Bo Wang, Sijin Tang, Pingjuan Tan, Baohong Mi, Weiheng Chen, Chunzhu Gong, Renxuan Yang, Qiushi Wei
This study provides a population-specific evidence map of AI applications in osteoporosis screening and fracture risk prediction among aging populations. The findings suggest that this field is moving from conventional risk assessment toward AI-assisted risk stratification, imaging-based screening, and clinically oriented decision support. Future research should prioritize aging-specific validation, multimodal integration, interpretability, fairness, external validation, and real-world evaluation before AI-based fracture risk tools can be routinely incorporated into endocrine and bone health practice.
OBJECTIVE: To map the development, knowledge structure, and clinical translation of artificial intelligence applications in osteoporosis screening and fracture risk prediction among aging populations, with a focus on risk stratification, prediction-oriented modeling, and endocrine bone health management.
METHODS: A bibliometric evidence-mapping study was conducted using publications retrieved from the Web of Science Core Collection and PubMed. VOSviewer and R/bibliometrix were used to analyze publication and citation trends, collaboration networks, reference co-citation knowledge bases, author keyword structures and bursts, keyword-knowledge-base consistency, and interdisciplinary knowledge-flow patterns at the discipline and journal levels.
RESULTS: A total of 371 publications were included. Among 10,934 cited references, 74 met the co-citation threshold and formed four major knowledge-base clusters related to early fracture risk assessment and epidemiological evidence, fracture risk model construction, guideline-based osteoporosis management, and machine-learning expansion. Author keyword analysis included 70 keywords and identified four application-oriented themes: aging-related osteoporosis and fracture-risk screening frameworks, machine-learning risk stratification in older fracture-related populations, imaging and radiomics markers of structural fragility, and deep-learning applications for prediction-oriented detection and triage. Cross-mapping between keyword themes and co-citation clusters showed that most current hotspots were supported by established fracture-risk, BMD, FRAX, guideline, or machine-learning knowledge bases, whereas deep-learning-related themes remained more distributed across clinical and methodological clusters. Annual publications increased from 15 in 2019 to 86 in 2025. China and the United States were the leading contributors, with 126 and 90 publications, respectively. Knowledge-flow analyses showed increasing input from Radiology & Imaging, Computer & AI, and Engineering, with strong convergence toward Clinical Medicine and continued aggregation in core osteoporosis and bone metabolism journals.
CONCLUSION: This study provides a population-specific evidence map of AI applications in osteoporosis screening and fracture risk prediction among aging populations. The findings suggest that this field is moving from conventional risk assessment toward AI-assisted risk stratification, imaging-based screening, and clinically oriented decision support. Future research should prioritize aging-specific validation, multimodal integration, interpretability, fairness, external validation, and real-world evaluation before AI-based fracture risk tools can be routinely incorporated into endocrine and bone health practice.