Shu-Ting Fan1
Li-Na Hao
Ren-Chao Dong- 1School of Information Science and Electrical Engineering (School of Artificial Intelligence), Shandong Jiaotong University, Jinan, China
- 2Department of Pharmacy, The Fourth People's Hospital of Jinan, Jinan, China
- 3Department of Traditional Chinese Medicine, Yantai Zhifu District Maternal and Child Health Hospital, Yantai, China
- 4Department of Pharmacy, Children’s Hospital Affiliated to Shandong University (Jinan Children’s Hospital), Jinan, China
Osteoporosis is a systemic skeletal disease defined by decreased bone mass, deteriorated bone microarchitecture, and increased fracture susceptibility, all of which substantially increase the risks of functional disability and overall mortality. Given its complex pathophysiology, chronic progression, and frequently asymptomatic clinical presentation, the diagnosis and treatment of osteoporosis present multiple challenges. Artificial intelligence (AI) technologies integrate radiomics and multi-omics data to develop intelligent frameworks that map disease progression, identify molecular biomarkers, and stratify individual risk, offering new strategies for molecular-level support in precision diagnosis, preventive monitoring, and personalized intervention. Notably, AI-driven models for drug target prediction and therapeutic response evaluation have also been developed, offering mechanism-based strategies for novel drug discovery. This study outlines how AI advances osteoporosis management by applying technological breakthroughs in foundational and clinical research across key areas, including optimizing preventive monitoring, developing precision diagnostics, supporting personalized treatment decisions, and enabling molecular-targeted drug discovery, thereby offering a strong theoretical basis and practical pathways within precision and personalized medicine.
Graphical Abstract.
1 Introduction
Osteoporosis (OP) is a systemic metabolic bone disease marked by low bone mineral density (BMD), deteriorated bone micro-structure, and increased fragility, which significantly raises the risk of fractures (1–4). With the population, osteoporosis has become a major public health issue. According to the International Osteoporosis Foundation (IOF), about 18.3% of people currently suffer from the condition. Among those over 50 years, one-third of women and one-fifth of men experience osteoporotic fractures (OPF), with a new fracture occurring every 3 s, and 50% of osteoporotic fracture patients will experience a recurrence (1, 5–7). Osteoporosis is caused by multiple factors, including estrogen deficiency, aging, genetics, calcium loss, long-term glucocorticoid use, and smoking or alcohol abuse (8, 9). Currently, diagnosis primarily relies on imaging modalities such as X-ray, computed tomography (CT), and magnetic resonance imaging (MRI) (10, 11). However, diagnostic efficacy is limited by image noise and individual variability. Treatments include bone resorption inhibitors and bone formation promoters, which have drawbacks such as side effects, limited long-term effectiveness, and the ability only to slow disease progression, preventing fractures but not reversing the condition (12, 13). The utilization of blood biomarkers, gene expression analysis, and predictive modeling can significantly enhance the diagnosis of osteoporosis, the monitoring of treatment efficacy, and the investigation of novel mechanisms and targeted therapies (14, 15). Given that early-stage osteoporosis is often asymptomatic, it is essential to integrate imaging, biochemical markers, genetic data, and clinical parameters into a comprehensive, multidimensional, and multimodal assessment framework.
Artificial intelligence (AI) was formally proposed by John McCarthy in 1956 (16). Recent advancements in technologies such as cloud computing and big data have enabled significant progress in osteoporosis research (17). The selection of AI algorithms for osteoporosis research needs to be based on data scale, feature types, interpretability requirements, and computational resources. Depending on the application scenarios, osteoporosis research primarily involves the fields of data science and image processing within AI (17–19). The identification of biomarkers through feature engineering and the application of algorithms such as support vector machines (SVM), gradient boosting machines (GBM), random forests (RF), and least absolute shrinkage and selection operator (LASSO) represents an effective approach for both diagnosing osteoporosis and advancing drug development for the condition. Further, deep neural network (DNN) technology, based on deep learning algorithms, enables precise classification and detection of medical images, as well as the prediction of skeletal diseases through hierarchical feature extraction. Notably, AI models integrating multimodal data can assess the progression of knee joint diseases, provide comprehensive technical support for the prevention of osteoporotic fractures and personalized diagnosis and treatment of bone metabolic disorders, and contribute to the advancement of osteoporosis drug development (20–22).
This paper synthesizes key advancements in AI applied to bone image recognition, pathological mechanism analysis, clinical decision support, and drug development. It presents a comprehensive review of AI research progress in the prediction, diagnosis, treatment, and novel drug development for osteoporosis. The analysis focuses on efficient target identification and efficacy assessment in drug discovery, disease risk modeling for preventive strategies, breakthroughs in medical imaging recognition for diagnostic applications, and the optimization of individualized intervention plans. Moreover, this paper highlights the potential of artificial intelligence to enable precise management of bone metabolic disorders, support the development of innovative prevention strategies, and deliver intelligent enhancements to existing healthcare systems.
2 Application of AI in the foundational research of osteoporosis
AI has advanced osteoporosis foundational research through multi-omics data integration and enhanced machine learning algorithms, achieving efficient target screening and drug development (23–25). Training models with multidimensional datasets has enabled the identification of crucial diagnostic markers, including serum proteins, metabolic markers, and cell death regulators, revealing core mechanisms within the metabolism-immune-cell death network. AI utilizes molecular biomarkers for data analytics, further accelerating drug development via target identification and clinical trial compound prediction.
2.1 Novel diagnostic molecular markers of osteoporosis
At present, AI can efficiently identify molecular markers by integrating multi-omics data, thereby significantly improving the efficiency of target identification (26). The identification of new molecular markers for osteoporosis relies on advanced computational methodologies. AI algorithms can be classified into supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning (27, 28). According to the multi-dimensional data characteristics, the identification of new diagnostic molecular markers mostly adopts supervised learning algorithms. This section focuses on the multi-dimensional research innovations and technological advancements in molecular marker screening with AI (Figure 1).
Figure 1. AI advances understanding of osteoporosis biomarkers and drug discovery. AI has significantly advanced foundational research in osteoporosis, elucidating disease mechanisms and enhancing drug development processes. By integrating multi-omics data with machine learning algorithms such as random forest (RF), support vector machine (SVM), gradient boosting machine (GBM), and least absolute shrinkage and selection operator (LASSO), AI facilitates the identification of crucial molecular biomarkers and target genes implicated in osteoporosis. By optimizing the exploration of pharmacological mechanisms by deep learning, AI contributes to the development of novel therapeutics and refines drug screening processes, thereby driving innovation in therapeutic strategies.
2.1.1 AI identifies primary osteoporosis biomarkers
AI significantly enhances the precision of molecular biomarker discovery by integrating multidimensional data from genomics, proteomics, and metabolomics, addressing the dimensional limitations inherent in traditional analyses (Table 1). The systematic analysis of the target functional network indicates that osteoporosis involves multiple pathways related to immunity, cell death, and metabolism.
Table 1. AI enhances the comprehension of osteoporosis molecular biomarkers.
The discovery of immune biomarkers for osteoporosis utilized supervised learning algorithms, including SVM, GBM, and linear regression methods. The SVM algorithm classifies the gene transcriptional data and identifies characteristic genes by finding the optimal separation plane. By integrating SVM algorithms and the similarity network fusion (SNF) method, ten characteristic genes were identified, such as GPR31, GATM, DDB2, ARMCX1, COQ9, and CD9, highlighting the significant role of metabolic heterogeneity in osteoporosis (29). Subsequent studies further utilized the SVM-REF algorithm combined with the RFE algorithm to eliminate unimportant features and combined LASSO algorithm to identify CCR5, IAPP, IFNA4, IGHV3-73, and PTGER1 of osteoporosis (30). Additionally, immune infiltration analysis confirmed that CCR5-mediated inflammation and bone imbalance are key drivers of osteoporosis progression. Moreover, the algorithm fusion of RF, SVM-RFE, and LASSO regression has achieved significant progress in the immune microenvironment (31). Immune-related differentially expressed genes were identified in peripheral blood mononuclear cells from PMOP patients, including JUN, HMOX1, CYSLTR2, TNFSF8, IL1R2, and SSTR5. Single-cell sequencing revealed that JUN and HMOX1 are differentially expressed in M1/M2 macrophages, highlighting the role of M2 polarization and its interactions with CD8 + T cells, Tregs, and fibroblasts in maintaining bone homeostasis.
The GBM algorithm aggregates heterogeneous data modalities into unified representations through iterative residual minimization guided by gradient descent optimization. Integrated analysis employing Light Gradient Boosting Machine (Light GBM), proteomics, and MR models identified 22 BMD-associated proteins, including A2MG, APOA1, PHLD, AOPL1, SHBG, and SAMP, with significant correlations at lumbar spine and femoral neck (32). MR analysis further substantiated the genetic causal effects of BCHE and APOL1 on bone mineral metrics, particularly revealing APOL1’s positive impact at the femoral neck, elucidating mechanistic links between serum proteome dynamics and bone mass. Similarly, the integration of Light GBM, MR, Cox proportional hazards regression, and plasma proteomics analysis enabled the discovery of 134 osteoporosis-associated plasma protein biomarkers through analysis of 42,325 prospectively enrolled UK Biobank (UKB) (33). ADIPOQ, CKB, FSHB and SOST were identified as high-potential predictors, introducing novel biomarkers for trimester screening of osteoporosis.
