Automated Localization and Identification of Recurrent (Secondary) Caries Around Dental Restorations Using a Hybrid CNN–Vision Transformer Framework with Grad-CAM Explainability
Computer and Decision Making: An International Journal · 08.11.2026 · 3 (8S) · 289-308
DOI: 10.1922/comdem.v3i8s.19157 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 08.11.2026. Verfasst von Dr. Anushree Raj und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Recurrent (secondary) caries around dental restorations can be difficult to diagnose, as restorative materials, margins and overlapping anatomic structures may mimic or mask carious lesions in the radiographic image. While AI has shown promise in the interpretation of dental images, traditional classification methods might not give enough information about the location of the lesion with respect to the restorations. This article outlines a restoration-aware and explainable deep-learning (DL) framework designed to assist in clinical evaluation of recurrent caries. The proposed architecture consists of radiographic preprocessing, automatic localization of the restoration using a one-stage object detector, peri-restoration lesion segmentation using attention-gated U-Net, and hybrid CNN–ViT classification. A module called ‘Grad-CAM' is added to give spatial explanations of intended classification outputs. The methodological specification includes 17 mathematical formulations that govern the preprocessing, localization, segmentation, classification, optimization, explainability and evaluation. A literature-based comparative evaluation of the detection, segmentation, classification, calibration, ablation, and statistical evaluation is followed by a proposed validation protocol. The integrated framework has not been trained end-to-end or clinically validated and no original diagnostic performance results are claimed. It makes a methodological contribution to the integration of image analysis and interpretable classification in the proposed clinical decision-support workflow for restoration. Empirical implementation is needed to evaluate diagnostic accuracy, explanation reliability and clinical utility.
Zitieren
APA
Raj, D. A., K, L. C., O, P. M., Shetty, K. A., & K, S. M. (2026). Automated Localization and Identification of Recurrent (Secondary) Caries Around Dental Restorations Using a Hybrid CNN–Vision Transformer Framework with Grad-CAM Explainability. Computer and Decision Making: An International Journal, 3(8S), 289-308. https://doi.org/10.1922/comdem.v3i8s.19157
Vancouver
Raj DA, K LC, O PM, Shetty KA, K SM. Automated Localization and Identification of Recurrent (Secondary) Caries Around Dental Restorations Using a Hybrid CNN–Vision Transformer Framework with Grad-CAM Explainability. Computer and Decision Making: An International Journal. 2026;3(8S):289-308. doi:10.1922/comdem.v3i8s.19157
BibTeX
@article{Raj2026,
author = {Raj, Dr. Anushree and K, Lokesh C and O, Pallavi M and Shetty, K Annapoorneshwari and K, Subin M},
title = {Automated Localization and Identification of Recurrent (Secondary) Caries Around Dental Restorations Using a Hybrid CNN–Vision Transformer Framework with Grad-CAM Explainability},
journal = {Computer and Decision Making: An International Journal},
year = {2026},
volume = {3},
number = {8S},
pages = {289--308},
doi = {10.1922/comdem.v3i8s.19157}
}
EUVIMED