Deep Learning Models for Predicting the Nugent Score to Diagnose Bacterial Vaginosis

Abstract

The Nugent score is a commonly used tool for diagnosing bacterial vaginosis; however, its accuracy depends on the skills of laboratory technicians. We aimed to evaluate the performance of deep learning models in predicting the Nugent score, with the goal of improving diagnostic consistency and accuracy. A total of 1,510 vaginal images collected from a hospital in Japan between 2021 and 2023 were assessed. Each image was annotated by laboratory technicians into one of four categories based on the Nugent—scorenormal vaginal flora, absence of vaginal flora, altered vaginal flora, or bacterial vaginosis. Deep learning models were developed to predict these categories, and their performance was evaluated by comparing the predicted scores with technician annotations. A high magnification model was further optimized and evaluated using an independent test set of 106 images to assess its performance relative to that of the technicians. The deep learning models demonstrated an accuracy of 84% at low magnification and 89% at high magnification in predicting the Nugent score categories. After optimization, the high magnification model achieved 94% accuracy, surpassing the average 92% accuracy of the technicians. The agreement between deep learning model predictions and technician annotations was 92% for normal vaginal flora, 100% for absence of vaginal flora, 91% for altered vaginal flora, and 100% for bacterial vaginosis. The deep learning models demonstrated accuracy comparable to that of laboratory technicians, which indicates their potential utility in improving the diagnostic accuracy of bacterial vaginosis.