Radiomics analysis of lumbar spine X-ray images for diagnosing facet joint osteoarthritis: a two-center study

Radiomics analysis of lumbar spine X-ray images for diagnosing facet joint osteoarthritis: a two-center study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Radiomics analysis of lumbar spine X-ray images for diagnosing facet joint osteoarthritis: a two-center study Lezhen Jiang, Yifan Guo, Zhichao Sun, Yuanzhe Li, Haonan Zhu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7515717/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 6 You are reading this latest preprint version Abstract Objectives This study aims to develop and validate radiomics models utilizing lumbar spine X-rays for the early identification of facet joint osteoarthritis (FJOA). Methods This retrospective two-center study enrolled 1,997 patients who underwent paired lumbar X-ray and CT imaging within one month. Data from one center were used for model training and validation, and data from the other center were used for external testing. Radiomic features were extracted from manually segmented facet joint regions on X-rays. Key features selected through the least absolute shrinkage and selection operator (LASSO) were used to develop models, specifically logistic regression, linear support vector classification (LinearSVC), and support vector machines (SVM). The model performance was primarily evaluated using the area under the ROC curve (AUC) and the area under the precision-recall curve (AUPRC). Results A total of 20 features were selected for modeling. The logistic regression model based on radiomic features demonstrated the highest AUC. In the external testing cohort, this model achieved an AUC of 0.971 (95% CI: 0.956–0.986), a sensitivity of 98.0%, a specificity of 75.0%, and an AUPRC of 0.839. It outperformed both the SVM model (AUC = 0.946, AUPRC = 0.793) and the LinearSVC model (AUC = 0.966, AUPRC = 0.813). Conclusion Radiomics models based on lumbar X-rays showed robust performance and hold promise as a non-invasive, accessible tool for early and accurate identification of FJOA, potentially enabling timely intervention. Lumbar facet joints Osteoarthritis Radiomics Radiography Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Low back pain is a prevalent musculoskeletal disorder, and its global burden is increasing. By 2050, an estimated 843 million individuals worldwide will be affected, representing a 36.4% increase compared to 2020 [ 1 ]. Facet joint osteoarthritis (FJOA) is a prevalent cause of chronic low back pain and spinal dysfunction. Previous studies indicate that approximately 15–45% of patients with low back pain experience pain attributed to facet joint lesions [ 2 , 3 ]. In cases of severe lumbar FJOA associated with significant pain and functional impairment, surgical intervention may be required [ 4 – 6 ]. The clinical manifestations of lumbar FJOA lack specificity, and its symptomatology significantly overlaps with that of disc degenerative disease and lumbar spinal stenosis [ 7 ]. Furthermore, the diagnosis lacks standardized quantitative parameters, which results in limited clinical assessment and a high rate of misdiagnosis [ 8 , 9 ]. Currently, diagnosis primarily depends on imaging examinations and clinical physical assessments. Conventional X-ray examinations exhibit low sensitivity for early lesions, and the assessment criteria lack quantifiability [ 10 ]. MRI is not the preferred modality for assessing FJOA due to its high cost and lengthy examination time [ 11 , 12 ].In contrast, CT is the imaging modality of choice for diagnosing FJOA and evaluating its severity, owing to its superior resolution of bony structures and clear visualization of osseous changes [ 10 ]. However, the relatively high radiation dose and cost associated with CT limit its utility as a first-line screening tool for large populations or patients presenting with low back pain [ 13 ]. The current manual image interpretation model struggles to meet the demands of precise clinical diagnosis and treatment. This inadequacy may result in delays in optimal treatment timing for some patients due to assessment bias. Radiomics method developed predictive models for diseases by extracting high-throughput quantitative features from medical images and integrating them with machine learning algorithms [ 14 ]. Recent studies have demonstrated that the application of radiomics to lumbar musculoskeletal imaging can significantly enhance the accuracy of disease prediction [ 15 – 17 ]. However, there exists a notable gap in radiomics research concerning FJOA, particularly in the absence of a standardized system for automated analysis based on the interpretation of X-ray images. This study aims to develop and test a radiomics-based screening model for lumbar FJOA by quantitatively analyzing texture, morphological, and structural features from X-ray images, thereby providing an objective and reproducible assessment tool and exploring new pathways for early diagnosis. Methods Study Participants Approval for this research was secured from the institutional review board in accordance with the Declaration of Helsinki, and the requirement for informed consent was exempted due to the study's retrospective design. This retrospective study included patients from two medical centers who underwent both lumbar spine X-ray and CT scans within a one-month interval between January 2019 and December 2024. The exclusion criteria are as follows: (1) Clinically diagnosed cases of ankylosing spondylitis, infectious arthritis, systemic metabolic joint diseases, traumatic joint disorders, and spinal tumors; (2) Structural abnormalities resulting from anatomical deformities, such as scoliosis and hemivertebrae; (3) Fractures or postoperative changes that impair visualization of the facet joint area; (4) Poor visualization of the lumbar facet joints due to artifacts or abnormal occlusion. Each facet joint of the vertebrae was defined as an independent sample unit. A stratified random sampling method was employed to divide the dataset from Center I into a training dataset and a validation dataset in a 7:3 ratio. The dataset from the Center II served as an external test dataset (Fig. 1). Image Acquisition Lumbar spine anteroposterior X-ray imaging was conducted using the uDR780i (UIH, Shanghai, China) and four Digital Diagnost DR systems (Philips, Hamburg, Germany; Philips, Amsterdam, Netherlands). Automatic tube voltage and tube current are used for lumbar spine X-rays. FJOA classification This study utilized an independent blinded assessment design, wherein a young physician with five years of experience in musculoskeletal imaging and a senior expert with fifteen years of experience in the same field independently evaluated the degree of degeneration of bilateral facet joints in the L2/3-L4/5 segments. Both readers were blinded to clinical diagnosis and to each other's assessments. The assessment was conducted based on the Weishaupt criteria using thin-slice CT images [18]. The resulting lumbar spine CT classifications served as the reference standard for subsequent spatial registration and classification of the X-ray image annotation results. In this study, the lumbar facet joint samples from Weishaupt grade 3 were designated as the positive group, referred to as the G3 group. Conversely, the samples from Weishaupt grade 0 to 2 were designated as the negative group, referred to as the G0-2 group. In cases of disagreement regarding the Weishaupt classification, resolution was achieved through negotiation. Image Segmentation Image processing was conducted using ITK-SNAP software (version 3.8.0, http://www.itksnap.org). A young radiologist completed the segmentation of the region of interest (ROI), which was subsequently reviewed and corrected by an experienced radiology expert. Manual segmentation was performed on the lumbar spine X-ray images, with the ROI encompassing the bilateral facet joints of the L2/3-L4/5 segments. On the X-ray image, the ROI was manually delineated as follows: the upper and lower boundaries were defined by the cranial and caudal margins of the superior articular process, respectively. The medial and lateral boundaries were defined by the medial margin of the inferior articular process and the lateral margin of the superior articular process, respectively. The shadow of the pedicle was excluded from the segmentation region. Radiomic Features Extraction In this study, the extraction of radiomic features was conducted using the Shukun Technology research platform (https://medresearch.shukun.net/). Prior to radiomic feature extraction, standardized preprocessing was applied to the X-ray images. All images were resampled to a uniform pixel spacing of 1.0 × 1.0 mm using cubic spline interpolation. Intensity normalization was performed to minimize inter-image variations, and a fixed bin width of 25 was applied for discretizing the gray-level values. A total of 1,698 quantitative imaging features were extracted, which include first-order statistical features (n = 324), gray level co-occurrence matrix (GLCM, n = 432), gray level dependence matrix (GLDM, n = 252), gray level run length matrix (GLRLM, n = 288), gray level size zone matrix (GLSZM, n = 288), neighboring gray tone difference matrix (NGTDM, n = 90), and three-dimensional morphological features (n = 14). Machine Learning Models Development First, standardize and preprocess the omics feature datasets. Perform Pearson correlation analysis among all features. If the absolute value of the correlation coefficient between any feature and another feature is ≥0.9, remove that feature. Subsequently, feature compression is conducted using the least absolute shrinkage and selection operator (LASSO) regression model, which is set to a maximum of 3000 iterations and a convergence threshold of 0.0001. This study employs three machine learning algorithms to develop diagnostic models, specifically