Related Experiment Video
Updated: May 9, 2026
Reliability of Artificial Intelligence-Based Cone Beam Computed Tomography Integration with Digital Dental Images
Published on: February 23, 2024
Knowledge and Awareness of Performance Metrics in Dental Image Analysis With Artificial Intelligence: A
Rishi Ramani1, Bree Jones1, Akhilanand Chaurasia2
1Dental Data and Artificial Intelligence (D2AI) research group, Melbourne Dental School, University of Melbourne, Victoria, Australia; Topic Group Oral Health, ITU/WHO/WIPO Global Initiative on Artificial Intelligence for Health, Geneva, Switzerland.
Introduction And Aim:
Appropriate performance metrics are essential for evaluating machine learning (ML) algorithm performance in dental image analysis. A broad variety of performance metrics are used in dental artificial intelligence (AI) research which could cause confusion regarding model performance. This study aimed to assess the knowledge, awareness, and self-reported confidence of oral health researchers and clinicians regarding AI performance metrics.
Methods:
This cross-sectional, questionnaire-based survey initially recruited participants (n = 100) from the ITU/WHO/WIPO Global Initiative on AI for Health Topic Group - Oral Health and affiliated professional global networks. An online self-administered questionnaire evaluated theoretical knowledge and applied knowledge. Participants also reported basic demographics and their confidence in using different metrics.
Results:
Overall performance was poor with participants achieving a mean score of 17.1% (SD: 15.2%, median: 16.7%, IQR: 16.7%) for fully correct responses. Responses to theoretical questions were rarely fully correct (range: 1%-15%; median: 5.5%). Applied scenario performance ranged from 17% to 37% correct responses (median: 29%). Participants demonstrated greater knowledge of common metrics (accuracy, specificity) but struggled with advanced measures (Intersection over Union & Dice Similarity Coefficient). Common errors included terminology confusion, familiarity bias towards accuracy, and task-metric misalignment.
Conclusion:
Substantial gaps exist in knowledge and awareness of both theory and practical applications of AI performance metrics amongst dental professionals and researchers. Frequent misapplication of commonly known metrics suggests overgeneralisation without consideration of task-specific requirements.
Clinical Relevance:
Inadequate knowledge or awareness of metric selection and interpretation may lead to the deployment of suboptimal algorithms, which can compromise diagnostic accuracy and patient outcomes. Enhanced understanding of performance metrics is critical for ensuring reliable, fair, safe, and effective AI systems in clinical dentistry.