The discovery of osteoporosis cell death biomarkers predominantly utilized SVM and RF algorithms. As reported, Machine learning analysis with the SVM-RFE/LASSO model identified apoptosis-related biomarkers, including DAP3, BIK, and ACAA2 (34). Mitochondrial oxidative phosphorylation pathways, regulated by transcription factors such as SETDB1 and ZNF281, play a crucial role in modulating both programmed cell death and the remodeling of the immune microenvironment. The combined SVM-RFE and RF algorithms model established the connection between iron homeostasis regulation and bone matrix synthesis damage. The algorithm model identified the FOXO3/DDIT3 regulatory axis and revealed that MAPK signaling and neutrophil activation amplify bone resorption, offering new insights into ROS accumulation caused by estrogen deficiency in postmenopausal osteoporosis (PMOP) (35). In addition, SVM and RF models have been applied to identify key ferroptosis genes, including HMOX1, HAMP, LPIN1, MAP3K5, and FLT3 in PMOP (36). Moreover, the proposed model demonstrated robust predictive performance based on COL1A1 expression, linking iron homeostasis to compromised bone matrix synthesis.
To handle high-dimensional data, the researchers also combined RF, Boruta, with the LASSO algorithm, and studied the interaction network between autophagy and cell apoptosis (37). Cross-validation of datasets (GSE56814/GSE56815) identified key autophagy regulatory factors, including PDPK1, MAP1LC3B, and ZFP36. This investigation has elucidated transcription factor-mediated immune response and cell death signaling pathways, signifying a paradigm shift from single-pathway examinations to multi-modal regulatory network analyses in programmed cell death (PCD) research. Investigations integrating differential gene screening, machine learning, and molecular network construction have established a framework linking biomarker discovery to pathological mechanisms. These findings highlight the cascading roles of programmed cell death, oxidative stress, ferroptosis, and autophagy in bone remodeling and improve marker screening accuracy through optimized algorithms.
The integration of multi-omics with machine learning has overcome traditional limitations in apoptosis pathways research, revealing novel mechanistic insights into cell death regulation. The dynamic balance between programmed cell death and oxidative stress networks was clarified in bone homeostasis. Collectively, AI substantially enhances the efficiency and reliability of biomarkers discovery through feature screening processes and subtype stratification strategies.
2.1.2 AI identifies secondary osteoporosis biomarkers
Recent cross-system studies of osteoporosis pathogenesis have made significant breakthroughs through the integration of AI approaches and multi-omics (38, 39). AI-driven systems biology overcomes limitations of single-omics studies and shifts from isolated pathways to integrated networks involving metabolism, immunity, and cell death, offering a new framework for studying bone remodeling. These studies have uncovered molecular networks linking osteoporosis with urinary, digestive, infectious, and immune system disorders. LASSO and SVM-RFE algorithms are currently central to biomarker identification in secondary osteoporosis.
The LASSO algorithm is employed to analyze osteoporosis data by eliminating irrelevant genetic features, thereby reducing noise and ensuring data independence. Its robustness in handling high-dimensional genomic datasets with inherent variability makes it particularly suitable for large-scale experimental studies. In urology, research utilizing the NHANES cohort has demonstrated a significant correlation between osteoporosis and osteopenia and the risk of kidney stones (40). Combining LASSO/Boruta algorithms and multi-omics analysis, pivotal regulatory genes, including WNT1, AKT1, and TNF, were identified, with the mTOR signaling pathway emerging as a shared mechanism underlying both conditions. Molecular docking studies further validated that drugs targeting mTOR exhibit dual regulatory effects, indicating potential novel therapeutic strategies for disorders related to bone metabolism and calcium deposition. Wang’s team (41) identified pivotal glutamine metabolism genes (IGKC, TMEM187, RPS11) and confirmed metabolic reprogramming mechanisms by tumor-associated pathway integration, substantiating the glutamine dependency hypothesis. Through the application of weighted gene co-expression network analysis (WGCNA) and iterative LASSO screening, they identified IGKC, TMEM187 five other core genes implicated in bone metabolism imbalance. The discovery extends the theoretical link between tumors and bone diseases and establishes a comprehensive framework from metabolic detection to targeted intervention. The discovery extends the theoretical link between tumors and bone diseases and establishes a comprehensive framework from metabolic detection to targeted intervention.
The use of LASSO for data selection and the establishment of models through the SVE-REF algorithm has become a current research hotspot. Xu’s team combined LASSO and SVE-REF, and through the application of bioinformatics, expanded the research on the underlying mechanisms of infectious diseases and osteoporosis to immune-mediated diseases (42). They were the first to reveal the connection between inflammatory bowel disease (IBD), osteoporosis and the HDAC6/IL-8/PPIF gene cluster, demonstrating that there is a cross-disease association between IBD and osteoporosis. Concurrently, advanced computational integration of protein–protein interaction networks and immune cell profiling techniques has elucidated patho-mechanistic links between ligamentum flavum ossification (OLF) and senile osteoporosis (SOP) through algorithmic unification (43). The identification of the IFNB1 regulatory axis extends SOP research into the domain of ectopic ossification, uncovering an inflammatory-immune-mineralization cascade central to degenerative osteoarticular disorders. Systems biology approaches have driven transformative advances in osteoporosis. On this basis, Lai’s team investigated PMOP through integrated application of WGCNA and machine learning algorithms (SVM-RFE, LASSO, RF, GBM, XG-Boost), establishing a methodological framework that enabled precise identification of ROCK1, KCNJ2, and HIPK1 as key targets from 1,278 candidate genes (44). ROCK1 exhibits dual pathological roles, demonstrating stability as a cross-dataset diagnostic marker while showing low-level expression correlation with activated innate immune pathways. Pan-cancer analyses reveal ROCK1’s implication in tumor progression, suggesting potential epigenetic crosstalk between bone metabolism and the tumor microenvironment, while methodologically aligning with prior glutamine metabolic gene through shared LASSO and SVM-RFE algorithm. While these studies all adopted a modeling approach combining LASSO and SVM-RFE, they focused on distinct mechanisms: one on the regulation of the immune microenvironment and the other on metabolic reprogramming. Recent studies have integrated the GBM-RFE algorithm to uncover the core pathological mechanisms underlying infectious diseases and osteoporosis. The tri-network diagnostic model for HBV infection, liver fibrosis, and bone metabolism disorders was enhanced through SVM-RFE, GBM-RFE, RF-RFE, and LASSO algorithms, identifying 16 gene biomarkers (USP10, ERAL1, ECM1) that achieved 79% diagnostic accuracy via rigorous gene screening, marking significant progress in hepatogenic bone disease research (45). By integrating omics data with algorithmic models, researchers are constructing a comprehensive knowledge network that links molecular interactions to clinical applications, thereby advancing bone metabolism disorder research beyond traditional paradigms.
2.1.3 AI identifies osteoporosis biomarkers induced by environmental pollutants
The integration of AI-powered multi-omics is revolutionizing the investigation of the effects of environmental pollutants on osteoporosis. Osteoporosis biomarker identification linked to environmental pollutants primarily relies on decision tree algorithms such as Boost and RF.
XG-Boost optimizes the traditional Boost algorithm, making it faster and using less memory when dealing with large datasets. The XG-Boost algorithm, when combined with the GBM algorithm, prevents overfitting of data and maintain better performance in cases with excessive data noise. Network toxicology, integrated with the Stepglm + GBM model, identified critical genes CKM and MMP12 associated with phthalate esters (BBP/DBP/DEHP) and osteoporosis (46). Molecular docking analyses demonstrated that these pollutants interact with TRP and GLU amino acids, disrupting the IL-17/TNF signaling pathway and influencing osteoclast differentiation. Establishing the inaugural molecular pathway linking chemical-gene interactions to imbalances in bone metabolism, XG-Boost, which integrates RF and SVM algorithms, intelligently combines the results of all models, reducing the risk of overfitting and significantly improving the model’s AUC (47). Building on this foundation, Huang’s team found that environmental endocrine disruptors (EDCs) affect TNF signaling by forming hydrogen bonds with FOXO3 and LUM. Among 13 EDCs, dexamethasone, perfluorooctanoic acid, and soy isoflavones bind to FOXO3, while genistein and prednisolone interact with LUM, disrupting bone-related gene networks and providing cross-scale insights of how EDCs contribute to osteoporosis pathogenesis.
The RF algorithm integrates SVM-RFE and LASSO for feature selection, enabling systematic identification of the optimal feature subset (48). This approach was applied to identify PANK2 as a key mediator in PFAS-induced osteogenesis. Also, simulations showed stable hydrogen bonds between PFOS and PANK2, and experiments confirmed its role in regulating chemokine pathways via miR-26a-5p, leading to imbalances in BMSCs’ osteogenic and lipogenic differentiation. This integrative methodology, encompassing molecular docking, machine learning, and experimental validation, demonstrates that while various pollutants engage distinct molecular mechanisms (CKM/MMP12, FOXO3/LUM), they collectively perturb osteogenesis, osteoclast homeostasis, and the equilibrium between osteogenesis and lipogenesis. These insights provide potential biomarkers for the early detection of osteoporosis and establish a foundation for developing targeted therapies and strategies for environmental risk mitigation. In terms of identifying osteoporosis molecular markers, the adoption of various algorithms to improve the model has become the current trend. The strategic integration of LASSO for data preprocessing, Boost algorithms for memory optimization, and ensemble methods (RF/SVM/GBM) for multi-perspective feature selection significantly enhances feature stability, thereby improving model performance and interpretability. These methodological advancements establish an interdisciplinary framework synergizing AI algorithms, multi-omics integration, and molecular biomarkers, catalyzing innovative approaches for osteoporosis biomarker discovery.