logistic regression, linear support vector classification (LinearSVC), and support vector machines (SVM). To mitigate overfitting and enhance model generalizability, a 10-fold cross-validation strategy was employed throughout model development and evaluation. The workflow of this research is illustrated in Fig. 2. Statistical Analyses All statistical analyses were performed using SPSS (version 25.0; IBM Corp) and R language (version 4.1.2; R Project, https://www.r-project.org). Continuous variables were described as mean ± standard deviation, determined based on the results of the Shapiro-Wilk normality test. Categorical variables were presented in frequency form, and inter-group comparisons were conducted using the χ² test or Fisher's exact test. The predictive models were comprehensively evaluated using the area under the receiver operating characteristic curve (AUC), alongside indicators such as sensitivity, specificity, and accuracy. Additionally, positive predictive value (PPV) and negative predictive value (NPV) were calculated to enhance clinical interpretability. To provide a more comprehensive evaluation of model performance under class imbalance, precision-recall curves (PRC) were additionally plotted. Calibration curves were also generated to assess the calibration of predicted probabilities. A p -value < 0.05 was considered statistically significant. Results Clinical Features This study included a total of 1,997 patients, with an average age of 57.35 ± 16.86 years, consisting of 981 males and 1,026 females. A total of 7,142 lumbar facet joints were analyzed. The facet joints of 1,720 patients from the Center I were stratified and randomly allocated into a training dataset (n = 4,470) and a validation dataset (n = 1,917). The external test dataset comprised facet joints from 277 patients at the Center II (n = 755). The clinical characteristics of the three datasets are presented in Table 1 . No significant differences were observed in the distributions of age and gender among the datasets. The proportion of G3 group samples in the internal training dataset, internal validation dataset, and external test dataset were 20.31% (895/4407), 20.03% (384/1917), and 13.24% (100/755), respectively. Table 1 Baseline in Training, Validation, and External Test Datasets Characteristics Training (n = 1202) Validation (n = 518) External Test (n = 277) p value Gender Female 622 (51.7%) 265 (51.2%) 139 (50.2%) 0.94 Male 580 (48.3%) 253 (48.8%) 138 (49.8%) Age [years] 57.41 ± 16.93 57.74 ± 16.74 56.10 ± 15.98 0.20 N(%), mean ± SD Development of radiomics Models Following standardized preprocessing, 1,697 radiomic features were initially extracted. Features exhibiting high inter-feature correlation with an absolute Pearson correlation coefficient ≥ 0.9 were removed, yielding 281 candidate features. Subsequently, LASSO regression was performed to select key features, resulting in 20 radiomic features with non-zero coefficients. The selected features and their corresponding coefficients in three models are presented in Figure S1 . To develop a high-performance classification model, three classic machine learning algorithms were systematically evaluated: logistic regression, LinearSVC, and SVM. The results indicate that all three models can accurately predict lumbar FJOA (Table 2 ), with the SVM model demonstrating superior performance in both the training and validation datasets (Fig. 3 ). Its ROC curve AUCs are 0.915 (95% CI: 0.905–0.924) and 0.862 (95% CI: 0.841–0.882), which surpass those of the logistic regression model and the LinearSVC model. Specifically, the AUCs for the logistic regression model in the training and validation sets are 0.875 (95% CI: 0.862–0.888) and 0.851 (95% CI: 0.830–0.872), respectively; the AUCs for the LinearSVC model are 0.875 (95% CI: 0.862–0.887) and 0.853 (95% CI: 0.833–0.874), respectively. Table 2 Predictive Performance of Each Model Model Dataset AUC 95% CI ACC SEN SPE PPV NPV Logistic Regression Training 0.875 0862–0.888 0.793 0.811 0.788 0.489 0.932 Logistic Regression Validation 0.851 0.851–0.872 0.764 0.779 0. 760 0.448 0.943 Logistic Regression External Test 0.971 0.956–0.986 0.780 0.980 0.750 0.374 0.996 SVM Training 0.915 0.905–0.924 0.845 0.837 0.847 0.578 0.954 SVM Validation 0.862 0.841–0.882 0.803 0.750 0.816 0.505 0.929 SVM External Test 0.946 0.925–0.968 0.840 0.900 0.831 0.448 0.982 LinearSVC Training 0.875 0.862–0.877 0.797 0.806 0.795 0.496 0.942 LinearSVC Validation 0.853 0.833–0.874 0.772 0.781 0.769 0.459 0.934 LinearSVC External Test 0.966 0.949–0.982 0.805 0.980 0.779 0403 0.996 AUC area under the curve, CI confidence interval, SEN Sensitivity, SPE Specificity, NPV negative predictive value, PPV positive predictive value. External test of machine learning models In the external test dataset, the logistic regression model exhibited superior diagnostic performance, achieving an AUC of 0.971 (95% CI: 0.956–0.986). This performance surpassed that of the SVM model, which recorded an AUC of 0.946 (95% CI: 0.925–0.968), and the LinearSVC model, which achieved an AUC of 0.966 (95% CI: 0.949–0.982). Further analysis of performance metrics revealed that the logistic regression model achieved a diagnostic specificity of 75.0%, a sensitivity of 98.0%, and an overall accuracy of 78.0% for G3 FJOA in the external test set. As illustrated in Fig. 4 , the PR curve analysis based on the external test set indicates that the curve distribution of the logistic regression model significantly approaches the upper right area. Quantitative evaluation reveals that the AUCPR of the logistic regression model in the training set is 0.839, which surpasses that of the SVM model with an AUCPR of 0.793 and the LinearSVC model with an AUCPR of 0.813. Figure S2 displays the calibration curves for three datasets, demonstrating that the all models fit reasonably well. Discussion This study successfully developed a radiomics diagnostic model for FJOA using lumbar spine X-rays. The results indicated that the radiomics model based on the logistic regression algorithm demonstrated the best performance, achieving an AUC of approximately 0.971 in the external test dataset. While the specificity of the logistic regression model was relatively low at 75.0%, it exhibited high sensitivity at 98.0%, thereby fulfilling the screening requirements for FJOA patients based on lumbar spine X-rays. The imaging assessment of traditional lumbar FJOA primarily depends on manual interpretation of X-rays, CT, and MRI [ 4 , 10 ]. CT exhibits high sensitivity in detecting subtle bone changes, attributed to its advantage of three-dimensional visualization. However, the associated radiation dose and cost restrict its use as a first-line screening tool [ 10 , 19 ]. MRI may underestimate the bony structural changes associated with FJOA [ 20 , 21 ], and the consistency with X-ray assessments is only moderate [ 22 ]. In recent years, some scholars have proposed the use of single photon emission computed tomography and hybrid imaging technology to assist in clinical diagnosis and treatment; however, their feasibility still requires further research [ 23 , 24 ]. Invasive facet joint injections, regarded as the gold standard for diagnosis, carry risks of complications and are time-consuming[ 25 , 26 ]. Characterized by its low cost, controllable radiation dose, high clinical prevalence, and superior spatial resolution, X-ray imaging provides distinct advantages in visualizing the bony structures of the lumbar spine, thereby establishing itself as an essential tool for routine orthopedic screening and diagnosis [ 27 , 28 ]. This study proposes the use of CT as an anatomical validation reference standard, aiming to mitigate the false positive interference caused by tissue overlap in X-rays by establishing a cross-modal feature mapping relationship between X-rays and CT. This method not only enhances the specificity of X-ray imaging in FJOA screening but also provides a more interpretable feature set for the subsequent development of automated diagnostic models based on X-rays. The radiomics model based on lumbar spine X-rays developed in this study enables early identification of FJOA. Positive cases can be referred for CT confirmation or MR soft tissue evaluation, thereby establishing a hierarchical diagnosis and treatment system. This approach not only optimizes the allocation of medical resources but also encourages timely intervention for early-stage patients, effectively delaying disease progression. Radiomics can quantify the heterogeneity of bone microstructure through the high-throughput extraction of texture features, morphological parameters, and gray-scale distribution information from X-ray images [ 29 , 30 ]. In the field of lumbar spine radiomics, studies have successfully developed diagnostic models utilizing X-ray features, thereby confirming the clinical value of this technology [ 31 , 32 ]. However, there remains a substantial gap in radiomics research concerning lumbar FJOA. Current methodologies predominantly depend on conventional imaging morphological parameters, which exhibit a weak correlation with the severity of pain [ 33 ]. Furthermore, traditional visual assessments exhibit low inter-observer consistency due to the overlapping characteristics present in X-ray imaging, which complicates the attainment of precise grading [ 22 ]. Therefore, it is imperative to develop more objective and quantifiable diagnostic tools for lumbar FJOA. This study systematically evaluated the performance of logistic regression model, SVM model, and LinearSVC model, and determining the efficacy of these algorithms in screening for lumbar FJOA. The results indicate that while the SVM model exhibited exceptional performance on the internal training and validation datasets, the logistic regression model displayed markedly superior diagnostic capabilities in the external test dataset. Consequently, the logistic regression model is