Role of Artificial Intelligence in Detecting Alveolar Bone Loss on Panoramic Radiography

Main Goals: To find, review, and summarize all the recent scientific studies that test how well AI can detect alveolar bone loss on panoramic radiographs. Based on the methodology and the preliminary findings from the included studies, this systematic review is expected to yield the following specific results and outputs: 1. A Critical Appraisal of Evidence Quality A key result will be the classification of the included studies by methodological rigor. The application of the CASP checklist is expected to output a clear distribution, likely showing that a portion of the studies are of high quality, while others may have moderate or low scores due to common limitations like small sample sizes, lack of external validation, or retrospective design. This appraisal is a crucial output that contextualizes the reported performance metrics. 2. A Qualitative Synthesis of Diagnostic Performance The primary output will be a consolidated narrative and tabular summary of the diagnostic accuracy of AI models for detecting periodontal bone loss. This will clearly present some key performance metrics, including Sensitivity, Specificity, Accuracy, Area Under the Curve (AUC), and F1-scores, reported across the eligible studies. The expectation is to confirm the trend of high performance, with metrics such as AUC values frequently exceeding 0.90, demonstrating the strong technical capability of these systems. 3. Identification of AI Model Evolution and Clinical Readiness The review is expected to output a clear analysis of how AI applications have evolved, from initial binary classification tasks to more advanced, clinically relevant functions like tooth-level segmentation and automated staging. This output will map the technological progression and highlight the most promising architectures (e.g., U-Net, YOLO variants) for specific diagnostic tasks. 4. A Defined Set of Research Gaps and Future Directions A major output will be a clearly articulated list of critical limitations and unanswered questions in the current evidence base. This is expected to include gaps such as the need for large-scale, prospective clinical trials; the importance of multi-center external validation to ensure generalizability; and the lack of studies evaluating AI integration into real-world clinical workflows and its impact on treatment decisions and patient outcomes. 5. Evidence-Based Conclusions and Recommendations The final, synthesized output will be a set of concise, evidence-based conclusions stating the current state of AI for this diagnostic purpose. Furthermore, the review will output practical recommendations for two key audiences: researchers (e.g., to adopt standardized reporting and focus on translational studies) and clinicians (e.g., that AI shows high potential as a decision support tool but requires further validation before routine independent use).