ABSTRACT
Background Bacterial vaginosis (BV) was the most common condition for women’s health caused by the disruption of normal vaginal flora and an overgrowth of certain disease-causing bacteria, affecting 30-50% of women at some time in their lives. Gram stain followed by Nugent scoring (NS) was long considered golden standard and based on bacterial morphotypes under the microscope. This conventional manual method often gave variable results among different technologists.
Methods We created a convolutional neural network (CNN), and evaluated its ability to auto-matic identify vaginal bacteria and classify Nugent scores from microscope images. All the CNN models were first trained with 23280 microscopic images diagnosed and archived either positive or negative for BV. A separate set of 5815 images were evaluated by the CNN model and technologists/obstetricians independently. The CNN model’s generalization ability was evaluated on total independent test sets of 1082 images collecting from three medical institutions.
Results Our model could classify Nugent Scores at the image-level with high sensitivity (82.4%) and specificity (96.6%), which was more consistent and had better diagnostic yield than the toplevel technologists and obstetricians in China. The speed of our CNN model was much faster than human reader. The generalization ability of our model was strong and the model could be deployed in more medical institutions.
Conclusion The CNN model over performed human readers on accuracy, efficiency and stability for BV diagnosis using microscopic image-based Nugent scores.
Competing Interest Statement
Authors Z. Wang, W. Mo, W. Wu and M. Li were employed by the company Suzhou Turing Microbial Technologies Co., Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.