identified as the optimal model developed in this study. However, it is noteworthy that the specificity of the logistic regression model in the external test set is relatively low. This may be attributed to the fact that, although radiomics can develop high-performance predictive models by extracting quantitative features such as texture and morphology from X-ray images using high-throughput methods, the inherent anatomical structure projection overlap on lumbar spine X-ray films limits its ability to clearly distinguish subtle bony structure changes, resulting in an increased false-positive prediction rate. Furthermore, given the low proportion of positive samples in the dataset of this study, and the fact that AUPRC is a more sensitive evaluation metric for minority class samples, it becomes particularly significant in such imbalanced scenarios. The superior AUPRC performance of the logistic regression model, which confirms its effectiveness in identifying high-risk FJOA, underscores the necessity of concurrently evaluating both AUC and AUPRC in scenarios characterized by class imbalance. Limitation However, this study has several limitations. Firstly, the absence of facet joint injection, which is regarded as the gold standard for diagnosing FJOA, introduces potential subjective bias in the manual grouping of positive and negative samples, which relied solely on CT-based Weishaupt grading. Secondly, anatomical overlap in lumbar spine X-rays may compromise facet joint ROI delineation, particularly at superior and inferior margins, due to spinal curvature variations or inconsistent positioning, ultimately limiting reproducibility. Conclusion This study demonstrates that radiomics facilitates the accurate identification of FJOA. The developed model effectively screens for FJOA in patients undergoing routine lumbar radiography, enabling a rapid determination of whether FJOA contributes to the etiology of their low back pain. This provides objective imaging support for orthopedic decision-making concerning further diagnostic or therapeutic interventions. Abbreviations ACC Accuracy AUC Area under the curve AUPRC Area under the precision-recall curve CI Confidence interval CT Computed tomography FJOA Facet joint osteoarthritis GLCM Gray level co-occurrence matrix GLDM Gray level dependence matrix GLRLM Gray level run length matrix GLSZM Gray level size zone matrix LASSO Least absolute shrinkage and selection operator LinearSVC Linear support vector classification MRI Magnetic resonance imaging NGTDM Neighboring gray tone difference matrix NPV Negative predictive value PPV Positive predictive value PRC Precision-recall curve ROC Receiver operating characteristic ROI Region of interest SEN Sensitivity SPE Specificity SVM Support vector machines Declarations Ethics approval and consent to participate This study was approved by the Institutional Review Board of the Second Affiliated Hospital of Fujian Medical University, China (No. 2024-519) in accordance with the Declaration of Helsinki. Due to the retrospective nature of the study, the requirement for written informed consent was waived by the same ethics committee. Consent for publication Not applicable. Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests. Funding This study has received funding by Joint funds for the innovation of science and technology,Fujian Province（No:2024Y9388）. Authors' contributions All authors contributed to the study's conception and design. Material preparation, data collection, and analysis were performed by L.Z.J., Y.F.G. and Z.C.S. The first draft of the manuscript was written by L.Z.J. and Y.F.G. Y.Z.L. and H.N.Z. supervised the project and were responsible for funding acquisition. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Acknowledgements None. References Global, regional, and national burden of low back pain, 1990-2020, its attributable risk factors, and projections to 2050: a systematic analysis of the Global Burden of Disease Study 2021. Lancet Rheumatol 2023, 5(6):e316-e329. Du R, Gao J, Wang B, Zhang J, Meng M, Wang J, Qu W, Li Z: Percutaneous radiofrequency ablation and endoscopic neurotomy for lumbar facet joint syndrome: are they good enough? Eur Spine J 2024, 33(2):463-473. Abd-Elsayed A, Azeem N, Chopra P, D'Souza RS, Sayed D, Deer T: An International Survey on the Practice of Lumbar Radiofrequency Ablation for Management of Zygapophyseal (Facet)-Mediated Low Back Pain. J Pain Res 2022, 15:1083-1090. Perolat R, Kastler A, Nicot B, Pellat JM, Tahon F, Attye A, Heck O, Boubagra K, Grand S, Krainik A: Facet joint syndrome: from diagnosis to interventional management. Insights Imaging 2018, 9(5):773-789. Gajjar AA, Dombrovsky DA, Singh R, Stonnington HO, Koester SW, Polavarapu H, George DD, Wakim A, Harland T, Mansfield K: Trends in Spinal Pain Procedure Volumes and Reimbursements: An Analysis of 20 Years of Medicare Data. Pain Pract 2025, 25(5):e70043. Park S, Park JH, Sokpeou N, Jang JN, Kim YU, Choi YS, Park S: Radiofrequency treatments for lumbar facet joint syndrome: a systematic review and network meta-analysis. Reg Anesth Pain Med 2024. Anaya JEC, Coelho SRN, Taneja AK, Cardoso FN, Skaf AY, Aihara AY: Differential Diagnosis of Facet Joint Disorders. Radiographics 2021, 41(2):543-558. Martins DE, Astur N, Kanas M, Ferretti M, Lenza M, Wajchenberg M: Quality assessment of systematic reviews for surgical treatment of low back pain: an overview. Spine J 2016, 16(5):667-675. Yoo YM, Kim KH: Facet joint disorders: from diagnosis to treatment. Korean J Pain 2024, 37(1):3-12. Du R, Xu G, Bai X, Li Z: Facet Joint Syndrome: Pathophysiology, Diagnosis, and Treatment. J Pain Res 2022, 15:3689-3710. Berg L, Thoresen H, Neckelmann G, Furunes H, Hellum C, Espeland A: Facet arthropathy evaluation: CT or MRI? Eur Radiol 2019, 29(9):4990-4998. Thiel CL, Vigil-Garcia M, Nande S, Meijer C, Gehrels J, Struk O, Thornander S, Pullella D, Omary RA, Carver DE et al : Environmental Life Cycle Assessment of a U.S. Hospital-based Radiology Practice. Radiology 2024, 313(2):e240398. De Wilde D, Zanier O, Da Mutten R, Jin M, Regli L, Serra C, Staartjes VE: Strategies for generating synthetic computed tomography-like imaging from radiographs: A scoping review. Med Image Anal 2025, 101:103454. Gillies RJ, Kinahan PE, Hricak H: Radiomics: Images Are More than Pictures, They Are Data. Radiology 2016, 278(2):563-577. Nian S, Zhao Y, Li C, Zhu K, Li N, Li W, Chen J: Development and validation of a radiomics-based model for predicting osteoporosis in patients with lumbar compression fractures. Spine J 2024, 24(9):1625-1634. Zhang B, Chen Z, Yan R, Lai B, Wu G, You J, Wu X, Duan J, Zhang S: Development and Validation of a Feature-Based Broad-Learning System for Opportunistic Osteoporosis Screening Using Lumbar Spine Radiographs. Acad Radiol 2024, 31(1):84-92. Wu Z, Wang H, Zheng Y, Fei H, Dong C, Wang Z, Ren W, Xu W, Bian T: Lumbar MR-based radiomics nomogram for detecting minimal residual disease in patients with multiple myeloma. Eur Radiol 2023, 33(8):5594-5605. Weishaupt D, Zanetti M, Boos N, Hodler J: MR imaging and CT in osteoarthritis of the lumbar facet joints. Skeletal Radiol 1999, 28(4):215-219. Saremi A, Goyal KK, Benzel EC, Orr RD: Evolution of lumbar degenerative spondylolisthesis with key radiographic features. Spine J 2024, 24(6):989-1000. Little JW, Grieve T, Cantu J, Bogar WC, Heiser R, Miley H, Cramer GD: 20 Reliability of Human Lumbar Facet Joint Degeneration Severity Assessed by Magnetic Resonance Imaging. J Manipulative Physiol Ther 2020, 43(1):43-49. Lacroix M, Nguyen C, Burns R, Laporte A, Rannou F, Feydy A: Degenerative Lumbar Spine Disease: Imaging and Biomechanics. Semin Musculoskelet Radiol 2022, 26(4):424-438. Liu X, Zhu Z, He S, Sun M, Zhao T, Liu L, Shi H, Hou Y, Shi G: Enhancing Clinical Insights: New Radiographic Grading for Lumbar Facet Joint Degeneration, A Reliability Study. JOR Spine 2025, 8(1):e70035. Yolcu YU, Lehman VT, Bhatti AUR, Goyal A, Alvi MA, Bydon M: Use of Hybrid Imaging Techniques in Diagnosis of Facet Joint Arthropathy: A Narrative Review of Three Modalities. World Neurosurg 2020, 134:201-210. Pendleton J, Ng A: SPECT/CT Scan: A New Diagnostic Tool in Pain Medicine. Curr Pain Headache Rep 2023, 27(11):729-735. Van Oosterwyck W, Vander Cruyssen P, Castille F, Van de Kelft E, Decaigny V: Lumbar Facet Joint Disease: What, Why, and When? Life (Basel) 2024, 14(11). Busse JW, Genevay S, Agarwal A, Standaert CJ, Carneiro K, Friedrich J, Ferreira M, Verbeke H, Brox JI, Xiao H et al : Commonly used interventional procedures for non-cancer chronic spine pain: a clinical practice guideline. Bmj 2025, 388:e079970. Pina Vegas L, van Lunteren M, Loeuille D, Morizot C, Newsum E, Ramiro S, van Gaalen F, Saraux A, Claudepierre P, Feydy A et al : Ten-year follow-up of degenerative spinal lesions on radiographs and MRI in axial spondyloarthritis: results of the DESIR (DEvenir des spondylarthropathies indifférenciées récentes) cohort. Eur Radiol 2025. McCormick RJ, Perloff MD: Xray prediction of MRI in low back pain. Am J Phys Med Rehabil 2025. Gitto S, Annovazzi A, Nulle K, Interlenghi M, Salvatore C, Anelli V, Baldi J, Messina C, Albano D, Di Luca F et al : X-rays radiomics-based machine learning classification of atypical cartilaginous tumour and high-grade chondrosarcoma of long bones. EBioMedicine 2024, 101:105018. von Schacky CE, Wilhelm NJ, Schäfer VS, Leonhardt Y, Jung M, Jungmann PM, Russe MF, Foreman SC, Gassert FG, Gassert FT et al : Development and evaluation of machine learning models based on X-ray radiomics for the classification and differentiation of malignant and benign bone tumors. Eur Radiol 2022, 32(9):6247-6257. Galbusera F, Cina A, O'Riordan D, Vitale JA, Loibl M, Fekete TF, Kleinstück F, Haschtmann D, Mannion AF: Estimating lumbar bone mineral density from conventional MRI and radiographs with deep learning in spine patients. Eur Spine J 2024, 33(11):4092-4103. Cheng L, Cai F, Xu M, Liu P, Liao J, Zong S: A diagnostic approach integrated multimodal radiomics with machine learning models based on lumbar spine CT and X-ray for osteoporosis. J Bone Miner Metab 2023, 41(6):877-889. Hu J, Zhang Y, Duan C, Peng X, Hu P, Lu H: Feasibility study for evaluating early lumbar facet joint degeneration using axial T(1) ρ, T(2) , and T2* mapping in cartilage. J Magn Reson Imaging 2017, 46(2):468-475. Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterials.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 12 Sep, 2025 Reviewers agreed at journal 12 Sep, 2025 Reviewers invited by journal 11 Sep, 2025 Editor assigned by journal 09 Sep, 2025 Submission checks completed at journal 09 Sep, 2025 First submitted to journal 02 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7515717","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":514959516,"identity":"18114705-a37b-4f08-a970-6275431d436b","order_by":0,"name":"Lezhen Jiang","email":"","orcid":"","institution":"The Second Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Lezhen","middleName":"","lastName":"Jiang","suffix":""},{"id":514959517,"identity":"225fe783-0c95-4820-aba1-1a513e718d59","order_by":1,"name":"Yifan Guo","email":"","orcid":"","institution":"The First Affiliated Hospital of Zhejiang Chinese Medical University, Zhejiang Provincial Hospital of Traditional Chinese Medicine)","correspondingAuthor":false,"prefix":"","firstName":"Yifan","middleName":"","lastName":"Guo","suffix":""},{"id":514959518,"identity":"b437cb3b-de02-4a4f-811d-53a067149f7f","order_by":2,"name":"Zhichao Sun","email":"","orcid":"","institution":"The First Affiliated Hospital of Zhejiang Chinese Medical University (Zhejiang Provincial Hospital of Traditional Chinese Medicine)","correspondingAuthor":false,"prefix":"","firstName":"Zhichao","middleName":"","lastName":"Sun","suffix":""},{"id":514959519,"identity":"8a1c8e90-4480-4c27-9e7a-71353483e980","order_by":3,"name":"Yuanzhe Li","email":"","orcid":"","institution":"The Second Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yuanzhe","middleName":"","lastName":"Li","suffix":""},{"id":514959520,"identity":"98666db2-bdf7-45c7-bb7e-4d72283a3155","order_by":4,"name":"Haonan Zhu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAqUlEQVRIiWNgGAWjYDACCQbGB1CmAdFamGFKidfCJkGaFvnZvccqf9RsS2xgb94mwVBzh7AWxjnn0m5IHLud2MBzrEyC4dgzwlqYJXLMbhg2ALUAGRKMDYcJa2EDqixIBGmRf0OkFh6gFoaDYFt4iNQiIZFjLNlw7LZxG09asUXCMSK0yM/IMfz4o+a2bD/74Y03PtQQoQUO2EBEAgkaRsEoGAWjYBTgAQAoBzWG53gSKQAAAABJRU5ErkJggg==","orcid":"","institution":"The First Affiliated Hospital of Zhejiang Chinese Medical University, Zhejiang Provincial Hospital of Traditional Chinese Medicine)","correspondingAuthor":true,"prefix":"","firstName":"Haonan","middleName":"","lastName":"Zhu","suffix":""}],"badges":[],"createdAt":"2025-09-02 09:08:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7515717/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7515717/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91843584,"identity":"8c5f9302-d33d-49f1-adcd-341af5d6bc0f","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9637356,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/d9a74e1b97cb97757e1928c9.docx"},{"id":91843580,"identity":"7dedaa3d-3c6d-4a88-b1e8-fdbadcace509","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6740,"visible":true,"origin":"","legend":"","description":"","filename":"b2f8d286a2fd475b95a434a8aa5a0b2f.json","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/ab99effb37a9c8bb0152381d.json"},{"id":91843587,"identity":"e2da5ce4-c63c-4fb5-8b03-60b25c7210c5","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8612864,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/9fc242f4ffc95aed3c38ae63.docx"},{"id":91843588,"identity":"9758d3bd-8e72-47ac-8fb9-c170b3867f9c","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":100954,"visible":true,"origin":"","legend":"","description":"","filename":"b2f8d286a2fd475b95a434a8aa5a0b2f1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/78ed776e16adf6028c428bfd.xml"},{"id":91843586,"identity":"172a03ee-951c-486a-bbab-3fd4184db722","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":117885,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/f876871cdcb06f3aec34fea2.png"},{"id":91843597,"identity":"40df9b0d-a0c0-4450-87d9-6a8d851cb254","added_by":"auto","created_at":"2025-09-22 10:06:43","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3053810,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/7164e68e3b4688198728b147.jpeg"},{"id":91843599,"identity":"fea43061-6a82-4389-8fb3-dced89fdc208","added_by":"auto","created_at":"2025-09-22 10:06:43","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1556234,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/ee6a289568a8665ac9609996.jpeg"},{"id":91843592,"identity":"b611cef2-29f7-4b7f-9525-84b48db34d0d","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4814834,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/8544c1f4202df9e1ca0b4964.jpeg"},{"id":91845921,"identity":"ca23292e-6f51-4b02-a340-8dbc6a93c0cf","added_by":"auto","created_at":"2025-09-22 10:14:43","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":28745,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/0698ced1858e2fb46049f6a0.png"},{"id":91843590,"identity":"e2989974-aa9a-444a-a362-1e7464e2dc42","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":58773,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/48c3cb207e16b08780fe7fab.png"},{"id":91843595,"identity":"af91a5eb-f437-4632-8fd9-6e1ae81ce4be","added_by":"auto","created_at":"2025-09-22 10:06:43","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16703,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/8f0140fccc94cda0527baad9.png"},{"id":91843591,"identity":"5998b571-43ef-46c2-bc26-f0b7e3c4be11","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":27159,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/d52243392ec7c0c1ced1da36.png"},{"id":91843594,"identity":"0a623ece-29ad-4b83-8339-6f378e34572c","added_by":"auto","created_at":"2025-09-22 10:06:43","extension":"xml","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":99400,"visible":true,"origin":"","legend":"","description":"","filename":"b2f8d286a2fd475b95a434a8aa5a0b2f1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/57e355df19f1360aacfbf226.xml"},{"id":91845923,"identity":"b55f872e-5adf-4e23-9bb3-80a3e54d236c","added_by":"auto","created_at":"2025-09-22 10:14:43","extension":"html","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":107364,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/27e28b3eb8263016b8c201bd.html"},{"id":91843579,"identity":"65fbccf0-3aa6-42e5-b3ff-3283f6950840","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":126641,"visible":true,"origin":"","legend":"\u003cp\u003eSummary of lumber facet joints recruitment and exclusions.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/57a15da03f8e0183782e83ac.png"},{"id":91843583,"identity":"ee83cff7-810c-4928-9677-862942421945","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":443405,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow of radiomics analysis. ROI, region of Interest; LASSO, shrinkage and selection shrinkage and selection operator; ROC, receiver operating characteristic\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/1e1ea754e569fb71007d98a2.png"},{"id":91843581,"identity":"63011b19-751a-456a-ba54-b13ea3973e8f","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":74773,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver operating characteristic curves of Logistic regression model (A), LinearSVC model (B), and SVM model (C) in training, validation, and external test datasets. AUC, area under the curve; CI, confidence interval\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/6ed8196702cddc5680bfbe3e.png"},{"id":91843585,"identity":"54cdf954-4096-4015-bff7-3bae3ab83284","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":106359,"visible":true,"origin":"","legend":"\u003cp\u003ePrecision-recall curve of Logistic regression model, SVM model, and LinearSVC model in external test dataset. AUPRC, area under the precision-recall curve\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/2d95d835f6a4831e500ae287.png"},{"id":91847429,"identity":"04a270d2-f5be-4a3b-9995-a08642b116d5","added_by":"auto","created_at":"2025-09-22 10:22:43","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1360019,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/bd53b138-cfbb-4e70-8d67-9ee924ec6e42.pdf"},{"id":91843593,"identity":"87bc4eda-0ef2-46b5-a2fb-7a139be7d275","added_by":"auto","created_at":"2025-09-22 10:06:42","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":8612864,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-7515717/v1/3655c8cc74425c3a6ada826c.