This invention discloses a method and
system for longitudinal
bone mineral density prediction and high-risk window identification, belonging to the field of intelligent
bone health assessment technology. Based on a large amount of localized
population-based
bone mineral density data assets, this invention preprocesses the data to construct a regionally adaptive dual-graph LSTM network for
temporal bone mineral density prediction. It quantifies prediction uncertainty and generates confidence intervals using a Bayesian method, employs the PELT
algorithm to detect
bone mineral density evolution change points, and identifies the critical point of accelerated decline. By integrating prediction results, confidence intervals, and change point locations, a three-dimensional high-risk window is delineated, and personalized intervention plans are output, supporting multi-platform deployment. This invention solves the problems of inapplicability to traditional assessment standards, difficulty in capturing nonlinear changes, and insufficient prediction reliability, significantly improving the accuracy of bone
mineral density prediction and the timeliness of clinical intervention. It is suitable for
osteoporosis risk screening and primary healthcare management.