2.2 Novel drug discovery for osteoporosis
Artificial intelligence has demonstrated significant potential in advancing osteoporosis therapeutics (49, 50). Drug discovery-driven innovative modeling approach, synergized with multi-source data integration, has substantially improved screening efficiency and reduced development timelines (51, 52). Furthermore, AI facilitates the optimization of clinical trial design and accelerates the overall drug development pipeline (Table 2).
Table 2. AI enhances osteoporosis drug discovery.
Drug discovery processes established machine learning algorithms that retain their efficacy in data analysis, while AI has achieved comprehensive integration throughout pharmaceutical development, spanning target identification, absorption distribution metabolism excretion (ADMET) prediction, and clinical trial execution (53). The RF/SVM-SFE algorithm-powered model emerges as a robust methodology for both core gene discovery and small molecule compound recognition. The integration of LASSO, SVM, and RF algorithms has identified BRSK2 and VPS35 as core genes associated with programmed cell death (54). By combining network toxicology with RF/SVM-SFE algorithms, MR, and single-cell sequencing, it was demonstrated that HMOX1-positive macrophages promote M2 polarization through the ANXA1/MIF signaling axis, while molecular docking analyses validated the binding potential of geniposide (55). Further, the algorithm integrating SVM-RFE, LASSO, and WGCNA establishes a multi-level model for analyzing transcriptomics and proteomics data. It has validated mitophagy core genes, NELFB, SFSWAP, and MAP3K3, and predicted four potential therapeutic agents: Vinclozolin, Bisphenol A, Benzo(a)pyrene, and Valproic acid (56). The model also investigated the link between endoplasmic reticulum stress (ERS) and mitochondrial dysfunction (57). Combining RF with these methods identified five key genes, AAAS, ESR1, SLC12A2, TAF15, and VAMP2, enriched in lipid metabolism, calcium transport, and ossification pathways. Bumetanide and elacestrant were further screened as candidate therapeutics via qRT-PCR and molecular docking.
Recently, deep learning models have also been applied to drug development. These models directly learn multi-level features from raw data, optimizing the exploration methods of pharmacological mechanisms. Network pharmacology was combined with DeepPurpose to show that salvianine affects bone metabolism via the CASP3/CTNNB1/ERBB2 pathways (58). Drug-target networks were predicted using DGIdb, revealing that BRSK2 binds to hesperadin and VPS35 interacts with melagatran, offering new insights into targeted therapies. The model was further refined through the development of a deep learning efficacy prediction system (DLEPS), which found that dihydroartemisinin enhances the stemness of BMSCs via GCN5-mediated H3K9 acetylation and led to the creation of a bone-targeted nano-delivery system that improves therapeutic outcomes (59).
The graph neural network model based on deep learning further predicts the connections among diseases, targets, and drugs by constructing a deeper graph network. The development of the Deep Transformer model employing graph neural networks has demonstrated remarkable efficacy (AUC = 0.9916) in predicting the bioactivities of puerarin and aucubin (60). By integrating chemical informatics with deep learning methodologies, the researchers have established a comprehensive pharmacological network for anti-osteoporosis herbal medicines, successfully identifying 89 active ingredients and 30 target genes (61). The study substantiates the roles of phytosterols, flavonoids, and alkaloids as key active components and confirms the interaction between YYH flavonoids and HSD17B2. Furthermore, the application of Chemprop’s graph neural network (AUC = 0.93) and molecular docking techniques facilitated the identification of potent CTSK inhibitors, including quercetin, γ-linolenic acid, and benzyl isothiocyanate (62). These compounds exert their inhibitory effects on CTSK by stabilizing the catalytic triad (Cys25/His162/Asn162) through hydrogen bonding. The development of new drugs for osteoporosis can utilize both established machine learning methods and deep learning methods. Established machine learning methods ensure the accuracy of gene and target characteristics, enabling drug development even with limited data. In deep learning algorithms, graph neural network algorithms integrate multi-omics data and can extract gene and target features from vast databases, demonstrating significant potential in advancing drug development and translational medicine. AI has achieved significant breakthroughs in molecular biomarkers, molecular mechanisms, and targeted drug development. The iterative research framework encompassing “target discovery,” “molecular mechanism,” and “clinical application” is accelerating paradigm innovation in translational medicine.
3 AI-driven strategies for osteoporosis management: clinical innovations and challenges
Osteoporosis, a metabolic bone disease associated with high disability and mortality, is benefiting from major advances in early risk screening and diagnosis through AI (63, 64). AI-driven models have surpassed the sensitivity of traditional imaging and are transforming the diagnosis of osteoporosis and fracture risk through intelligent multimodal imaging analysis. They fuse cross-modal data (CT/MRI/X-ray) with deep learning and optimize interpretable features (65–67). By integrating biomechanical simulations and reinforcement learning, these systems accurately predict bone strength and assess fracture risk, enabling an intelligent diagnostic process that connects imaging, biomechanics, and clinical decisions (Figure 2).
Figure 2. AI improves osteoporosis diagnosis and treatment. AI integrates multimodal imaging, bone morphology, clinical profiles, and serology with risk assessment frameworks to enhance osteoporosis risk stratification, facilitate precise diagnosis, and enable proactive warning systems. Through deep learning architectures, including Artificial Neural Networks (ANNs), Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs), and Graph Neural Networks (GNNs). AI synthesizes personalized therapeutic regimens. By incorporating surgical prognosis prediction and drug efficacy assessment, AI establishes a comprehensive intelligent management system encompassing osteoporosis screening, intervention, and long-term monitoring.
3.1 Application of AI in osteoporosis diagnosis
AI is reshaping osteoporosis screening systems by advancing image processing in osteoporosis diagnosis through deep feature extraction based on artificial neural network methods (Table 3). Meanwhile, researchers overcome DXA scan limitations by developing multimodal imaging-based intelligent diagnostic frameworks (68, 69).
Table 3. Application of AI in the risk prediction and diagnosis of osteoporosis.
The artificial neural networks (ANN) method eliminates the need for manual feature design and has broad applications in osteoporosis diagnosis. For example, existing research combines ANN with ultrasound frequency-dependent attenuation to estimate microstructural parameters of cortical bone, such as pore diameter (φ), pore density (ρ), and porosity (ν) (70). The DNN algorithm, built upon the foundation of ANN, enables the extraction of complex and abstract features, making it particularly well-suited for analyzing intricate bone imaging data and more closely aligned with real-world clinical applications. Leveraging the DNN algorithm, a model incorporating hip joint images and Grad-CAM visualization achieved an accuracy of 71.8%, sensitivity of 83.7%, specificity of 38.7%, and an AUC of 0.700 (71). To address the challenges posed by high noise and low contrast in X-ray images, an innovative Frequency Channel-Wise Transformer Network (FCoTNet) was developed, integrating multi-scale feature extraction and convolutional fusion across diverse receptive fields (72). The model achieved an AUC of 0.8718, with accuracy, sensitivity, and specificity reaching 78.29, 69.72, and 88.92%, respectively. Furthermore, it enhanced clinicians’ diagnostic performance by 18.5% in accuracy, 18.93% in sensitivity, and 9.83% in specificity. Multimodal data fusion has emerged as a promising strategy to improve diagnostic efficiency.
Convolutional neural networks (CNNs) are a specialized type of DNN designed for grid-structured data like medical images. They are better suited for bone image analysis than conventional DNNs and are less prone to overfitting. Leveraging these advantages, researchers have developed intelligent diagnostic systems using multimodal imaging that surpassed the limitations of DXA scans. Researchers assessed bone density and demonstrated robust cross-racial performance, with an AUC of 0.88, accuracy of 77.69%, specificity of 74.19%, and sensitivity of 86.16%, indicating the potential for a more universally applicable screening approach (73). Breakthroughs in advanced 3D image analysis and micro-parameter measurement technologies have enabled the development of a multi-tier CNN architecture utilizing lumbar X-ray images, significantly enhancing opportunistic CT screening capabilities. A CNN-based DeepDXA model demonstrates high accuracy in estimating bone mineral density from pelvic X-ray images, exhibiting strong predictive performance with an accuracy of 84%, sensitivity of 76%, specificity of 87%, and an AUC of 0.893 (74). Advances in AI have driven the development of innovative osteoporosis diagnostic methods, with model interpretability emerging as a critical factor for clinical adoption (75). As reported (10), researchers improved a deep U-Net-based diagnostic algorithm that integrates softmax prediction with a cross-entropy loss function to construct an energy function, enabling more accurate bone mineral density measurement with an automatic recognition rate exceeding 81%, demonstrating strong clinical potential. A deep learning model combining hand X-ray and DXA images achieved 94.37% accuracy, 93.45% sensitivity, 97.05% specificity, and an AUC of 0.9525, showing potential for early screening (76).
