The application of deep learning in early enamel demineralization detection

Objective: The study aims to develop a diagnostic model using intraoral photographs to accurately detect and classify early detection of enamel demineralization on tooth surfaces.

Methods: A retrospective analysis was conducted with 208 patients aged 14 to 44. A total of 624 high-quality digital images captured under standardized conditions were used to construct a deep learning model based on the Mask region-based convolutional neural network (Mask R-CNN). The model was trained to automate the detection of enamel demineralization. Its performance was compared to two junior dentists' diagnostic abilities.

Results: The model achieved an F1-score of 0.856 for detecting demineralized teeth on the validation set, a metric that reflects comprehensive diagnostic performance, demonstrating performance close to that of senior dentists. With the the model's assistance, the junior dentists' average F1-scores improved significantly-from 0.713 and 0.689 to 0.897 and 0.949, respectively (p < 0.05). The model accurately segmented tooth surfaces and detected demineralized areas, allowing for precise detection of demineralized areas and monitoring of lesion progression.

Conclusion: Deep learning can accurately segment tooth surfaces and lesion contours, enhancing the precision, accuracy, and efficiency of enamel demineralization diagnosis and area delineation.