Progress in machine learning-assisted medical imaging for osteoarthritis and osteoporosis diagnosis: a narrative...

Osteoarthritis (OA) and osteoporosis (OP) are prevalent musculoskeletal disorders with substantial global health and economic burdens. Imaging is central to their diagnosis and monitoring, yet manual interpretation is vulnerable to inter-reader variability and workload-related fatigue. Artificial intelligence (AI), including machine learning (ML) and deep learning (DL), provides data-driven approaches to enhance the accuracy, efficiency, and objectivity of image interpretation. This review summarizes AI-assisted imaging advances for OA and OP over the past decade and discusses translational opportunities and challenges.

A literature search was conducted in Web of Science, PubMed, and Scopus for English-language studies published between January 2015 and August 2025. Search terms included osteoarthritis, osteoporosis, X-ray, computed tomography (CT), magnetic resonance imaging (MRI), machine learning, deep learning, detection, classification, and diagnosis. Titles and abstracts were screened, and selected full texts were reviewed to summarize advances and diagnostic performance across modalities.

Across X-ray, CT, and MRI, ML/DL approaches enable more objective quantification of OA- and OP-related abnormalities. Using public and cohort-based datasets, studies have evolved from radiomics-based ML pipelines to end-to-end DL frameworks for screening, classification, and grading. For OA, radiographs dominate Kellgren-Lawrence (KL) grading and large-scale screening, complemented by MRI for early tissue biomarkers and CT for quantifying subchondral bone remodeling. For OP, X-ray/CT captures bone texture and trabecular microarchitecture to support detection and classification, with MRI mainly used to assess marrow- and soft-tissue-related markers. Overall, DL typically improves automation and representation learning, while ML remains interpretable and competitive in smaller datasets. Emerging studies suggest that multimodal fusion and longitudinal modeling for progression assessment and prediction may further improve performance.

AI-assisted imaging is reshaping OA and OP assessment by enabling earlier detection and more objective longitudinal monitoring. However, clinical translation is hindered by limited interpretability of many DL models and substantial data heterogeneity. Future research should prioritize standardized multicenter datasets and explainable AI frameworks. Prospective clinical studies and rigorous external validation are needed to bridge the gap between research and practice and to advance personalized musculoskeletal care.