Back to search

University of British Columbia

Artificial intelligence-driven escalation in caries diagnosis and treatment decisions

Abstract

dc:description

Objectives: Artificial intelligence (AI) is increasingly used to support radiographic caries detection; however, its influence on clinical decision-making remains unclear. This study evaluated whether AI assistance influenced pediatric dentists’ lesion identification, radiographic severity classification, treatment recommendations, and diagnostic reliability when assessing caries in primary teeth. Methods: Thirty pediatric bitewing radiographs were evaluated independently by 10 board-certified pediatric dentists during three assessment sessions separated by two-week washout periods. During Attempts 1 and 2, radiographs were assessed without AI assistance. During Attempt 3, the same radiographs were evaluated with Diagnocat™ AI assistance. At each session, lesion severity was recorded using a modified ICDAS system, and treatment recommendations were categorized as no treatment, non-invasive treatment, or invasive treatment. Intra- and inter-rater reliability were assessed using Cohen’s κ, Fleiss’ κ, and Gwet’s AC1. AI-associated changes in lesion severity and treatment recommendations were evaluated using lowest prior severity, average prior severity, and highest prior severity comparison frameworks. Results: Pediatric dentists demonstrated high intra-rater reliability for lesion severity classification (mean κ = 0.75) and treatment category assignment (mean κ = 0.72). Inter-rater agreement was moderate to very good at baseline (lesion severity: Fleiss’ κ = 0.46, AC1 = 0.87; treatment category: Fleiss’ κ = 0.58, AC1 = 0.90). AI assistance increased lesion identification, with raters identifying an average of 86.3 lesions compared with 66.5 at baseline. Using the average prior severity framework, the proportion of surfaces assigned a more severe lesion classification increased from 9.7% to 26.1% following AI exposure. Treatment escalation occurred in 19.2% of AI-flagged assessments, while 7.8% were de-escalated. The most common escalation was from no treatment to invasive treatment. Inter-rater agreement did not meaningfully improve following AI assistance. Conclusions: Diagnocat™ AI assistance influenced radiographic interpretation and treatment recommendations for carious lesions in primary teeth. AI exposure was associated with increased lesion identification, upward shifts in severity classification, and more frequent treatment escalation, but did not improve agreement among experienced pediatric dentists. These findings highlight the importance of evaluating AI systems not only by diagnostic performance, but also by their effects on clinical judgment and treatment decision-making.

Degree

thesis:*
Name thesis:degree_name
Master of Science - MSc
Level thesis:degree_level
master's
Discipline thesis:degree_discipline
Craniofacial Science
Grantor dc:publisher
University of British Columbia
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Mathew, Teris
Contributors dc:contributor
  • Vora, Siddharth

Rights

dc:rights
Statement dc:rights
  • Attribution-NonCommercial-NoDerivatives 4.0 International
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/95114

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Mathew, Teris. Artificial intelligence-driven escalation in caries diagnosis and treatment decisions. master's thesis, University of British Columbia, 2026. http://hdl.handle.net/2429/95114