Model interpretability has become a key factor in clinical adoption (77, 78). By integrating Grad-CAM technology, a CNN-based HarDNet deep learning model was introduced for non-invasive bone mineral density prediction using hand X-ray images, enhanced interpretability through heat map visualization, achieving a sensitivity of 78.9, 80.6, specificity of 94.7, 90.5, accuracy of 91.4, 87.6, and AUC of 0.96, 0.92, showing promise for early screening, showing promise for early screening. Superfluity deep learning model analyzing knee X-ray images achieved an AUC of 0.6833, an accuracy of 74.51%, and a specificity of 79.33%, demonstrating potential for multi-site skeletal evaluation (79). Further, advances in AI have led to the development of innovative methods for osteoporosis diagnosis (80). An improved DenseNet121 model was applied to perform 3D segmentation and predict bone mineral density of T12, L1, and L2 vertebrae using low-dose chest CT images. Bland–Altman analysis confirmed the system’s high performance, with sensitivity over 86%, specificity over 98%, and an accuracy rate of 86.6%. The system is compatible with images from different scanners and unaffected by age or gender, demonstrating strong cross-device stability and broad applicability. A model using CT attenuation values showed that each 10 HU increase was associated with a 32–44% lower risk of osteoporosis (AUC = 0.831), with particular relevance for menopausal women (66). New algorithm architectures have improved the fusion of clinical information in data integration. Another study applied graph neural networks (GNN) to model personal health and lifestyle data as topological structures, enabling dynamic risk assessment and personalized prevention strategies (81).
The hybrid deep learning framework (HDLF) combines various types of ANN algorithms and enhances the integration of clinical information in data consolidation. Research combined pelvic X-rays, vertebral data, and clinical variables to predict BMD and classify disease with high accuracy (AUC 0.97, accuracy 96%, specificity 97%, sensitivity 88%) (82). Notably, this approach was further refined in later research, where a generalized additive model (GA2M) identified that the interaction between spinal BMD and body weight was a key predictor of 10-year osteoporosis risk (AUC 0.83) (83). On model interpretability, researchers used SHAP/LIME/ELI5 tools to analyze a multi-layer ensemble model and found that low femoral neck and thoracolumbar T values (≤ − 2.5) were major risk factors, with SHAP values strongly negatively correlated with disease risk (84). According to research findings (71, 82), models developed using the DNN algorithm demonstrate higher sensitivity but relatively lower accuracy, whereas those utilizing the HDLF algorithm achieve the highest accuracy. The SHAP tool has high computational complexity, resulting in significant computational costs for large models. The LIME tool requires subjective selection of kernel functions and kernel width, which can significantly affect the interpretation results, as different settings may highlight different features. Compared to SHAP and LIME, ELI5 has relatively weaker interpretability capabilities for complex data such as images and text.
Accurate diagnosis of osteoporosis constitutes a critical component of its clinical management, offering substantial value in slowing disease progression and preventing fragility fractures (85). Within the domain of imaging diagnostics, intelligent diagnostic approaches are advancing osteoporosis assessment toward enhanced precision and efficiency. As more clinical evidence emerges, AI-assisted diagnostic systems are poised to play a central role in osteoporosis management.
3.2 Application of AI in osteoporotic fracture risk assessment
Osteoporotic fractures represent the most common complication of osteoporosis and a leading cause of disability and mortality (86). The application of AI in osteoporosis risk assessment is transforming traditional diagnostic paradigms towards greater precision. Future research should emphasize its role in fracture risk evaluation to strengthen clinical decision support (87). The Fracture Risk Assessment Tool (FRAX) system is the most widely adopted tool for global fracture risk assessment, providing a ten-year probability model based on clinical factors to enable effective risk stratification (88). With advances in deep learning, intelligent assessment leveraging medical imaging has evolved from qualitative interpretation to quantitative prediction.
By integrating hip and spine X-ray images with an ANN model, researchers achieved high diagnostic accuracy for hip osteoporosis (91.7%), lumbar osteoporosis (86.2%), 10-year osteoporotic fracture prediction (95%), and high-risk hip fractures (90%) (89). These results showed no significant difference compared to the FRAX system, confirming the reliability of AI in standardized risk assessment. The development of dynamic risk models introduces a novel temporal dimension for disease monitoring. The integration of TWIST and ANN neural networks with a femoral bone strain index (Femoral BSI) demonstrated a specificity of 82.93%, sensitivity of 75.76%, accuracy of 79.34%, and an AUC of 0.846 in identifying vertebral fractures among postmenopausal women (90). Compared to the conventional ANN algorithm, the improved DNN algorithm employing DeepHit enables non-parametric data analysis without reliance on predefined distribution assumptions. A dynamic scoring system combining DeepHit and Cox proportional hazards models successfully predicted hip fracture risk in a discovery cohort, achieving a Harrell’s C index of 0.860 and accurately forecasting a 10-year incidence of severe fractures at 68.8%, thereby significantly enhancing early warning capabilities for critical fracture events (91). A deep learning model for vertebral fracture classification based on spinal X-rays attained an AUC of 0.948, with a sensitivity of 54.5%, a specificity of 99.7%, and an accuracy of 89.8% in independent testing (92). An AUC-ROC exceeding 0.90 confirms its diagnostic performance, which is comparable to that of experienced radiologists.
The integration of radiomics and 3D modeling has expanded the applicability of CNN algorithms in bone morphological analysis. A CNN-based thoracolumbar spine CT analysis system achieved 83.9% accuracy in identifying compression fractures by extracting discriminative imaging features (AUC = 0.883) (93). A 3D CT-based dynamic prediction model for hip fractures, built on the DenseNet architecture, maintained an AUC of 0.73–0.74 over 1 year, outperforming conventional clinical prediction models (94). For special populations, multi-omics data fusion is enabling more precise preventive strategies. A gradient boosting model integrating genetic risk scores achieved 88% prediction accuracy in elderly men (AUC = 0.71), surpassing the FRAX method and demonstrating the value of multi-omics integration in personalized risk assessment (95). Opportunistic screening using routine imaging data provides essential technical support for establishing a tiered diagnosis and treatment framework. Further optimization led to the development of a GoogLeNet-based model for fracture severity grading, designed specially to identify moderate to severe fractures (96). This model achieved an AUC-ROC of 0.99 and an area under the precision-recall curve (AUPRC) of 0.82, balancing a high positive predictive value (91.2%) with moderate sensitivity (59.8%). Through optimized probability thresholds, it minimizes unnecessary interventions. This hierarchical early warning mechanism establishes the algorithmic foundation for automated screening, marking a significant advancement toward the clinical implementation of AI-assisted diagnostic systems.
3.3 Application of AI in osteoporosis treatment and prognosis assessment
Improving the precision of pharmacological interventions and optimizing prognostic evaluation remain key priorities in clinical research (97). Notably, AI exhibits significant potential in predicting disease progression, developing personalized treatment strategies, managing postoperative care, and assessing drug efficacy via its advanced data processing and pattern recognition capabilities (98, 99).
AI also supports the development of personalized treatment strategies and offers broad applications in osteoporosis care (100). An RF model utilizing 8,981 clinical variables achieved an AUC of 0.70 and an accuracy of 69% in predicting responses to 11 distinct treatments (101). The model demonstrated a 9.54% improvement over observed clinical outcomes, thereby reinforcing the potential for more individualized therapeutic approaches. Further, a clinical decision support system that integrated DXA results with clinical data through hierarchical feature extraction and novel rule-based evaluation methods achieved decision accuracy of 90%, providing an actionable strategy for prevention and treatment planning (102). Different machine learning algorithms exhibit varied efficacy in distinct applications. Comparative analyses have revealed that among ANN, RF, SVM, and LR models, the RF model achieved the highest accuracy at 75.0%, while the LR model yielded the highest AUC at 0.731. These findings indicate that algorithm selection should be tailored to specific clinical objectives and evaluation metrics (103). This principle is further validated by a multimodal evaluation framework that elucidated the regulatory mechanisms of the miRNA-m6A network and accurately predicted drug sensitivity to clozapine and aspirin by identifying key genes DEFA4 and HLA-DPB1 using the SVM algorithm (104). Intelligent algorithms have demonstrated substantial efficacy in postoperative prognosis management (105). An RF model with an AUC of 0.953 demonstrated superior performance compared to traditional logistic regression (AUC = 0.831) in predicting vertebral fracture recurrence among 529 patients undergoing percutaneous kyphoplasty (PKP), highlighting the potential of machine learning to enhance surgical decision-making (106). Moreover, an SVM classifier, achieving an AUC of 0.85, accuracy of 81%, and a sensitivity of 98%, effectively identified multidimensional risk factors for PKP recurrence through multivariate logistic regression, establishing a quantitative foundation for preventive interventions (107). Deep learning is increasingly employed in the assessment of pharmacological efficacy. An ANN analysis of spinal radiographs and bone strength parameters revealed that teriparatide enhanced BSI by 13.9%, trabecular texture bone score (TBS) by 5.08%, and BMD by 8.36%, thereby demonstrating its effectiveness in mitigating fracture risk (108). AI is transforming osteoporosis diagnosis and treatment, from risk assessment and diagnosis to clinical solution optimization and efficacy evaluation by creating an intelligent decision support network. As algorithms integrate with multimodal data, intelligent systems are enhancing clinical decision accuracy and advancing personalized, dynamic management of osteoporosis.
