Cornea · 2025
Deep Learning Model for Extensive Diagnosis of Corneal Deposits.
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.
Taki Yosuke, Ueno Yuta, Oda Masahiro, Kitaguchi Yoshiyuki, Maehara Hiroki, Yamaguchi Takefumi
Аннотация
PURPOSE: We developed a comprehensive artificial intelligence DL model (CorneAI) to classify cataracts and corneal diseases into 9 categories. In this study, we aimed to enhance CorneAI by developing a deep learning (DL) model to diagnose the following conditions associated with corneal deposition: amyloidosis, band keratopathy, granular corneal dystrophy (GCD), lattice corneal dystrophy (LCD), gelatinous drop-like dystrophy (GDLD), Schnyder corneal dystrophy (SCD), and macular corneal dystrophy (MCD).
METHODS: We retrospectively analyzed 1546 slit-lamp images of 7 corneal opacities using You Only Look Once version 5 (YOLOv5). The model was trained with 5-fold cross-validation and tested on 303 images. DL model performance was compared with specialists and residents and evaluated.
RESULTS: The positive predictive value (PPV) for all categories was 0.94 [95% confidence interval (CI): 0.93-0.96]. The highest PPVs for the individual diseases were relatively good, with 0.95 (0.93-0.97) for band keratopathy, 0.95 (0.94-0.97) for GCD, 0.83 (0.76-0.89) for LCD, 0.99 (0.97-0.99) for SCD, and 0.88 (0.65-0.96) for MCD. However, the PPVs for amyloid and GDLD were relatively low, at 0.75 (0.53-0.88) and 0.75 (0.46-0.91), respectively. When comparing DL model and human performance using a test dataset of 303 images, the PPVs were 0.96 (0.94-0.98) for our DL model, 0.89 (0.85-0.92) for specialists, and 0.67(0.61-0.72) for residents.
CONCLUSIONS: Our DL model accurately diagnosed band keratopathy, GCD, LCD, SCD, and MCD although it was not precise enough to classify amyloidosis or GDLD. Although its performance first needs to be improved, this will be the first DL model to be installed in CorneAI for future.
METHODS: We retrospectively analyzed 1546 slit-lamp images of 7 corneal opacities using You Only Look Once version 5 (YOLOv5). The model was trained with 5-fold cross-validation and tested on 303 images. DL model performance was compared with specialists and residents and evaluated.
RESULTS: The positive predictive value (PPV) for all categories was 0.94 [95% confidence interval (CI): 0.93-0.96]. The highest PPVs for the individual diseases were relatively good, with 0.95 (0.93-0.97) for band keratopathy, 0.95 (0.94-0.97) for GCD, 0.83 (0.76-0.89) for LCD, 0.99 (0.97-0.99) for SCD, and 0.88 (0.65-0.96) for MCD. However, the PPVs for amyloid and GDLD were relatively low, at 0.75 (0.53-0.88) and 0.75 (0.46-0.91), respectively. When comparing DL model and human performance using a test dataset of 303 images, the PPVs were 0.96 (0.94-0.98) for our DL model, 0.89 (0.85-0.92) for specialists, and 0.67(0.61-0.72) for residents.
CONCLUSIONS: Our DL model accurately diagnosed band keratopathy, GCD, LCD, SCD, and MCD although it was not precise enough to classify amyloidosis or GDLD. Although its performance first needs to be improved, this will be the first DL model to be installed in CorneAI for future.