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Radiomics analysis of lumbar spine X-ray images for diagnosing facet joint osteoarthritis: a two-center study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eLow back pain is a prevalent musculoskeletal disorder, and its global burden is increasing. By 2050, an estimated 843\u0026nbsp;million individuals worldwide will be affected, representing a 36.4% increase compared to 2020 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Facet joint osteoarthritis (FJOA) is a prevalent cause of chronic low back pain and spinal dysfunction. Previous studies indicate that approximately 15\u0026ndash;45% of patients with low back pain experience pain attributed to facet joint lesions [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. In cases of severe lumbar FJOA associated with significant pain and functional impairment, surgical intervention may be required [\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe clinical manifestations of lumbar FJOA lack specificity, and its symptomatology significantly overlaps with that of disc degenerative disease and lumbar spinal stenosis [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Furthermore, the diagnosis lacks standardized quantitative parameters, which results in limited clinical assessment and a high rate of misdiagnosis [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Currently, diagnosis primarily depends on imaging examinations and clinical physical assessments. Conventional X-ray examinations exhibit low sensitivity for early lesions, and the assessment criteria lack quantifiability [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. MRI is not the preferred modality for assessing FJOA due to its high cost and lengthy examination time [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].In contrast, CT is the imaging modality of choice for diagnosing FJOA and evaluating its severity, owing to its superior resolution of bony structures and clear visualization of osseous changes [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. However, the relatively high radiation dose and cost associated with CT limit its utility as a first-line screening tool for large populations or patients presenting with low back pain [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. The current manual image interpretation model struggles to meet the demands of precise clinical diagnosis and treatment. This inadequacy may result in delays in optimal treatment timing for some patients due to assessment bias.\u003c/p\u003e\u003cp\u003eRadiomics method developed predictive models for diseases by extracting high-throughput quantitative features from medical images and integrating them with machine learning algorithms [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Recent studies have demonstrated that the application of radiomics to lumbar musculoskeletal imaging can significantly enhance the accuracy of disease prediction [\u003cspan additionalcitationids=\"CR16\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. However, there exists a notable gap in radiomics research concerning FJOA, particularly in the absence of a standardized system for automated analysis based on the interpretation of X-ray images.\u003c/p\u003e\u003cp\u003eThis study aims to develop and test a radiomics-based screening model for lumbar FJOA by quantitatively analyzing texture, morphological, and structural features from X-ray images, thereby providing an objective and reproducible assessment tool and exploring new pathways for early diagnosis.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eStudy Participants\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eApproval for this research was secured from the institutional review board in accordance with the Declaration of Helsinki, and the requirement for informed consent was exempted due to the study\u0026apos;s retrospective design. This retrospective study included patients from two medical centers who underwent both lumbar spine X-ray and CT scans within a one-month interval between January 2019 and December 2024. The exclusion criteria are as follows: (1) Clinically diagnosed cases of ankylosing spondylitis, infectious arthritis, systemic metabolic joint diseases, traumatic joint disorders, and spinal tumors; (2) Structural abnormalities resulting from anatomical deformities, such as scoliosis and hemivertebrae; (3) Fractures or postoperative changes that impair visualization of the facet joint area; (4) Poor visualization of the lumbar facet joints due to artifacts or abnormal occlusion. Each facet joint of the vertebrae was defined as an independent sample unit. A stratified random sampling method was employed to divide the dataset from Center I into a training dataset and a validation dataset in a 7:3 ratio. The dataset from the Center II served as an external test dataset (Fig. 1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImage Acquisition\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eLumbar spine anteroposterior X-ray imaging was conducted using the uDR780i (UIH, Shanghai, China) and four Digital Diagnost DR systems (Philips, Hamburg, Germany; Philips, Amsterdam, Netherlands). Automatic tube voltage and tube current are used for lumbar spine X-rays.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFJOA classification\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study utilized an independent blinded assessment design, wherein a young physician with five years of experience in musculoskeletal imaging and a senior expert with fifteen years of experience in the same field independently evaluated the degree of degeneration of bilateral facet joints in the L2/3-L4/5 segments. Both readers were blinded to clinical diagnosis and to each other\u0026apos;s assessments. The assessment was conducted based on the Weishaupt criteria using thin-slice CT images [18]. The resulting lumbar spine CT classifications served as the reference standard for subsequent spatial registration and classification of the X-ray image annotation results. In this study, the lumbar facet joint samples from Weishaupt grade 3 were designated as the positive group, referred to as the G3 group. Conversely, the samples from Weishaupt grade 0 to 2 were designated as the negative group, referred to as the G0-2 group. In cases of disagreement regarding the Weishaupt classification, resolution was achieved through negotiation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImage Segmentation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eImage processing was conducted using ITK-SNAP software (version 3.8.0, http://www.itksnap.org). A young radiologist completed the segmentation of the region of interest (ROI), which was subsequently reviewed and corrected by an experienced radiology expert. Manual segmentation was performed on the lumbar spine X-ray images, with the ROI encompassing the bilateral facet joints of the L2/3-L4/5 segments. On the X-ray image, the ROI was manually delineated as follows: the upper and lower boundaries were defined by the cranial and caudal margins of the superior articular process, respectively. The medial and lateral boundaries were defined by the medial margin of the inferior articular process and the lateral margin of the superior articular process, respectively. The shadow of the pedicle was excluded from the segmentation region.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRadiomic Features Extraction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, the extraction of radiomic features was conducted using the Shukun Technology research platform (https://medresearch.shukun.net/). Prior to radiomic feature extraction, standardized preprocessing was applied to the X-ray images. All images were resampled to a uniform pixel spacing of 1.0 \u0026times; 1.0 mm using cubic spline interpolation. Intensity normalization was performed to minimize inter-image variations, and a fixed bin width of 25 was applied for discretizing the gray-level values. A total of 1,698 quantitative imaging features were extracted, which include first-order statistical features (n = 324), gray level co-occurrence matrix (GLCM, n = 432), gray level dependence matrix (GLDM, n = 252), gray level run length matrix (GLRLM, n = 288), gray level size zone matrix (GLSZM, n = 288), neighboring gray tone difference matrix (NGTDM, n = 90), and three-dimensional morphological features (n = 14).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine Learning Models Development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFirst, standardize and preprocess the omics feature datasets. Perform Pearson correlation analysis among all features. If the absolute value of the correlation coefficient between any feature and another feature is \u0026ge;0.9, remove that feature. Subsequently, feature compression is conducted using the least absolute shrinkage and selection operator (LASSO) regression model, which is set to a maximum of 3000 iterations and a convergence threshold of 0.0001.\u003c/p\u003e\n\u003cp\u003eThis study employs three machine learning algorithms to develop diagnostic models, specifically logistic regression, linear support vector classification (LinearSVC), and support vector machines (SVM).\u0026nbsp;To mitigate overfitting and enhance model generalizability, a 10-fold cross-validation strategy was employed throughout model development and evaluation. The workflow of this research is illustrated in Fig. 2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical Analyses\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll statistical analyses were performed using SPSS (version 25.0; IBM Corp) and R language (version 4.1.2; R Project, https://www.r-project.org). Continuous variables were described as mean \u0026plusmn; standard deviation, determined based on the results of the Shapiro-Wilk normality test. Categorical variables were presented in frequency form, and inter-group comparisons were conducted using the \u0026chi;\u0026sup2; test or Fisher\u0026apos;s exact test. The predictive models were comprehensively evaluated using the area under the receiver operating characteristic curve (AUC), alongside indicators such as sensitivity, specificity, and accuracy. Additionally, positive predictive value (PPV) and negative predictive value (NPV) were calculated to enhance clinical interpretability. To provide a more comprehensive evaluation of model performance under class imbalance, precision-recall curves (PRC) were additionally plotted. Calibration curves were also generated to assess the calibration of predicted probabilities. A \u003cem\u003ep\u003c/em\u003e-value \u0026lt; 0.05 was considered statistically significant.