Applications of Artificial Intelligence in Bone Tissue Histology: a Scoping Review of Diagnostic and Structural...

Artificial intelligence (AI) has been applied to bone tissue histopathology for tasks such as bone tumor classification, bone microstructure segmentation, and identification of viable and necrotic tissue. However, methodological heterogeneity, variability in evaluation metrics, and limited comparative validation still hinder clinical application. Therefore, this scoping review systematically mapped the literature on AI applications in bone tissue histopathology. A comprehensive search was conducted across five databases for studies published up to June 2026. Eligible studies included diagnostic accuracy investigations using AI for bone disease classification, as well as human and animal studies applying AI to bone microstructural analysis. Among the 39 eligible studies, 35 focused on diagnostic accuracy and 4 on bone structure characterization. CNN-based models predominated, although hybrid, ensemble, generative, and Transformer-based approaches were also explored. Overall, 32 diagnostic studies reported high accuracy (≥ 80%), whereas only two showed moderate accuracy (60-79%). For osteosarcoma, most models showed high diagnostic performance, with the best-performing approaches reaching accuracies above 99%. Hybrid models combining deep feature extraction, feature selection, and machine learning classifiers, as well as attention-based and Transformer-based architectures were particularly effective. For other bone tumors, the Efficient and Enhanced Network (EENet) achieved accuracy, F1-score, sensitivity, specificity, and precision values above 99%. Studies focused on bone structure characterization primarily involved animal or fossil specimens and explored automated segmentation approaches; however, performance metrics were inconsistently reported across studies. In conclusion, AI-based models included in this review showed strong potential for accurate and automated bone histopathological analysis, supporting their future integration into diagnostic and tissue characterization workflows.