Quantum-inspired fused explainable deep learning framework for early enamel caries classification in intraoral photographs.

The visual detection of early enamel caries in intraoral photographs is challenging due to subtle lesion appearance and diagnostic subjectivity. This study aimed to develop and validate an explainable, quantum-simulated deep learning framework for the automated classification of enamel caries severity from intraoral photographs. A hybrid framework was developed, integrating deep features from two custom models: a lightweight DentXCaries Convolutional Neural Network (CNN) and a modified attention-based ResNet50. A novel quantum-simulated entanglement fusion strategy combined these features, which were subsequently classified by several machine learning algorithms. The models were trained and evaluated on the public "Caries-Spectra" dataset (2,000 images across three classes: Sound Enamel, Early-Stage Enamel Caries, and Advanced Enamel Caries). Explainable AI (Grad-CAM) provided visual interpretability. Performance was assessed via accuracy, precision, recall, F1-score, and ROC-AUC. The fused framework achieved peak performance with Neural Network and Random Forest tree classifiers, attaining 99.33% accuracy and F1-score. The fusion mechanism significantly reduced inter-class confusion compared to individual models. The proposed framework demonstrates superior accuracy and crucial visual explainability for enamel caries classification. It represents a step towards a reliable, transparent AI-assisted diagnostic tool, although a high-rigour reference standard and future multicentre validation are required to confirm clinical generalizability.