Scholarlink

Aim <p>Evaluating dental plaque is a fundamental task for periodontal health care, but it is subjective, time-consuming, and cumbersome. Therefore, this study aimed to develop and validate a web-based deep learning application capable of objectively quantifying tooth plaque in fluorescence images by calculating the plaque–tooth area ratio.</p> Methods <p>A total of 2,490 fluorescence image of the Lingual surfaces of mandibular anterior teeth from 498 participants were used to train and test a YOLO v11 model with optimized hyperparameters for detecting tooth and plaque. After the model was developed, 30 additional participants were recruited, and their fluorescence image were evaluated for clinical validation by calculating the plaque–tooth area ratio. A web application was developed for clinical use, and validation compared AI predictions with clinical ratings via intraclass correlation coefficient analysis.</p> Results <p>The deep learning model accurately detected and segmented teeth and dental plaque, with F1 scores of 0.81 for both tasks. Mean average precision at an intersection over union threshold of 0.50 (mAP50) was 0.83 and 0.84, respectively. The model achieved average precision scores of 0.969 for teeth and 0.706 for plaque, with an overall mAP50 of 0.838. Clinical validation showed strong agreement with expert assessments (ICC = 0.947) and a 97.9% reduction in evaluation time.</p> Conclusions <p>The web application demonstrated high accuracy in identifying and quantifying tooth plaque objectively in fluorescence images, supporting its potential as an oral hygiene assessment tool for the prevention of periodontal disease.</p> Clinical significance <p>This deep learning-based web application offers an effective, and objectively scalable solution for dental plaque quantification, enhancing diagnostic precision and supporting timely periodontal intervention. Its integration into clinical workflows might improve treatment planning, promote patient compliance, and enable standardised monitoring of oral hygiene status, ultimately contributing to improved periodontal outcomes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Clinical validation of a deep learning based application for quantitative assessment of dental plaque in fluorescence imaging

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Hang-Nga Mai,
  • Sohee Kang,
  • Hyeonjeong Go,
  • Youn-Hee Choi,
  • Eun Young Park,
  • Eun-Kyong Kim

摘要

Aim

Evaluating dental plaque is a fundamental task for periodontal health care, but it is subjective, time-consuming, and cumbersome. Therefore, this study aimed to develop and validate a web-based deep learning application capable of objectively quantifying tooth plaque in fluorescence images by calculating the plaque–tooth area ratio.

Methods

A total of 2,490 fluorescence image of the Lingual surfaces of mandibular anterior teeth from 498 participants were used to train and test a YOLO v11 model with optimized hyperparameters for detecting tooth and plaque. After the model was developed, 30 additional participants were recruited, and their fluorescence image were evaluated for clinical validation by calculating the plaque–tooth area ratio. A web application was developed for clinical use, and validation compared AI predictions with clinical ratings via intraclass correlation coefficient analysis.

Results

The deep learning model accurately detected and segmented teeth and dental plaque, with F1 scores of 0.81 for both tasks. Mean average precision at an intersection over union threshold of 0.50 (mAP50) was 0.83 and 0.84, respectively. The model achieved average precision scores of 0.969 for teeth and 0.706 for plaque, with an overall mAP50 of 0.838. Clinical validation showed strong agreement with expert assessments (ICC = 0.947) and a 97.9% reduction in evaluation time.

Conclusions

The web application demonstrated high accuracy in identifying and quantifying tooth plaque objectively in fluorescence images, supporting its potential as an oral hygiene assessment tool for the prevention of periodontal disease.

Clinical significance

This deep learning-based web application offers an effective, and objectively scalable solution for dental plaque quantification, enhancing diagnostic precision and supporting timely periodontal intervention. Its integration into clinical workflows might improve treatment planning, promote patient compliance, and enable standardised monitoring of oral hygiene status, ultimately contributing to improved periodontal outcomes.