4 Conclusion and perspective
AI transforms healthcare by integrating multimodal data and advanced technologies. In diagnosis, a combination of multi-omics data, including genomics, proteomics, metabolomics, and radiomics, is utilized to build dynamic visual pathological models that enable disease risk stratification and early warning systems. In clinical decision-making, intelligent monitoring systems analyze biomolecules and radiomic features to create a virtual-real integrated diagnosis and treatment ecosystem with predictive intervention. Furthermore, machine learning and network pharmacology models enhance the understanding of the molecular biomarkers and pathological mechanisms underlying osteoporosis. Deep learning-driven molecular dynamics simulations accelerate new drug discovery. The AI system has established a novel digital precision medicine paradigm by integrating biomarker monitoring, clinical imaging analysis, personalized medication plans, and drug innovation while optimizing clinical decisions to advance osteoporosis diagnosis and treatment toward enhanced precision and intelligence.
Nonetheless, while AI holds significant potential in clinical applications, it still carries certain limitations. Regarding data standardization, different data subsets are influenced by gradient descent algorithms during preprocessing, which can easily induce model drift and lead to degraded model performance. In terms of interpretability, algorithmic bias means that some training data may not represent the target population, resulting in reduced diagnostic accuracy. From an ethical standpoint, the clinical application of artificial intelligence raises significant concerns regarding patient data privacy. These challenges pose substantial barriers to the regulatory approval of AI-integrated healthcare technologies. Future research should foster collaboration between AI researchers and clinicians to integrate AI findings into clinical workflows, improving model performance and interpretability. Incorporating emerging approaches such as federated learning and multi-modal data coordination, and addressing real-world integration challenges can help resolve ethical concerns in the clinical use of AI.
Author contributions
S-TF: Conceptualization, Data curation, Investigation, Methodology, Software, Supervision, Writing – original draft, Writing – review & editing. ML: Conceptualization, Writing – original draft. J-LD: Conceptualization, Data curation, Writing – original draft. Y-LL: Data curation, Writing – review & editing. L-NH: Data curation, Methodology, Writing – original draft. R-CD: Conceptualization, Data curation, Methodology, Writing – original draft. M-DH: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing.
Funding
The author(s) declare that financial support was received for the research and/or publication of this article. This work was supported by the Shandong Key Laboratory of Technologies and Systems for Intelligent Construction Equipment. We express our gratitude to Figdraw, as the figures in this review were created with its assistance (ID: ISYSY41482, POWYR3d3db, ISOTO7c229).
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The authors declare that no Gen AI was used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1. Wang, J, Shu, B, Tang, D, Li, C, Xie, X, Jiang, L, et al. The prevalence of osteoporosis in china, a community based cohort study of osteoporosis. Front Public Health. (2023) 11:1084005. doi: 10.3389/fpubh.2023.1084005,
2. Cruz, A, Lins, H, Medeiros, R, Filho, J, and da Silva, S. Artificial intelligence on the identification of risk groups for osteoporosis, a general review. Biomed Eng Online. (2018) 17:12. doi: 10.1186/s12938-018-0436-1,
3. Aibar-Almazán, A, Voltes-Martínez, A, Castellote-Caballero, Y, Afanador-Restrepo, D, Carcelén-Fraile, M, and López-Ruiz, E. Current status of the diagnosis and management of osteoporosis. Int J Mol Sci. (2022) 23:9465. doi: 10.3390/ijms23169465,
4. Coughlan, T, and Dockery, F. Osteoporosis and fracture risk in older people. Clin Med. (2014) 14:187–91. doi: 10.7861/clinmedicine.14-2-187,
5. Roussy, J, Bessette, L, Bernatsky, S, Rahme, E, and Lachaine, J. Rates of non-vertebral osteoporotic fractures in rheumatoid arthritis and postfracture osteoporosis care in a period of evolving clinical practice guidelines. Calcif Tissue Int. (2014) 95:8–18. doi: 10.1007/s00223-014-9856-5,
6. Oberkircher, L, Ruchholtz, S, Rommens, P, Hofmann, A, Bücking, B, and Krüger, A. Osteoporotic pelvic fractures. Dtsch Arztebl Int. (2018) 115:70–80. doi: 10.3238/arztebl.2018.0070,
7. Paranthaman, M. Linking bone marrow fat with decreased bone mineral density among indian patients with osteoporotic fracture. Bioinformation. (2024) 20:49–53. doi: 10.6026/973206300200049,
8. Zhao, J, Shi, H, Jiang, D, Wang, L, Chen, S, and Jia, W. Analysis of combined indicators for risk of osteoporotic hip fracture in elderly women. Orthop Surg. (2021) 13:1205–12. doi: 10.1111/os.12974,
9. Roux, C, Thomas, T, Paccou, J, Bizouard, G, Crochard, A, Toth, E, et al. Refracture and mortality following hospitalization for severe osteoporotic fractures: the fractos study. JBMR Plus. (2021) 5:e10507. doi: 10.1002/jbm4.10507,
10. Liu, J, Wang, J, Ruan, W, Lin, C, and Chen, D. Diagnostic and gradation model of osteoporosis based on improved deep u-net network. J Med Syst. (2020) 44:15. doi: 10.1007/s10916-019-1502-3,
11. Ye, X, Jiang, H, Wang, Y, Ji, Y, and Jiang, X. A correlative studies between osteoporosis and blood cell composition. Medicine. (2020) 99:e20864. doi: 10.1097/MD.0000000000020864,
12. Khosla, S. Current state of science on drug therapies for osteoporotic fracture prevention. Innovation. Aging. (2019) 3:S748–8. doi: 10.1093/geroni/igz038.2744
13. Jung, Y, Ko, Y, Kim, H, Ha, Y, Lee, Y, Kim, T, et al. Gender differences in anti-osteoporosis drug treatment after osteoporotic fractures. J Bone Miner Metab. (2019) 37:134–41. doi: 10.1007/s00774-018-0904-5,
14. Kim, B, Lee, S, and Koh, J. Potential biomarkers to improve the prediction of osteoporotic fractures. Endocrinol Metab. (2020) 35:55–63. doi: 10.3803/EnM.2020.35.1.55,
15. Ramírez-Salazar, E, Carrillo-Patiño, S, Hidalgo-Bravo, A, Rivera-Paredez, B, Quiterio, M, Ramírez-Palacios, P, et al. Serum mirnas mir-140-3p and mir-23b-3p as potential biomarkers for osteoporosis and osteoporotic fracture in postmenopausal mexican-mestizo women. Gene. (2018) 679:19–27. doi: 10.1016/j.gene.2018.08.074,
16. Grzybowski, A, Pawlikowska-Łagód, K, and Lambert, W. A history of artificial intelligence. Clin Dermatol. (2024) 42:221–9. doi: 10.1016/j.clindermatol.2023.12.016,
17. Shin, Y, Kim, S, and Lee, Y. Ai musculoskeletal clinical applications: how can ai increase my day-to-day efficiency? Skeletal Radiol. (2022) 51:293–304. doi: 10.1007/s00256-021-03876-8,
18. Jakhar, D, and Kaur, I. Artificial intelligence, machine learning and deep learning: definitions and differences. Clin Exp Dermatol. (2020) 45:131–2. doi: 10.1111/ced.14029,
19. LeCun, Y, Bengio, Y, and Hinton, G. Deep learning. Nature. (2015) 521:436–44. doi: 10.1038/nature14539,
20. Meng, J, Du, H, Lv, H, Lu, J, Li, J, and Yao, J. Identification of the osteoarthritis signature gene pdk1 by machine learning and its regulatory mechanisms on chondrocyte autophagy and apoptosis. Front Immunol. (2023) 13:1072526. doi: 10.3389/fimmu.2022.1072526,
