Osteoporosis and its consequent fragility fractures represent a major public health challenge in aging populations worldwide. Although dual-energy X-ray absorptiometry (DXA) remains the diagnostic gold standard, its utility for opportunistic screening is limited by equipment accessibility, low screening coverage, and inability to assess bone microarchitecture. Multislice computed tomography (MSCT), routinely performed for various clinical indications, offers a quantitative measure, trabecular bone attenuation in Hounsfield Units (HU), that can be opportunistically leveraged for osteoporosis evaluation. Recent advances in natural language processing (NLP) and large language models (LLMs) enable automated extraction of unstructured clinical data from electronic health records and integration of multimodal information. This review systematically summarizes the evidence base for MSCT-derived HU values in osteoporosis diagnosis, explores the application of NLP for extracting key variables from radiology reports and clinical notes, and discusses the potential of LLMs for multimodal data fusion and predictive modeling. We propose a novel framework combining HU measurements, NLP-extracted clinical features, and LLM-driven analysis to construct a Nomogram model for opportunistic osteoporosis screening and fracture risk prediction. This approach may expand the screened population beyond postmenopausal women to include men and individuals with chronic diseases, ultimately enabling opportunistic, individualized fracture risk assessment integrated into routine clinical workflows.