Deep Learning Photo Processing for Periodontitis Screening.

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Late detection of periodontitis leads to significant health issues.
• Oral image screening presents an accessible, nonclinical diagnostic approach.
• Deep learning algorithms show promise in medical diagnostics.

Purpose of the Study:

• To evaluate a deep learning model's accuracy in detecting periodontitis (Stage II-IV) using oral images.
• To assess the model's performance compared to clinical examinations and human experts.
• To explore the potential of AI-driven oral image analysis for periodontal screening.

Main Methods:

• A deep learning model (DLM) using ResNet50 was developed and tested on oral digital twins.
• The study involved 387 subjects for internal testing and 183 for external validation.
• Performance was measured using AUROC, sensitivity, and specificity, with class activation heatmaps for interpretability.

Main Results:

• The DLM achieved high accuracy (AUROC=0.93) in both internal and external datasets for periodontitis detection.
• Regions of interest identified by the DLM showed high consistency with periodontist assessments.
• The DLM demonstrated superior sensitivity and specificity compared to clinicians.

Conclusions:

• Deep learning model processing of oral images shows significant potential for periodontal health screening.
• AI effectively identifies key diagnostic features in oral images, sometimes surpassing human clinical observation.
• Further validation across diverse populations is necessary for global implementation as a screening tool.