21. Xu, Z, Chen, Q, Zhou, Z, Sun, J, Tian, G, Liu, C, et al. Screening risk factors for the occurrence of wedge effects in intramedullary nail fixation for intertrochanteric fractures in older people via machine learning and constructing a prediction model: a retrospective study. BMC Musculoskelet Disord. (2025) 26:403. doi: 10.1186/s12891-025-08619-7,
22. Hardalaç, F, Uysal, F, Peker, O, Çiçeklidağ, M, Tolunay, T, Tokgöz, N, et al. Fracture detection in wrist x-ray images using deep learning-based object detection models. Sensors. (2022) 22:1285. doi: 10.3390/s22031285,
23. Nielsen, R, Monfeuga, T, Kitchen, R, Egerod, L, Leal, L, Schreyer, A, et al. Data-driven identification of predictive risk biomarkers for subgroups of osteoarthritis using interpretable machine learning. Nat Commun. (2024) 15:2817. doi: 10.1038/s41467-024-46663-4,
24. Eraslan, G, Avsec, Ž, Gagneur, J, and Theis, F. Deep learning: new computational modelling techniques for genomics. Nat Rev Genet. (2019) 20:389–403. doi: 10.1038/s41576-019-0122-6,
25. Romiti, S, Vinciguerra, M, Saade, W, Anso Cortajarena, I, and Greco, E. Artificial intelligence (ai) and cardiovascular diseases: an unexpected alliance. Cardiol Res Pract. (2020) 2020:1–8. doi: 10.1155/2020/4972346,
26. Lee, S, Mohr, N, Street, N, and Nadkarni, P. Machine learning in relation to emergency medicine clinical and operational scenarios: an overview. Western J Emergency Med. (2019) 20:219–27. doi: 10.5811/westjem.2019.1.41244,
27. Alam, M, Le, D, Lim, J, Chan, R, and Yao, X. Supervised machine learning based multi-task artificial intelligence classification of retinopathies. J Clin Med. (2019) 8:872. doi: 10.3390/jcm8060872,
28. Chai, H, Liang, Y, Wang, S, and Shen, H. A novel logistic regression model combining semi-supervised learning and active learning for disease classification. Sci Rep. (2018) 8:13009. doi: 10.1038/s41598-018-31395-5,
29. Guo, J, Huang, Q, Zhou, Y, Xu, Y, Zong, C, Shen, P, et al. Typing characteristics of metabolism-related genes in osteoporosis. Front Pharmacol. (2022) 13:999157. doi: 10.3389/fphar.2022.999157,
30. Hao, S, Xinqi, M, Weicheng, X, Shiwei, Y, Lumin, C, Xiao, W, et al. Identification of key immune genes of osteoporosis based on bioinformatics and machine learning. Front Endocrinol. (2023) 14:1118886. doi: 10.3389/fendo.2023.1118886,
31. Fang, S, Ni, H, Zhang, Q, Dai, J, He, S, Min, J, et al. Integrated single-cell and bulk rna sequencing analysis reveal immune-related biomarkers in postmenopausal osteoporosis. Heliyon. (2024) 10:e38022. doi: 10.1016/j.heliyon.2024.e38022,
32. Xu, J, Cai, X, Miao, Z, Yan, Y, Chen, D, Yang, Z, et al. Proteome‐wide profiling reveals dysregulated molecular features and accelerated aging in osteoporosis: a 9.8‐year prospective study. Aging Cell. (2024) 23:e14035. doi: 10.1111/acel.14035,
33. Zheng, Y, Li, J, Li, Y, Wang, J, Suo, C, Jiang, Y, et al. Plasma proteomic profiles reveal proteins and three characteristic patterns associated with osteoporosis: a prospective cohort study. J Adv Res. (2025) 75:491–503. doi: 10.1016/j.jare.2024.10.019,
34. Yang, X, Zhang, Z, Lu, Y, Chen, H, Wang, H, Lin, T, et al. Identification and experimental validation of programmed cell death- and mitochondria-associated biomarkers in osteoporosis and immune microenvironment. Front Genet. (2024) 15:1439171. doi: 10.3389/fgene.2024.1439171,
35. Liu, D, Zhijun, H, Tang, Z, Li, P, Yuan, W, Li, F, et al. Identification of biomarkers associated with oxidative stress-related genes in postmenopausal osteoporosis. Cell Mol Biol. (2023) 69:186–92. doi: 10.14715/cmb/2023.69.6.28,
36. Hu, Y, Han, J, Ding, S, Liu, S, and Wang, H. Identification of ferroptosis-associated biomarkers for the potential diagnosis and treatment of postmenopausal osteoporosis. Front Endocrinol. (2022) 13:986384. doi: 10.3389/fendo.2022.986384,
37. Huo, Y, Guo, M, Li, Y, Yao, X, Tian, Q, and Liu, T. Identification of programmed cell death-related biomarkers for the potential diagnosis and treatment of osteoporosis. Endocr Metab Immune Disord Drug Targets. (2025) 25:864–78. doi: 10.2174/0118715303326112241021061805
38. Huang, X, Ma, J, Wei, Y, Chen, H, and Chu, W. Identification of biomarkers associated with diagnosis of postmenopausal osteoporosis patients based on bioinformatics and machine learning. Front Genet. (2023) 14:1198417. doi: 10.3389/fgene.2023.1198417,
39. Wang, X, Pei, Z, Hao, T, Ariben, J, Li, S, He, W, et al. Prognostic analysis and validation of diagnostic marker genes in patients with osteoporosis. Front Immunol. (2022) 13:987937. doi: 10.3389/fimmu.2022.987937,
40. Luo, D, Xie, L, Zhang, J, and Liu, C. Exploring the association between osteoporosis and kidney stones: a clinical to mechanistic translational study based on big data and bioinformatics. Biol Direct. (2025) 20:42. doi: 10.1186/s13062-025-00627-w,
41. Wang, L, Deng, C, Wu, Z, Zhu, K, and Yang, Z. Bioinformatics and machine learning were used to validate glutamine metabolism-related genes and immunotherapy in osteoporosis patients. J Orthop Surg Res. (2023) 18:685. doi: 10.1186/s13018-023-04152-2,
42. Xu, G, Zhang, W, Yang, J, Sun, N, and Qu, X. Identification of neutrophil extracellular traps and crosstalk genes linking inflammatory bowel disease and osteoporosis by integrated bioinformatics analysis and machine learning. Sci Rep. (2023) 13:23054. doi: 10.1038/s41598-023-50488-4,
43. Zhang, Y, Huang, H, Chen, H, Zhang, P, Liu, Y, Gan, Y, et al. Unearths ifnb1 immune infiltrates in sop-related ossification of ligamentum flavum pathogenesis. Heliyon. (2023) 9:e16722. doi: 10.1016/j.heliyon.2023.e16722,
44. Lai, B, Jiang, H, Gao, Y, and Zhou, X. Identification of rock1 as a novel biomarker for postmenopausal osteoporosis and pan-cancer analysis. Aging. (2023) 15:8873–907. doi: 10.18632/aging.205004,
45. Yang, J, Yang, W, Hu, Y, Tong, L, Liu, R, Liu, L, et al. Screening of genes co-associated with osteoporosis and chronic hbv infection based on bioinformatics analysis and machine learning. Front Immunol. (2024) 15:1472354. doi: 10.3389/fimmu.2024.1472354,
46. Zhang, X, Zhu, X, Gu, W, Li, X, Niu, T, Mao, P, et al. Elucidating the mechanism of phthalates induced osteoporosis through network toxicology and molecular docking. Ecotoxicol Environ Saf. (2025) 291:117820. doi: 10.1016/j.ecoenv.2025.117820,
47. Huang, X, Huang, H, Chen, H, and Wei, Y. Identification of endocrine-disrupting chemicals targeting key op-associated genes via bioinformatics and machine learning. Ecotoxicol Environ Saf. (2024) 286:117155. doi: 10.1016/j.ecoenv.2024.117155,
48. Shao, Y, and Fan, W. Multi-level evidence reveals pank2 as a potential target of pfoa/pfos-induced bone metabolism disruption: from network toxicology to in vitro validation. Ecotoxicol Environ Saf. (2025) 302:118593. doi: 10.1016/j.ecoenv.2025.118593,
49. Xu, J, Wen, X, Sun, L, Xing, K, Xue, L, Zhou, S, et al. Large model era: deep learning in osteoporosis drug discovery. J Chem Inf Model. (2025) 65:2232–44. doi: 10.1021/acs.jcim.4c02264,
50. Noorain, L, Nguyen, V, Kim, H, and Nguyen, L. A machine learning approach for plga nanoparticles in antiviral drug delivery. Pharmaceutics. (2023) 15:495. doi: 10.3390/pharmaceutics15020495,
51. Yan, R, Yang, Y, and Chen, Y. Pharmacokinetics of chinese medicines: strategies and perspectives. Chin Med. (2018) 13:24. doi: 10.1186/s13020-018-0183-z,
