were trained and evaluated using 5 performance metrics: mean squared error (MSE), mean absolute error (MAE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and Fréchet inception distance (FID). The goal was to predict VF outcomes based on OCT-derived RNFL thickness maps.
MAIN OUTCOME MEASURES: Accuracy and image quality of predicted VF maps compared with ground truth VF maps, assessed by MSE, MAE, PSNR, SSIM, and FID.
RESULTS: R2 U-Net outperformed all other architectures, with the lowest MSE and MAE values and the highest SSIM and PSNR scores, indicating superior accuracy and image quality. Nested U-Net and Dense U-Net lagged, with Dense U-Net showing the lowest predictive accuracy.
CONCLUSIONS: This study is the first to apply generative artificial intelligence (AI) models, such as R2 U-Net, to predict VF loss based on OCT data, with both models demonstrating exceptional performance. These findings highlight the potential of generative AI to enhance glaucoma diagnosis and facilitate personalized treatment planning.
FINANCIAL DISCLOSURES: The authors have no proprietary or commercial interest in any materials discussed in this article.