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eClinical Features\u003c/h2\u003e\u003cp\u003eThis study included a total of 1,997 patients, with an average age of 57.35\u0026thinsp;\u0026plusmn;\u0026thinsp;16.86 years, consisting of 981 males and 1,026 females. A total of 7,142 lumbar facet joints were analyzed. The facet joints of 1,720 patients from the Center I were stratified and randomly allocated into a training dataset (n\u0026thinsp;=\u0026thinsp;4,470) and a validation dataset (n\u0026thinsp;=\u0026thinsp;1,917). The external test dataset comprised facet joints from 277 patients at the Center II (n\u0026thinsp;=\u0026thinsp;755). The clinical characteristics of the three datasets are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. No significant differences were observed in the distributions of age and gender among the datasets. The proportion of G3 group samples in the internal training dataset, internal validation dataset, and external test dataset were 20.31% (895/4407), 20.03% (384/1917), and 13.24% (100/755), respectively.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eBaseline in Training, Validation, and External Test Datasets\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCharacteristics\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;1202)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eValidation\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;518)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eExternal Test\u003c/p\u003e\u003cp\u003e(n\u0026thinsp;=\u0026thinsp;277)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cem\u003ep\u003c/em\u003e value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e622 (51.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e265 (51.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e139 (50.2%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.94\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e580 (48.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e253 (48.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e138 (49.8%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge [years]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e57.41\u0026thinsp;\u0026plusmn;\u0026thinsp;16.93\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e57.74\u0026thinsp;\u0026plusmn;\u0026thinsp;16.74\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e56.10\u0026thinsp;\u0026plusmn;\u0026thinsp;15.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003eN(%), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eDevelopment of radiomics Models\u003c/h2\u003e\u003cp\u003eFollowing standardized preprocessing, 1,697 radiomic features were initially extracted. Features exhibiting high inter-feature correlation with an absolute Pearson correlation coefficient\u0026thinsp;\u0026ge;\u0026thinsp;0.9 were removed, yielding 281 candidate features. Subsequently, LASSO regression was performed to select key features, resulting in 20 radiomic features with non-zero coefficients. The selected features and their corresponding coefficients in three models are presented in Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e. To develop a high-performance classification model, three classic machine learning algorithms were systematically evaluated: logistic regression, LinearSVC, and SVM. The results indicate that all three models can accurately predict lumbar FJOA (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), with the SVM model demonstrating superior performance in both the training and validation datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Its ROC curve AUCs are 0.915 (95% CI: 0.905\u0026ndash;0.924) and 0.862 (95% CI: 0.841\u0026ndash;0.882), which surpass those of the logistic regression model and the LinearSVC model. Specifically, the AUCs for the logistic regression model in the training and validation sets are 0.875 (95% CI: 0.862\u0026ndash;0.888) and 0.851 (95% CI: 0.830\u0026ndash;0.872), respectively; the AUCs for the LinearSVC model are 0.875 (95% CI: 0.862\u0026ndash;0.887) and 0.853 (95% CI: 0.833\u0026ndash;0.874), respectively.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePredictive Performance of Each Model\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"9\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDataset\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAUC\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e95% CI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eACC\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSEN\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eSPE\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003ePPV\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e\u003cp\u003eNPV\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLogistic Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.875\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0862\u0026ndash;0.888\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.793\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.811\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.788\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.489\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.932\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLogistic Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eValidation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.851\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.851\u0026ndash;0.872\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.764\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.779\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0. 760\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.448\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.943\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLogistic Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal Test\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.971\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.956\u0026ndash;0.986\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.780\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.980\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.750\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.374\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.996\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.915\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.905\u0026ndash;0.924\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.845\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.837\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.847\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.578\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.954\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eValidation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.862\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.841\u0026ndash;0.882\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.803\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.750\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.816\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.505\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.929\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal Test\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.946\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.925\u0026ndash;0.968\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.840\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.900\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.831\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.448\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.982\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLinearSVC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.875\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.862\u0026ndash;0.877\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.797\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.806\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.795\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.496\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.942\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLinearSVC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eValidation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.853\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.833\u0026ndash;0.874\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.772\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.781\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.769\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0.459\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.934\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLinearSVC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal Test\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.966\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.949\u0026ndash;0.982\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.805\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.980\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.779\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0403\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.996\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"9\"\u003eAUC area under the curve, CI confidence interval, SEN Sensitivity, SPE Specificity, NPV negative predictive value, PPV positive predictive value.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eExternal test of machine learning models\u003c/h2\u003e\u003cp\u003eIn the external test dataset, the logistic regression model exhibited superior diagnostic performance, achieving an AUC of 0.971 (95% CI: 0.956\u0026ndash;0.986). This performance surpassed that of the SVM model, which recorded an AUC of 0.946 (95% CI: 0.925\u0026ndash;0.968), and the LinearSVC model, which achieved an AUC of 0.966 (95% CI: 0.949\u0026ndash;0.982). Further analysis of performance metrics revealed that the logistic regression model achieved a diagnostic specificity of 75.0%, a sensitivity of 98.0%, and an overall accuracy of 78.0% for G3 FJOA in the external test set. As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the PR curve analysis based on the external test set indicates that the curve distribution of the logistic regression model significantly approaches the upper right area. Quantitative evaluation reveals that the AUCPR of the logistic regression model in the training set is 0.839, which surpasses that of the SVM model with an AUCPR of 0.793 and the LinearSVC model with an AUCPR of 0.813. Figure S2 displays the calibration curves for three datasets, demonstrating that the all models fit reasonably well.