52. Guo, C, and Li, H. Application of 5g network combined with ai robots in personalized nursing in china: a literature review. Front Public Health. (2022) 10:948303. doi: 10.3389/fpubh.2022.948303,
53. Wu, Z, Lei, T, Shen, C, Wang, Z, Cao, D, and Hou, T. Admet evaluation in drug discovery. 19. reliable prediction of human cytochrome p450 inhibition using artificial intelligence approaches. J Chem Inf Model. (2019) 59:4587–601. doi: 10.1021/acs.jcim.9b00801.s001
54. Li, M, Wang, X, Guo, M, Zhang, W, Li, T, and Zheng, J. Identification of potential cell death-related biomarkers for diagnosis and treatment of osteoporosis. BMC Musculoskelet Disord. (2024) 25:235. doi: 10.1186/s12891-024-07349-6,
55. Li, Y, Liang, X, Rong, Y, Jiang, K, Zhang, J, and Li, G. Investigating the potential risk of cadmium exposure on osteoporosis: an integrated multi-omics approach. Ecotoxicol Environ Saf. (2025) 301:118502. doi: 10.1016/j.ecoenv.2025.118502,
56. Su, Y, Yu, G, Li, D, Lu, Y, Ren, C, Xu, Y, et al. Identification of mitophagy-related biomarkers in human osteoporosis based on a machine learning model. Front Physiol. (2024) 14:1289976. doi: 10.3389/fphys.2023.1289976,
57. Chen, Y, Bi, K, Zhang, C, Gu, J, Yu, Z, Lu, J, et al. Identification of endoplasmic reticulum stress and mitochondrial dysfunction related biomarkers in osteoporosis. Hereditas. (2025) 162:21. doi: 10.1186/s41065-025-00387-7,
58. Tang, Y, Zhou, D, Gan, F, Yao, Z, and Zeng, Y. Exploring the mechanisms of sanguinarine in the treatment of osteoporosis by integrating network pharmacology analysis and deep learning technology. Curr Comput Aided Drug Des. (2025) 21:83–93. doi: 10.2174/0115734099282231240214095025,
59. Wang, R, Wang, Y, Niu, Y, He, D, Jin, S, Li, Z, et al. Deep learning-predicted dihydroartemisinin rescues osteoporosis by maintaining mesenchymal stem cell stemness through activating histone 3 lys 9 acetylation. ACS Central Science. (2023) 9:1927–43. doi: 10.1021/acscentsci.3c00794,
60. Liu, Y, Jiang, G, Sun, M, Zhou, Z, Liang, P, and Chang, Q. Deeptransformer: node classification research of a deep graph network on an osteoporosis graph based on graphtransformer. Curr Comput Aided Drug Des. (2025) 21:28–37. doi: 10.2174/0115734099266731231115065030,
61. Lin, Z, Wang, S, Liu, Z, Liu, B, Xie, L, and Zhou, J. Exploring anti-osteoporosis medicinal herbs using cheminformatics and deep learning approaches. Comb Chem High Throughput Screen. (2023) 26:1802–11. doi: 10.2174/1386207325666220905155923
62. Li, Q, Han, X, Zhou, S, Lu, Y, Wang, Y, and Yang, J. Discovery of novel cathepsin k inhibitors for osteoporosis treatment using a deep learning-based strategy. Expert Opin Drug Discov. (2025) 20:1345–56. doi: 10.1080/17460441.2025.2527686,
63. Shen, L, Gao, C, Hu, S, Kang, D, Zhang, Z, Xia, D, et al. Using artificial intelligence to diagnose osteoporotic vertebral fractures on plain radiographs. J Bone Miner Res. (2023) 38:1278–87. doi: 10.1002/jbmr.4879,
64. Namireddy, S, Gill, S, Peerbhai, A, Kamath, A, Ramsay, D, Ponniah, H, et al. Artificial intelligence in risk prediction and diagnosis of vertebral fractures. Sci Rep. (2024) 14:30560. doi: 10.1038/s41598-024-75628-2,
65. Kong, S, Cho, W, Park, S, Choo, J, Kim, J, Kim, S, et al. A computed tomography–based fracture prediction model with images of vertebral bones and muscles by employing deep learning: development and validation study. J Med Internet Res. (2024) 26:e48535. doi: 10.2196/48535,
66. Yang, J, Liao, M, Wang, Y, Chen, L, He, L, Ji, Y, et al. Opportunistic osteoporosis screening using chest ct with artificial intelligence. Osteoporos Int. (2022) 33:2547–61. doi: 10.1007/s00198-022-06491-y,
67. Khadivi, G, Akhtari, A, Sharifi, F, Zargarian, N, Esmaeili, S, Ahsaie, M, et al. Diagnostic accuracy of artificial intelligence models in detecting osteoporosis using dental images: a systematic review and meta-analysis. Osteoporos Int. (2025) 36:1–19. doi: 10.1007/s00198-024-07229-8,
68. Subramaniam, S, Ima-Nirwana, S, and Chin, K. Performance of osteoporosis self-assessment tool (ost) in predicting osteoporosis—a review. Int J Environ Res Public Health. (2018) 15:1445. doi: 10.3390/ijerph15071445,
69. Chin, K. A review on the performance of osteoporosis self-assessment tool for asians in determining osteoporosis and fracture risk. Postgrad Med. (2017) 129:734–46. doi: 10.1080/00325481.2017.1353394,
70. Mohanty, K, Yousefian, O, Karbalaeisadegh, Y, Ulrich, M, Grimal, Q, and Muller, M. Artificial neural network to estimate micro-architectural properties of cortical bone using ultrasonic attenuation: a 2-d numerical study. Comput Biol Med. (2019) 114:103457. doi: 10.1016/j.compbiomed.2019.103457,
71. Jang, R, Choi, J, Kim, N, Chang, J, Yoon, P, and Kim, C. Prediction of osteoporosis from simple hip radiography using deep learning algorithm. Sci Rep. (2021) 11:19997. doi: 10.1038/s41598-021-99549-6,
72. Xue, L, Qin, G, Chang, S, Luo, C, Hou, Y, Xia, Z, et al. Osteoporosis prediction in lumbar spine x-ray images using the multi-scale weighted fusion contextual transformer network. Artif Intell Med. (2023) 143:102639. doi: 10.1016/j.artmed.2023.102639,
73. Jang, M, Kim, M, Bae, S, Lee, S, Koh, J, and Kim, N. Opportunistic osteoporosis screening using chest radiographs with deep learning: development and external validation with a cohort dataset. J Bone Miner Res. (2020) 37:369–77. doi: 10.1002/jbmr.4477,
74. Ho, C, Chen, Y, Fan, T, Kuo, C, Yen, T, Liu, Y, et al. Application of deep learning neural network in predicting bone mineral density from plain x-ray radiography. Arch Osteoporos. (2021) 16:153. doi: 10.1007/s11657-021-00985-8,
75. Wu, Y, Yang, X, Wang, M, Lian, Y, Hou, P, Chai, X, et al. Artificial intelligence assisted automatic screening of opportunistic osteoporosis in computed tomography images from different scanners. Eur Radiol. (2024) 35:2287–95. doi: 10.1007/s00330-024-11046-2
76. Mohammadi, F, and Sebro, R. Opportunistic screening for osteoporosis using hand radiographs: a preliminary study. Stud Health Technol Inform. (2023) 302:911–2. doi: 10.3233/shti230306
77. Ho, C, Fan, T, Kuo, C, Yen, T, Chang, S, Pei, Y, et al. Hardnet-based deep learning model for osteoporosis screening and bone mineral density inference from hand radiographs. Bone. (2025) 190:117317. doi: 10.1016/j.bone.2024.117317,
78. Lin, C, Tsai, D, Wang, C, Chao, Y, Huang, J, Lin, C, et al. Osteoporotic precise screening using chest radiography and artificial neural network: the opscan randomized controlled trial. Radiology. (2024) 311:e231937. doi: 10.1148/radiol.231937,
79. Naguib, S, Saleh, M, Hamza, H, Hosny, K, and Kassem, M. A new superfluity deep learning model for detecting knee osteoporosis and osteopenia in x-ray images. Sci Rep. (2024) 14:25434. doi: 10.1038/s41598-024-75549-0,
80. Mao, L, Xia, Z, Pan, L, Chen, J, Liu, X, Li, Z, et al. Deep learning for screening primary osteopenia and osteoporosis using spine radiographs and patient clinical covariates in a Chinese population. Front Endocrinol. (2022) 13:971877. doi: 10.3389/fendo.2022.971877,
81. Wang, Z, Li, Y, and Xu, Y. Osteoporosis risk prediction method based on relational network and GNN In: IEEE 6th Information Technology, Networking, Electronic and Automation Control Conference (ITNEC), vol. 2023 (2023). 1242–7.
