The ocular surface · 2026
Beyond diagnosis: AI-based surgical risk assessment in Acanthamoeba keratitis with real-world validation.
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.
Asadigandomani Hassan, Maizel Michael, Rahdar Amir, Soleimani Mohammad
Аннотация
PURPOSE: Acanthamoeba keratitis (AK) is a sight-threatening infection in which delayed diagnosis can lead to surgical intervention. This study aimed to develop and externally validate a deep learning (DL) model to distinguish patients requiring surgical management from those managed non-surgically within 2 weeks of presentation using slit-lamp photographs.
METHODS: A total of 1410 slit-lamp images from 104 culture-confirmed AK patients were labeled based on the need for tectonic keratoplasty within 2 weeks of presentation. Images underwent standardized preprocessing and augmentation before training a ResNet50-based convolutional neural network (CNN) using stratified 5-fold cross-validation. Performance was evaluated using confusion matrices, precision, recall, F1-scores, area under the ROC curve (AUC), and average precision (AP), with external validation conducted on an independent test set of 110 images using a predefined decision threshold of 0.44.
RESULTS: The model achieved a true-positive rate of 82% for surgical AK and a true-negative rate of 94% for non-surgical cases. Across folds, the mean AUC was 0.968 and the mean AP was 0.919, indicating strong discriminatory performance and high precision across thresholds. On external validation, the model achieved an accuracy of 89.1%, sensitivity of 86.7%, specificity of 92.0%, and an AUC of 0.968.
CONCLUSIONS: This study presents one of the first DL-based approaches to classify AK severity using slit-lamp photography and demonstrate robust generalizability through external validation. The model offers a practical tool for early identification of patients at risk for surgical progression, with potential to improve clinical triage, reduce diagnostic delays, and enhance access to timely care.
METHODS: A total of 1410 slit-lamp images from 104 culture-confirmed AK patients were labeled based on the need for tectonic keratoplasty within 2 weeks of presentation. Images underwent standardized preprocessing and augmentation before training a ResNet50-based convolutional neural network (CNN) using stratified 5-fold cross-validation. Performance was evaluated using confusion matrices, precision, recall, F1-scores, area under the ROC curve (AUC), and average precision (AP), with external validation conducted on an independent test set of 110 images using a predefined decision threshold of 0.44.
RESULTS: The model achieved a true-positive rate of 82% for surgical AK and a true-negative rate of 94% for non-surgical cases. Across folds, the mean AUC was 0.968 and the mean AP was 0.919, indicating strong discriminatory performance and high precision across thresholds. On external validation, the model achieved an accuracy of 89.1%, sensitivity of 86.7%, specificity of 92.0%, and an AUC of 0.968.
CONCLUSIONS: This study presents one of the first DL-based approaches to classify AK severity using slit-lamp photography and demonstrate robust generalizability through external validation. The model offers a practical tool for early identification of patients at risk for surgical progression, with potential to improve clinical triage, reduce diagnostic delays, and enhance access to timely care.