OR05 - Evaluation of Periapical Lesion Detection by Anatomical Region Using a No-Code Artificial Intelligence System

Wednesday, April 2, 2025

10:00 AM – 10:15 AM EST

Location: Hynes Convention Center, Floor 3, Room 310

Presenter(s)

Kevin W. Yu, D.M.D.

Resident
Marquette University
Milwaukee, Wisconsin

Disclosure information not submitted.

Co-Author(s)

ZB

Zaid Badr, D.D.S., M.S.

Clinical Assistant Professor & Director of the Technological Innovation Center
Marquette University
Milwaukee, Wisconsin

Disclosure information not submitted.
SS

Stephanie J. Sidow, D.D.S.

Clinical Associate Professor & Director, Postgraduate Program in Endodontics
Marquette University
Milwaukee, Wisconsin

Disclosure information not submitted.
MH

Manal H. Hamdan, D.D.S., M.S.

Assistant Professor & Director, Predoctoral Program in Oral & Maxillofacial Radiology
Marquette University
Milwaukee, Wisconsin

Disclosure information not submitted.

Objective: Radiographic evaluation of periapical lesions is subjective and may be hindered by anatomical structures present and radiographic techniques used. The purpose of this study was to evaluate the impact of anatomical stratification on periapical lesion detection by a no-code artificial intelligence (AI) model based on anatomical region.

Methods: 312 intraoral periapical radiographs were retrospectively analyzed. A no-code computer vision model was trained on annotated periapical radiographs using the LandingLens (Landing AI LLC, Palo Alto, CA) platform. Images were stratified by anatomical region and sensitivity, specificity, accuracy, precision, and F1 score were calculated. Statistical analysis was completed using chi-squared and pairwise chi-squared tests (α=0.05).

Results: The model’s sensitivity was significantly higher in the maxilla (0.931) than the mandible (0.837; p=0.046), while specificity was higher in the mandible (0.947; p=0.035). No significant difference was observed between sensitivity and specificity in the anterior and posterior teeth (p>0.05). Sensitivity was highest in the anterior maxilla (0.977), while specificity was significantly higher in the anterior mandible (0.974) than other regions (p < 0.05). No significant difference was observed with other metrics (p>0.05).

Conclusion: Anatomical location influences AI detection of periapical lesions, particularly in sensitivity for the maxilla and specificity for the mandible. These findings suggest that anatomical stratification can affect AI model performance.