82. Zhou, K, Zhu, Y, Luo, X, Yang, S, Xin, E, Zeng, Y, et al. A novel hybrid deep learning framework based on biplanar x-ray radiography images for bone density prediction and classification. Osteoporos Int. (2025) 36:521–30. doi: 10.1007/s00198-024-07378-w,
83. Zeitlin, J, Parides, M, Lane, J, Russell, L, and Kunze, K. A clinical prediction model for 10-year risk of self-reported osteoporosis diagnosis in pre- and perimenopausal women. Arch Osteoporos. (2023) 18:78. doi: 10.1007/s11657-023-01292-0,
84. Khanna, V, Chadaga, K, Sampathila, N, Chadaga, R, Prabhu, S, Swathi, KS, et al. A decision support system for osteoporosis risk prediction using machine learning and explainable artificial intelligence. Heliyon. (2023) 9:e22456. doi: 10.1016/j.heliyon.2023.e22456,
85. Kong, S. Clinical observation of acupuncture treatment for sleep disorders. Minerva Med. (2024). doi: 10.23736/s0026-4806.24.09104-3
86. Trémollieres, F. Assessment and hormonal management of osteoporosis. Climacteric. (2019) 22:122–6. doi: 10.1080/13697137.2018.1555582,
87. Kitcharanant, N, Chotiyarnwong, P, Tanphiriyakun, T, Vanitcharoenkul, E, Mahaisavariya, C, Boonyaprapa, W, et al. Development and internal validation of a machine-learning-developed model for predicting 1-year mortality after fragility hip fracture. BMC Geriatr. (2022) 22:451. doi: 10.1186/s12877-022-03152-x,
88. Kanis, J, Johnell, O, De Laet, C, Jonsson, B, Oden, A, and Ogelsby, A. International variations in hip fracture probabilities: implications for risk assessment. J Bone Miner Res. (2002) 17:1237–44. doi: 10.1359/jbmr.2002.17.7.1237,
89. Hsieh, C, Zheng, K, Lin, C, Lu, L, Li, W, Chen, F, et al. Automated and precise bone mineral density prediction and fracture risk assessment using hip/lumbar spine plain radiographs via learning deep image signatures and correlations (2021). doi: 10.21203/rs.3.rs-371880/v1,
90. Ulivieri, F, Rinaudo, L, Messina, C, Piodi, L, Capra, D, Lupi, B, et al. Bone strain index predicts fragility fracture in osteoporotic women: an artificial intelligence-based study. Eur Radiol Exp. (2021) 5:47. doi: 10.1186/s41747-021-00242-0,
91. Kong, S, Kim, S, Kim, Y, Kim, J, Kim, K, and Shin, C. Development and validation of common data model-based fracture prediction model using machine learning algorithm. Osteoporos Int. (2023) 34:1437–51. doi: 10.1007/s00198-023-06787-7,
92. Dong, Q, Luo, G, Lane, N, Lui, L, Marshall, L, Johnston, S, et al. Generalizability of deep learning classification of spinal osteoporotic compression fractures on radiographs using an adaptation of the modified-2 algorithm-based qualitative criteria. Acad Radiol. (2023) 30:2973–87. doi: 10.1016/j.acra.2023.04.023,
93. Hu, X, Zhu, Y, Qian, Y, Huang, R, Yin, S, Zeng, Z, et al. Cover picture: prediction of subsequent osteoporotic vertebral compression fracture on ct radiography via deep learning (view 6/2022). VIEW. (2022) 3. doi: 10.1002/VIW.20220012
94. Kim, Y, Kim, Y, Park, J, Kim, B, Shin, Y, Kong, S, et al. A ct-based deep learning model for predicting subsequent fracture risk in patients with hip fracture. Radiology. (2024) 310:e230614. doi: 10.1148/radiol.230614,
95. Wu, Q, Nasoz, F, Jung, J, Bhattarai, B, and Han, M. Machine learning approaches for fracture risk assessment: a comparative analysis of genomic and phenotypic data in 5130 older men. Calcif Tissue Int. (2020) 107:353–61. doi: 10.1007/s00223-020-00734-y,
96. Dong, Q, Luo, G, Lane, N, Lui, L, Marshall, L, Kado, D, et al. Deep learning classification of spinal osteoporotic compression fractures on radiographs using an adaptation of the genant semiquantitative criteria. Acad Radiol. (2022) 29:1819–32. doi: 10.1016/j.acra.2022.02.020,
97. Lei, C, Song, J, Li, S, Zhu, Y, Liu, M, Wan, M, et al. Advances in materials-based therapeutic strategies against osteoporosis. Biomaterials. (2023) 296:122066. doi: 10.1016/j.biomaterials.2023.122066,
98. Anish, R, and Nair, A. Osteoporosis management-current and future perspectives – a systemic review. J Orthop. (2024) 53:101–13. doi: 10.1016/j.jor.2024.03.002,
99. Zou, Q, and Ma, Q. The application of machine learning to disease diagnosis and treatment. Math Biosci. (2020) 320:108305. doi: 10.1016/j.mbs.2019.108305,
100. Orimo, H, Nakamura, T, Hosoi, T, Iki, M, Uenishi, K, Endo, N, et al. Japanese 2011 guidelines for prevention and treatment of osteoporosis—executive summary. Arch Osteoporos. (2012) 7:3–20. doi: 10.1007/s11657-012-0109-9
101. Tanphiriyakun, T, Rojanasthien, S, and Khumrin, P. Bone mineral density response prediction following osteoporosis treatment using machine learning to aid personalized therapy. Sci Rep. (2021) 11:13811. doi: 10.1038/s41598-021-93152-5,
102. Bonaccorsi, G, Giganti, M, Nitsenko, M, Pagliarini, G, Piva, G, and Sciavicco, G. Predicting treatment recommendations in postmenopausal osteoporosis. J Biomed Inform. (2021) 118:103780. doi: 10.1016/j.jbi.2021.103780,
103. Lin, Y, Chu, C, Hung, K, Lu, C, Bednarczyk, E, and Chen, H. Can machine learning predict pharmacotherapy outcomes? an application study in osteoporosis. Comput Methods Prog Biomed. (2022) 225:107028. doi: 10.1016/j.cmpb.2022.107028,
104. Wang, S, Zhu, J, Liu, W, and Liu, A. A machine learning framework for screening plasma cell-associated feature genes to estimate osteoporosis risk and treatment vulnerability. Biochem Genet. (2025) 63:3117–38. doi: 10.1007/s10528-024-10861-y,
105. Atik, O. Artificial intelligence, machine learning, and deep learning in orthopedic surgery. Joint Diseases and Related Surgery. (2022) 33:484–5. doi: 10.52312/jdrs.2022.57906,
106. Ma, Y, Lu, Q, Yuan, F, and Chen, H. Comparison of the effectiveness of different machine learning algorithms in predicting new fractures after pkp for osteoporotic vertebral compression fractures. J Orthop Surg Res. (2023) 18:62. doi: 10.1186/s13018-023-03551-9,
107. Dong, S, Zhu, J, Yang, H, Huang, G, Zhao, C, and Yuan, B. Development and internal validation of supervised machine learning algorithm for predicting the risk of recollapse following minimally invasive kyphoplasty in osteoporotic vertebral compression fractures. Front Public Health. (2022) 10:874672. doi: 10.3389/fpubh.2022.874672,
108. Messina, C, Piodi, L, Grossi, E, Eller-Vainicher, C, Bianchi, M, Ortolani, S, et al. Artificial neural network analysis of bone quality dxa parameters response to teriparatide in fractured osteoporotic patients. PLoS One. (2020) 15:e0229820. doi: 10.1371/journal.pone.0229820,
Glossary
ADMET - absorption distribution metabolism excretion
AI - artificial intelligence
ANN - artificial neural network
ANXA1 - Annexin A1
AUC - area under the curve
AUPRC - area under the precision-recall curve
BMD - bone mineral density
BPX - biplanar X-ray images of the pelvis
BSI - bone strain index
CASP3 - Cysteine aspartic acid protease 3
CNN - convolutional neural networks
CT - computerized tomography
CTSK - Cathepsin K
DL - deep learning
HDLF - hybrid deep learning framework
DLEPS - deep leaning-based therapeutic effect prediction system
DNN - deep neural network
DXA - dual-energy x-ray absorptiometry
EDCs - Endocrine disrupting chemicals
ERS - endoplasmic reticulum stress
ERAP2 - endoplasmic reticulum aminopeptidase 2
FCoTNet - Frequency Channel-Wise Transformer Network
FOXO3 - Forkhead box protein O3
FRAX - Fracture Risk Assessment Tool
GA2M - generalized additive model
GBM - gradient boosting machines
GNN - graph neural network
HMOX1 - Heme oxygenase 1
IBD - inflammatory bowel disease
IOF - international osteoporosis foundation
HSD17B2 - Estradiol 17-beta-dehydrogenase 2
LASSO - least absolute shrinkage and selection operator
LR - logistic regression
Light GBM - Light Gradient Boosting Machine
MAE - mean absolute error
MAPK - Mitogen activated protein kinases
MD - mitochondrial dysfunction
MIF - Macrophage migration inhibitory factor
ML - machine learning
MMP12 - Matrix metalloproteinase 12
MR - Mendelian randomization
MRI - magnetic resonance imaging
OLF - ossification of ligamentum flavum
OP - osteoporosis
OPF - osteoporotic fracture
PANK2 - Pantothenate kinase 2
PCD - programmed cell death
PFAS - Perfluoroalkyl sulfonate
PFOA - Perfluorooctanoic acid
PFOS - Perfluorooctane sulfonates
PKP - percutaneous kyphoplasty
PMOP - postmenopausal osteoporosis
P-SAMPNN - pre-trained self-attentive message passing neural network
RA - rheumatoid arthritis
RF - random forest
RFE - recursive feature elimination
ROC curve - receiver operator characteristic curve
SNF - similarity network fusion
SOP - senile osteoporosis
SOST - sclerostin
SVM - support vector machine
TBS - trabecular bone score
UKB - UK Biobank
WGCNA - weighted gene co-expression network analysis
XG-Boost - eXtreme Gradient Boosting.
Keywords: artificial intelligence, deep learning, interdisciplinarity, osteoporosis, bone metabolism, precision medicine
Citation: Fan S-T, Lu M, Dong J-L, Li Y-L, Hao L-N, Dong R-C and Hou M-D (2025) Application of artificial intelligence in osteoporosis: a review. Front. Med. 12:1718554. doi: 10.3389/fmed.2025.1718554
Edited by:
Chunyou Wan, Tianjin University, ChinaReviewed by:
Poonam Raturi, Graphic Era Deemed to be University, IndiaSoheil Shahbazi, University of California, Los Angeles, United States
Rachasak Somyanonthanakul, Thammasat University - Rangsit Campus, Thailand
Copyright © 2025 Fan, Lu, Dong, Li, Hao, Dong and Hou. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Ming-Dong Hou, aG91bWluZ2RvbmdAc2RqdHUuZWR1LmNu; Ren-Chao Dong, ZHJjaGFvMTAxQDE2My5jb20=