Truace
Subscribe

AI-Based Clinical Decision Support Systems for Secondary Caries Detection and Staging on Bitewings: A Multi-Algorithm Comparison

Introduction Radiographic detection of caries lesions adjacent to restorations is challenging because of the limitations of two-dimensional imaging and difficulties distinguishing true lesions from restorative or anatomical radiolucencies. Artificial intelligence (AI)-based clinical decision support systems (CDSSs) have been introduced to assist radiographic interpretation; however, different AI tools may yield variable diagnostic outputs, and their comparative performance remains unclear. Objective To compare t…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1673· cite & verify
AI-Based Clinical Decision Support Systems for Secondary Caries Detection and Staging on Bitewings: A Multi-Algorithm Comparison
"22-0008-010 Camp Lejeune" by NavyMedicine is marked with Public Domain Mark 1.0. To view the terms, visit https://creativecommons.org/publicdomain/mark/1.0/.

In brief

Researchers tested five commercial and three experimental AI systems on 200 anonymized bitewing radiographs containing 885 restored tooth surfaces. The tools were compared against a consensus reference standard for finding and staging caries next to restorations.

The tools performed alike and conservatively. They rarely flagged healthy surfaces but missed many true lesions, which limits their use as a standalone check. The authors suggest more diverse datasets, better reference standards, and greater explainability could improve reliability, though this was not tested here.

Main points

  1. The cross-sectional study used 200 bitewings covering 885 restored tooth surfaces.
  2. Specificity ranged from 0.957 to 0.986 across all systems.
  3. Sensitivity ranged only from 0.327 to 0.487, with most misses in dentin-stage lesions.
  4. Accuracy of 0.882 to 0.917 did not differ significantly among systems (p ≥ 0.05).

The problem

Across eight commercial and experimental AI systems, sensitivity for secondary caries on bitewings was only moderate, so many lesions, especially dentin-stage ones, went undetected.

The rundown

The systems evaluated were Second Opinion, CranioCatch, Diagnocat, DIO Inteligência and Align X-ray Insights, plus three experimental models built on Mask R-CNN and Mask DINO architectures. Each restored surface was classified as non-carious, secondary caries, or indeterminate, and lesions were staged as enamel or dentin.

Performance was measured with sensitivity, specificity, accuracy, and positive and negative predictive values with 95% confidence intervals. Generalized estimating equations adjusted for clustering of surfaces within images.

Errors between non-carious and carious surfaces were often linked to radiographic overlap, restoration-related radiolucencies, and cervical artifacts. The authors conclude that architecture and commercial status made little difference to diagnostic capability.

Sources

  1. Peer-reviewedCaries Research2026-10-09

The debate