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study successfully developed a radiomics diagnostic model for FJOA using lumbar spine X-rays. The results indicated that the radiomics model based on the logistic regression algorithm demonstrated the best performance, achieving an AUC of approximately 0.971 in the external test dataset. While the specificity of the logistic regression model was relatively low at 75.0%, it exhibited high sensitivity at 98.0%, thereby fulfilling the screening requirements for FJOA patients based on lumbar spine X-rays.\u003c/p\u003e\u003cp\u003eThe imaging assessment of traditional lumbar FJOA primarily depends on manual interpretation of X-rays, CT, and MRI [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. CT exhibits high sensitivity in detecting subtle bone changes, attributed to its advantage of three-dimensional visualization. However, the associated radiation dose and cost restrict its use as a first-line screening tool [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. MRI may underestimate the bony structural changes associated with FJOA [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], and the consistency with X-ray assessments is only moderate [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. In recent years, some scholars have proposed the use of single photon emission computed tomography and hybrid imaging technology to assist in clinical diagnosis and treatment; however, their feasibility still requires further research [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Invasive facet joint injections, regarded as the gold standard for diagnosis, carry risks of complications and are time-consuming[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Characterized by its low cost, controllable radiation dose, high clinical prevalence, and superior spatial resolution, X-ray imaging provides distinct advantages in visualizing the bony structures of the lumbar spine, thereby establishing itself as an essential tool for routine orthopedic screening and diagnosis [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. This study proposes the use of CT as an anatomical validation reference standard, aiming to mitigate the false positive interference caused by tissue overlap in X-rays by establishing a cross-modal feature mapping relationship between X-rays and CT. This method not only enhances the specificity of X-ray imaging in FJOA screening but also provides a more interpretable feature set for the subsequent development of automated diagnostic models based on X-rays. The radiomics model based on lumbar spine X-rays developed in this study enables early identification of FJOA. Positive cases can be referred for CT confirmation or MR soft tissue evaluation, thereby establishing a hierarchical diagnosis and treatment system. This approach not only optimizes the allocation of medical resources but also encourages timely intervention for early-stage patients, effectively delaying disease progression.\u003c/p\u003e\u003cp\u003eRadiomics can quantify the heterogeneity of bone microstructure through the high-throughput extraction of texture features, morphological parameters, and gray-scale distribution information from X-ray images [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. In the field of lumbar spine radiomics, studies have successfully developed diagnostic models utilizing X-ray features, thereby confirming the clinical value of this technology [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. However, there remains a substantial gap in radiomics research concerning lumbar FJOA. Current methodologies predominantly depend on conventional imaging morphological parameters, which exhibit a weak correlation with the severity of pain [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Furthermore, traditional visual assessments exhibit low inter-observer consistency due to the overlapping characteristics present in X-ray imaging, which complicates the attainment of precise grading [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Therefore, it is imperative to develop more objective and quantifiable diagnostic tools for lumbar FJOA.\u003c/p\u003e\u003cp\u003eThis study systematically evaluated the performance of logistic regression model, SVM model, and LinearSVC model, and determining the efficacy of these algorithms in screening for lumbar FJOA. The results indicate that while the SVM model exhibited exceptional performance on the internal training and validation datasets, the logistic regression model displayed markedly superior diagnostic capabilities in the external test dataset. Consequently, the logistic regression model is identified as the optimal model developed in this study. However, it is noteworthy that the specificity of the logistic regression model in the external test set is relatively low. This may be attributed to the fact that, although radiomics can develop high-performance predictive models by extracting quantitative features such as texture and morphology from X-ray images using high-throughput methods, the inherent anatomical structure projection overlap on lumbar spine X-ray films limits its ability to clearly distinguish subtle bony structure changes, resulting in an increased false-positive prediction rate. Furthermore, given the low proportion of positive samples in the dataset of this study, and the fact that AUPRC is a more sensitive evaluation metric for minority class samples, it becomes particularly significant in such imbalanced scenarios. The superior AUPRC performance of the logistic regression model, which confirms its effectiveness in identifying high-risk FJOA, underscores the necessity of concurrently evaluating both AUC and AUPRC in scenarios characterized by class imbalance.\u003c/p\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eLimitation\u003c/h2\u003e\u003cp\u003eHowever, this study has several limitations. Firstly, the absence of facet joint injection, which is regarded as the gold standard for diagnosing FJOA, introduces potential subjective bias in the manual grouping of positive and negative samples, which relied solely on CT-based Weishaupt grading. Secondly, anatomical overlap in lumbar spine X-rays may compromise facet joint ROI delineation, particularly at superior and inferior margins, due to spinal curvature variations or inconsistent positioning, ultimately limiting reproducibility.\u003c/p\u003e\u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrates that radiomics facilitates the accurate identification of FJOA. The developed model effectively screens for FJOA in patients undergoing routine lumbar radiography, enabling a rapid determination of whether FJOA contributes to the etiology of their low back pain. This provides objective imaging support for orthopedic decision-making concerning further diagnostic or therapeutic interventions.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eACC Accuracy\u003c/p\u003e\n\u003cp\u003eAUC Area under the curve\u003c/p\u003e\n\u003cp\u003eAUPRC Area under the precision-recall curve\u003c/p\u003e\n\u003cp\u003eCI Confidence interval\u003c/p\u003e\n\u003cp\u003eCT Computed tomography\u003c/p\u003e\n\u003cp\u003eFJOA Facet joint osteoarthritis\u003c/p\u003e\n\u003cp\u003eGLCM Gray level co-occurrence matrix\u003c/p\u003e\n\u003cp\u003eGLDM Gray level dependence matrix\u003c/p\u003e\n\u003cp\u003eGLRLM Gray level run length matrix\u003c/p\u003e\n\u003cp\u003eGLSZM Gray level size zone matrix\u003c/p\u003e\n\u003cp\u003eLASSO Least absolute shrinkage and selection operator\u003c/p\u003e\n\u003cp\u003eLinearSVC Linear support vector classification\u003c/p\u003e\n\u003cp\u003eMRI Magnetic resonance imaging\u003c/p\u003e\n\u003cp\u003eNGTDM Neighboring gray tone difference matrix\u003c/p\u003e\n\u003cp\u003eNPV Negative predictive value\u003c/p\u003e\n\u003cp\u003ePPV Positive predictive value\u003c/p\u003e\n\u003cp\u003ePRC Precision-recall curve\u003c/p\u003e\n\u003cp\u003eROC Receiver operating characteristic\u003c/p\u003e\n\u003cp\u003eROI Region of interest\u003c/p\u003e\n\u003cp\u003eSEN Sensitivity\u003c/p\u003e\n\u003cp\u003eSPE Specificity\u003c/p\u003e\n\u003cp\u003eSVM Support vector machines\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Institutional Review Board of the Second Affiliated Hospital of Fujian Medical University, China (No. 2024-519) in accordance with the Declaration of Helsinki. Due to the retrospective nature of the study, the requirement for written informed consent was waived by the same ethics committee.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study has received funding by Joint funds for the innovation of science and technology,Fujian \u0026nbsp;Province（No:2024Y9388）.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors contributed to the study\u0026apos;s conception and design. Material preparation, data collection, and analysis were performed by L.Z.J., Y.F.G. and Z.C.S. The first draft of the manuscript was written by L.Z.J. and Y.F.G. Y.Z.L. and H.N.Z. supervised the project and were responsible for funding acquisition. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eGlobal, regional, and national burden of low back pain, 1990-2020, its attributable risk factors, and projections to 2050: a systematic analysis of the Global Burden of Disease Study 2021. \u003cem\u003eLancet Rheumatol \u003c/em\u003e2023, 5(6):e316-e329.