Background:

Periodontal bone loss is a key indicator of periodontitis severity and progression. Reliable detection is essential for timely intervention, yet manual assessment of cone-beam computed tomography (CBCT) scans is time-consuming and prone to interobserver variability. Artificial intelligence (AI), particularly deep learning, has shown promise in automating image interpretation; however, the “black-box” nature of many AI systems limits clinical trust. Explainable AI (XAI) techniques may enhance transparency by providing visual and quantitative insights into model reasoning. This study evaluates an XAI-enabled framework for automated detection of periodontal bone loss on CBCT scans and examines clinicians’ perceptions of its usability and interpretability.

Methodology:

A convolutional neural network (CNN) was trained on 2,500 annotated CBCT scans from 1,000 patients, labeled by two expert periodontists. XAI tools, including Gradient-weighted Class Activation Mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP), were applied to highlight image regions contributing to model predictions. Model performance was assessed using sensitivity, specificity, F1-score, and AUC. To evaluate interpretability, 32 clinicians (18 periodontists, 14 radiologists) reviewed 120 AI-assisted cases. Perceptions were measured using a validated 5-point Likert-scale questionnaire adapted from the System Usability Scale and the Trust in Automation framework, complemented by two open-ended questions. Responses were analyzed using descriptive statistics and thematic coding.

Results:

The CNN achieved 91% sensitivity, 88% specificity, an F1-score of 0.895, and an AUC of 0.94. Grad-CAM and SHAP visualizations aligned with expert annotations in 87% of cases. Clinicians reported improved diagnostic confidence (mean 4.3/5) and perceived transparency (4.2/5), while 78% indicated that XAI outputs facilitated faster case review without increasing cognitive load.

Conclusions:

The proposed XAI-based system accurately detects periodontal bone loss on CBCT scans and provides interpretable visual feedback that supports clinician trust and decision-making. Future work will include prospective validation and workflow integration in routine periodontal diagnostics.