Blog
This study proposes a deep learning-based pipeline for detecting early signs of gingivitis in orthodontic patients using intraoral photographic images acquired during routine treatment. Gingivitis is highly prevalent in orthodontic patients due to plaque accumulation around brackets and archwires, and early detection is essential for preventing progression to periodontitis. However, clinical visual inspection is subjective and may miss subtle inflammation. The authors therefore developed an object detection model based on Faster R-CNN to automatically identify early gingival inflammation in standardized clinical photographs. A total of 134 photographic images were collected from orthodontic patients at different treatment stages (T0, T1, T2). After exclusion of unusable images, the dataset was split into 107 training images and 27 testing images, with the split performed at the patient level to prevent leakage between train and test sets. Three key visual indicators of gingivitis—color changes, swelling, and bleeding—were annotated manually by trained clinicians, forming bounding-box labels for the model. The Faster R-CNN architecture (with a ResNet backbone) was optimized using standard hyperparameters and trained on the annotated dataset. The model demonstrated strong performance in detecting early gingivitis, achieving an accuracy of 90% in classifying images as “healthy” or “early gingivitis.” The object detection module reached a mean Average Precision (mAP) of 87%, showing effective localization of inflamed regions. Visual result maps revealed clear bounding boxes around affected gingival margins, supporting interpretability. The study concludes that Faster R-CNN is a promising tool for supporting early gingivitis detection in orthodontic settings. The model’s ability to detect subtle inflammatory changes can help clinicians intervene earlier, improving oral hygiene guidance and reducing the risk of periodontal deterioration throughout orthodontic treatment. With larger datasets and more granular labeling, the framework could evolve into a real-time clinical decision support system.
gingivitis; orthodontics; deep learning; Faster R-CNN; detection; intraoral images
Supports early gingivitis detection in orthodontic patients, improving preventive care and periodontal monitoring.