Eye (London, England) · 2025
Deep learning algorithms for timely diagnosis of retinopathy of prematurity requiring treatment.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
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.
Shoeibi Nasser, Ameri Nafise, Hoseinkhani Mohammad Reza, Ansari Astaneh Mohammad Reza, Motamed Shariati Mehrdad, Hosseini Seyedeh Maryam, Abrishami Mojtaba, Abrishami Majid, Zamani Ghodsieh, Heidarzadeh Hamid Reza
Аннотация
BACKGROUND/OBJECTIVES: To evaluate the effectiveness of deep learning (DL) algorithms in diagnosing Retinopathy of Prematurity (ROP) cases that requires treatment using fundus images submitted to the ROP clinic as part of a telemedicine consultation system.
SUBJECTS/METHODS: This retrospective cross-sectional study analysed 1700 RetCam fundus images from 141 preterm infants screened for ROP at Khatam-Al-Anbia Eye Hospital. The images underwent preprocessing using Contrast Limited Adaptive Histogram Equalisation (CLAHE), Automated Multiscale Retinex (AMSR), and a machine learning-based optimisation approach (ML). Various convolutional neural network (CNN) models such as MobileNet, ResNet-18, ResNet-50, and DenseNet-121, were evaluated for their diagnostic performance utilising accuracy, sensitivity, specificity, and F1-score metrics.
RESULTS: Among the models tested, MobileNet with CLAHE preprocessing achieved the highest accuracy (91.39%) and sensitivity (94.90%), establishing it as the most effective model for ROP detection. DenseNet-121 with CLAHE preprocessing showcased high sensitivity (94.26%) but slightly lower accuracy (90.98%). Additionally, ResNet-50 with AMSR preprocessing also demonstrated high accuracy (90.58%) and sensitivity (91.44%). These findings underscore the feasibility of DL models for real-time ROP screening in telemedicine environments.
CONCLUSION: MobileNet with CLAHE preprocessing exhibited the highest diagnostic performance in identifying treatment-requiring ROP, positioning it as a promising tool for AI-assisted screening. Further validation in varied clinical settings is necessary to confirm its real-world applicability.
SUBJECTS/METHODS: This retrospective cross-sectional study analysed 1700 RetCam fundus images from 141 preterm infants screened for ROP at Khatam-Al-Anbia Eye Hospital. The images underwent preprocessing using Contrast Limited Adaptive Histogram Equalisation (CLAHE), Automated Multiscale Retinex (AMSR), and a machine learning-based optimisation approach (ML). Various convolutional neural network (CNN) models such as MobileNet, ResNet-18, ResNet-50, and DenseNet-121, were evaluated for their diagnostic performance utilising accuracy, sensitivity, specificity, and F1-score metrics.
RESULTS: Among the models tested, MobileNet with CLAHE preprocessing achieved the highest accuracy (91.39%) and sensitivity (94.90%), establishing it as the most effective model for ROP detection. DenseNet-121 with CLAHE preprocessing showcased high sensitivity (94.26%) but slightly lower accuracy (90.98%). Additionally, ResNet-50 with AMSR preprocessing also demonstrated high accuracy (90.58%) and sensitivity (91.44%). These findings underscore the feasibility of DL models for real-time ROP screening in telemedicine environments.
CONCLUSION: MobileNet with CLAHE preprocessing exhibited the highest diagnostic performance in identifying treatment-requiring ROP, positioning it as a promising tool for AI-assisted screening. Further validation in varied clinical settings is necessary to confirm its real-world applicability.