Li Wei, Li Linlin, Xu Wenchong, Guo Yuting, Xu Min, Huang Shengyuan, Dai Dong, Lu Chang, Li Shuai, Lin Jiang
Department of Stomatology, Beijing Tongren Hospital, Capital Medical University, Beijing, China.
State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing, China.
Int Dent J. 2025 Jun;75(3):2104-2114. doi: 10.1016/j.identj.2025.01.002. Epub 2025 Jan 27.
The assessment of gingival inflammation surface features mainly depends on subjective judgment and lacks quantifiable and reproducible indicators. Therefore, it is a need to acquire objective identification information for accurate monitoring and diagnosis of gingival inflammation. This study aims to develop an automated method combining intraoral scanning (IOS) and deep learning algorithms to identify the surface features of gingival inflammation and evaluate its accuracy and correlation with clinical indicators.
The study included the periodontal probing data and intraoral scan images of 120 patients with periodontitis. The deep learning model GC-U-Net was used to automatically segment and identify the gingival inflammation regions. The performance of the model was evaluated by the Dice coefficient (Dice), intersection over union (IoU), and pixel accuracy (PA), and the correlation between the identification performance and the periodontal examination index was analysed.
The GC-U-Net model showed high recognition accuracy, with a Dice of 77.8%, an IoU of 65.4%, and a PA of 93.7%. This model demonstrated a strong positive correlation with the sulcus bleeding index (SBI; r = 0.836, P < .001), a moderately strong positive correlation with the bleeding index (BI; r = 0.618, P < .001), and a negative correlation with the probe depth (PD; r = - 0.425, P < .001).
The study has successfully developed an automatic identification method for surface characteristics of gingival inflammation based on deep learning and IOS technology, providing a standardised and automated auxiliary tool for clinical gingival inflammation examination with high accuracy and significant correlation with clinical indicators.
This method can reduce subjective judgment in the clinical assessment of gingival inflammation, improve the consistency and reliability of identification, and play an important auxiliary role in clinical diagnosis and treatment planning.
牙龈炎症表面特征的评估主要依赖主观判断，缺乏可量化和可重复的指标。因此，需要获取客观识别信息以准确监测和诊断牙龈炎症。本研究旨在开发一种结合口内扫描（IOS）和深度学习算法的自动化方法，以识别牙龈炎症的表面特征，并评估其准确性以及与临床指标的相关性。
该研究纳入了120例牙周炎患者的牙周探诊数据和口内扫描图像。使用深度学习模型GC-U-Net自动分割并识别牙龈炎症区域。通过Dice系数（Dice）、交并比（IoU）和像素准确率（PA）评估模型性能，并分析识别性能与牙周检查指标之间的相关性。
GC-U-Net模型显示出较高的识别准确率，Dice为77.8%，IoU为65.4%，PA为93.7%。该模型与龈沟出血指数（SBI；r = 0.836，P <.001）呈强正相关，与出血指数（BI；r = 0.618，P <.001）呈中等强度正相关，与探诊深度（PD；r = - 0.425，P <.001）呈负相关。
本研究成功开发了一种基于深度学习和IOS技术的牙龈炎症表面特征自动识别方法，为临床牙龈炎症检查提供了一种标准化、自动化的辅助工具，具有较高的准确性且与临床指标具有显著相关性。
该方法可减少牙龈炎症临床评估中的主观判断，提高识别的一致性和可靠性，并在临床诊断和治疗计划中发挥重要辅助作用。
