An Efficient Dual Branch Multi-Scale CNN With Reconstruction Learning and Explainable AI for Dental Caries Classification
Journal of Innovative Image Processing · 16.07.2026 · 8 (3) · 1022-1045
DOI: 10.36548/jiip.2026.3.014 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Innovative Image Processing, online erschienen am 16.07.2026. Verfasst von Shashikala K.S. und 1 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben.
Abstract (Originalsprache)
Preventive oral healthcare depends on the early detection of dental caries. Delicate visual differences between healthy tooth and early-stage lesions make automated disease diagnosis difficult. This work introduces a comparative analysis between a Single-Branch Convolutional Neural Network (SBCNN) and a proposed Dual-Branch Multi-Scale Convolutional Neural Network (DBMSCNN) for multiple class dental caries classification. It integrates Sharpness-Aware Minimization (SAM), multi-scale feature extraction, and self-supervised reconstruction learning to enhance feature robustness and generalization. We utilized a three-class dental caries dataset consisting of No Enamel Caries, Advanced Caries and Early Stage Caries. A total of 447 samples were used for testing in each fold. The proposed DBMSCNN framework achieved a mean classification accuracy of 92.43% across five-fold stratified cross-validation. The model demonstrated strong discriminative capability and a low false-positive rate, with a mean specificity of 96.14% and a mean precision of 92.59%. Reconstruction-based self-supervised pretraining improved anatomical feature learning, resulting in mean Dice and IoU scores of 0.8530 and 0.7577 respectively. Feature-level analysis using t-SNE is done for qualitative visualization of latent feature distributions for the proposed model, while Grad-CAM and saliency maps show meaningful and moderate localization of carious regions. The comparative results validate the effectiveness of multi-scale learning, reconstruction-based regularization, and SAM optimization in improving diagnostic reliability for dental caries classification.
Zitieren
APA
K.S., S., & S., U. (2026). An Efficient Dual Branch Multi-Scale CNN With Reconstruction Learning and Explainable AI for Dental Caries Classification. Journal of Innovative Image Processing, 8(3), 1022-1045. https://doi.org/10.36548/jiip.2026.3.014
Vancouver
K.S. S, S. U. An Efficient Dual Branch Multi-Scale CNN With Reconstruction Learning and Explainable AI for Dental Caries Classification. Journal of Innovative Image Processing. 2026;8(3):1022-1045. doi:10.36548/jiip.2026.3.014
BibTeX
@article{KS2026,
author = {K.S., Shashikala and S., Umamaheswaran},
title = {An Efficient Dual Branch Multi-Scale CNN With Reconstruction Learning and Explainable AI for Dental Caries Classification},
journal = {Journal of Innovative Image Processing},
year = {2026},
volume = {8},
number = {3},
pages = {1022--1045},
doi = {10.36548/jiip.2026.3.014}
}
EUVIMED