\u003c/li\u003e\n\u003cli\u003eDu R, Gao J, Wang B, Zhang J, Meng M, Wang J, Qu W, Li Z: Percutaneous radiofrequency ablation and endoscopic neurotomy for lumbar facet joint syndrome: are they good enough? \u003cem\u003eEur Spine J \u003c/em\u003e2024, 33(2):463-473.\u003c/li\u003e\n\u003cli\u003eAbd-Elsayed A, Azeem N, Chopra P, D\u0026apos;Souza RS, Sayed D, Deer T: An International Survey on the Practice of Lumbar Radiofrequency Ablation for Management of Zygapophyseal (Facet)-Mediated Low Back Pain. \u003cem\u003eJ Pain Res \u003c/em\u003e2022, 15:1083-1090.\u003c/li\u003e\n\u003cli\u003ePerolat R, Kastler A, Nicot B, Pellat JM, Tahon F, Attye A, Heck O, Boubagra K, Grand S, Krainik A: Facet joint syndrome: from diagnosis to interventional management. \u003cem\u003eInsights Imaging \u003c/em\u003e2018, 9(5):773-789.\u003c/li\u003e\n\u003cli\u003eGajjar AA, Dombrovsky DA, Singh R, Stonnington HO, Koester SW, Polavarapu H, George DD, Wakim A, Harland T, Mansfield K: Trends in Spinal Pain Procedure Volumes and Reimbursements: An Analysis of 20\u0026thinsp;Years of Medicare Data. \u003cem\u003ePain Pract \u003c/em\u003e2025, 25(5):e70043.\u003c/li\u003e\n\u003cli\u003ePark S, Park JH, Sokpeou N, Jang JN, Kim YU, Choi YS, Park S: Radiofrequency treatments for lumbar facet joint syndrome: a systematic review and network meta-analysis. \u003cem\u003eReg Anesth Pain Med \u003c/em\u003e2024.\u003c/li\u003e\n\u003cli\u003eAnaya JEC, Coelho SRN, Taneja AK, Cardoso FN, Skaf AY, Aihara AY: Differential Diagnosis of Facet Joint Disorders. \u003cem\u003eRadiographics \u003c/em\u003e2021, 41(2):543-558.\u003c/li\u003e\n\u003cli\u003eMartins DE, Astur N, Kanas M, Ferretti M, Lenza M, Wajchenberg M: Quality assessment of systematic reviews for surgical treatment of low back pain: an overview. \u003cem\u003eSpine J \u003c/em\u003e2016, 16(5):667-675.\u003c/li\u003e\n\u003cli\u003eYoo YM, Kim KH: Facet joint disorders: from diagnosis to treatment. \u003cem\u003eKorean J Pain \u003c/em\u003e2024, 37(1):3-12.\u003c/li\u003e\n\u003cli\u003eDu R, Xu G, Bai X, Li Z: Facet Joint Syndrome: Pathophysiology, Diagnosis, and Treatment. \u003cem\u003eJ Pain Res \u003c/em\u003e2022, 15:3689-3710.\u003c/li\u003e\n\u003cli\u003eBerg L, Thoresen H, Neckelmann G, Furunes H, Hellum C, Espeland A: Facet arthropathy evaluation: CT or MRI? \u003cem\u003eEur Radiol \u003c/em\u003e2019, 29(9):4990-4998.\u003c/li\u003e\n\u003cli\u003eThiel CL, Vigil-Garcia M, Nande S, Meijer C, Gehrels J, Struk O, Thornander S, Pullella D, Omary RA, Carver DE\u003cem\u003e et al\u003c/em\u003e: Environmental Life Cycle Assessment of a U.S. Hospital-based Radiology Practice. \u003cem\u003eRadiology \u003c/em\u003e2024, 313(2):e240398.\u003c/li\u003e\n\u003cli\u003eDe Wilde D, Zanier O, Da Mutten R, Jin M, Regli L, Serra C, Staartjes VE: Strategies for generating synthetic computed tomography-like imaging from radiographs: A scoping review. \u003cem\u003eMed Image Anal \u003c/em\u003e2025, 101:103454.\u003c/li\u003e\n\u003cli\u003eGillies RJ, Kinahan PE, Hricak H: Radiomics: Images Are More than Pictures, They Are Data. \u003cem\u003eRadiology \u003c/em\u003e2016, 278(2):563-577.\u003c/li\u003e\n\u003cli\u003eNian S, Zhao Y, Li C, Zhu K, Li N, Li W, Chen J: Development and validation of a radiomics-based model for predicting osteoporosis in patients with lumbar compression fractures. \u003cem\u003eSpine J \u003c/em\u003e2024, 24(9):1625-1634.\u003c/li\u003e\n\u003cli\u003eZhang B, Chen Z, Yan R, Lai B, Wu G, You J, Wu X, Duan J, Zhang S: Development and Validation of a Feature-Based Broad-Learning System for Opportunistic Osteoporosis Screening Using Lumbar Spine Radiographs. \u003cem\u003eAcad Radiol \u003c/em\u003e2024, 31(1):84-92.\u003c/li\u003e\n\u003cli\u003eWu Z, Wang H, Zheng Y, Fei H, Dong C, Wang Z, Ren W, Xu W, Bian T: Lumbar MR-based radiomics nomogram for detecting minimal residual disease in patients with multiple myeloma. \u003cem\u003eEur Radiol \u003c/em\u003e2023, 33(8):5594-5605.\u003c/li\u003e\n\u003cli\u003eWeishaupt D, Zanetti M, Boos N, Hodler J: MR imaging and CT in osteoarthritis of the lumbar facet joints. \u003cem\u003eSkeletal Radiol \u003c/em\u003e1999, 28(4):215-219.\u003c/li\u003e\n\u003cli\u003eSaremi A, Goyal KK, Benzel EC, Orr RD: Evolution of lumbar degenerative spondylolisthesis with key radiographic features. \u003cem\u003eSpine J \u003c/em\u003e2024, 24(6):989-1000.\u003c/li\u003e\n\u003cli\u003eLittle JW, Grieve T, Cantu J, Bogar WC, Heiser R, Miley H, Cramer GD: 20 Reliability of Human Lumbar Facet Joint Degeneration Severity Assessed by Magnetic Resonance Imaging. \u003cem\u003eJ Manipulative Physiol Ther \u003c/em\u003e2020, 43(1):43-49.\u003c/li\u003e\n\u003cli\u003eLacroix M, Nguyen C, Burns R, Laporte A, Rannou F, Feydy A: Degenerative Lumbar Spine Disease: Imaging and Biomechanics. \u003cem\u003eSemin Musculoskelet Radiol \u003c/em\u003e2022, 26(4):424-438.\u003c/li\u003e\n\u003cli\u003eLiu X, Zhu Z, He S, Sun M, Zhao T, Liu L, Shi H, Hou Y, Shi G: Enhancing Clinical Insights: New Radiographic Grading for Lumbar Facet Joint Degeneration, A Reliability Study. \u003cem\u003eJOR Spine \u003c/em\u003e2025, 8(1):e70035.\u003c/li\u003e\n\u003cli\u003eYolcu YU, Lehman VT, Bhatti AUR, Goyal A, Alvi MA, Bydon M: Use of Hybrid Imaging Techniques in Diagnosis of Facet Joint Arthropathy: A Narrative Review of Three Modalities. \u003cem\u003eWorld Neurosurg \u003c/em\u003e2020, 134:201-210.\u003c/li\u003e\n\u003cli\u003ePendleton J, Ng A: SPECT/CT Scan: A New Diagnostic Tool in Pain Medicine. \u003cem\u003eCurr Pain Headache Rep \u003c/em\u003e2023, 27(11):729-735.\u003c/li\u003e\n\u003cli\u003eVan Oosterwyck W, Vander Cruyssen P, Castille F, Van de Kelft E, Decaigny V: Lumbar Facet Joint Disease: What, Why, and When? \u003cem\u003eLife (Basel) \u003c/em\u003e2024, 14(11).\u003c/li\u003e\n\u003cli\u003eBusse JW, Genevay S, Agarwal A, Standaert CJ, Carneiro K, Friedrich J, Ferreira M, Verbeke H, Brox JI, Xiao H\u003cem\u003e et al\u003c/em\u003e: Commonly used interventional procedures for non-cancer chronic spine pain: a clinical practice guideline. \u003cem\u003eBmj \u003c/em\u003e2025, 388:e079970.\u003c/li\u003e\n\u003cli\u003ePina Vegas L, van Lunteren M, Loeuille D, Morizot C, Newsum E, Ramiro S, van Gaalen F, Saraux A, Claudepierre P, Feydy A\u003cem\u003e et al\u003c/em\u003e: Ten-year follow-up of degenerative spinal lesions on radiographs and MRI in axial spondyloarthritis: results of the DESIR (DEvenir des spondylarthropathies indiff\u0026eacute;renci\u0026eacute;es r\u0026eacute;centes) cohort. \u003cem\u003eEur Radiol \u003c/em\u003e2025.\u003c/li\u003e\n\u003cli\u003eMcCormick RJ, Perloff MD: Xray prediction of MRI in low back pain. \u003cem\u003eAm J Phys Med Rehabil \u003c/em\u003e2025.\u003c/li\u003e\n\u003cli\u003eGitto S, Annovazzi A, Nulle K, Interlenghi M, Salvatore C, Anelli V, Baldi J, Messina C, Albano D, Di Luca F\u003cem\u003e et al\u003c/em\u003e: X-rays radiomics-based machine learning classification of atypical cartilaginous tumour and high-grade chondrosarcoma of long bones. \u003cem\u003eEBioMedicine \u003c/em\u003e2024, 101:105018.\u003c/li\u003e\n\u003cli\u003evon Schacky CE, Wilhelm NJ, Sch\u0026auml;fer VS, Leonhardt Y, Jung M, Jungmann PM, Russe MF, Foreman SC, Gassert FG, Gassert FT\u003cem\u003e et al\u003c/em\u003e: Development and evaluation of machine learning models based on X-ray radiomics for the classification and differentiation of malignant and benign bone tumors. \u003cem\u003eEur Radiol \u003c/em\u003e2022, 32(9):6247-6257.\u003c/li\u003e\n\u003cli\u003eGalbusera F, Cina A, O\u0026apos;Riordan D, Vitale JA, Loibl M, Fekete TF, Kleinst\u0026uuml;ck F, Haschtmann D, Mannion AF: Estimating lumbar bone mineral density from conventional MRI and radiographs with deep learning in spine patients. \u003cem\u003eEur Spine J \u003c/em\u003e2024, 33(11):4092-4103.\u003c/li\u003e\n\u003cli\u003eCheng L, Cai F, Xu M, Liu P, Liao J, Zong S: A diagnostic approach integrated multimodal radiomics with machine learning models based on lumbar spine CT and X-ray for osteoporosis. \u003cem\u003eJ Bone Miner Metab \u003c/em\u003e2023, 41(6):877-889.\u003c/li\u003e\n\u003cli\u003eHu J, Zhang Y, Duan C, Peng X, Hu P, Lu H: Feasibility study for evaluating early lumbar facet joint degeneration using axial T(1) \u0026rho;, T(2) , and T2* mapping in cartilage. \u003cem\u003eJ Magn Reson Imaging \u003c/em\u003e2017, 46(2):468-475.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Lumbar facet joints, Osteoarthritis, Radiomics, Radiography","lastPublishedDoi":"10.21203/rs.3.rs-7515717/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7515717/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjectives\u003c/h2\u003e\u003cp\u003eThis study aims to develop and validate radiomics models utilizing lumbar spine X-rays for the early identification of facet joint osteoarthritis (FJOA).\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eThis retrospective two-center study enrolled 1,997 patients who underwent paired lumbar X-ray and CT imaging within one month. Data from one center were used for model training and validation, and data from the other center were used for external testing. Radiomic features were extracted from manually segmented facet joint regions on X-rays. Key features selected through the least absolute shrinkage and selection operator (LASSO) were used to develop models, specifically logistic regression, linear support vector classification (LinearSVC), and support vector machines (SVM). The model performance was primarily evaluated using the area under the ROC curve (AUC) and the area under the precision-recall curve (AUPRC).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eA total of 20 features were selected for modeling. The logistic regression model based on radiomic features demonstrated the highest AUC. In the external testing cohort, this model achieved an AUC of 0.971 (95% CI: 0.956\u0026ndash;0.986), a sensitivity of 98.0%, a specificity of 75.0%, and an AUPRC of 0.839. It outperformed both the SVM model (AUC\u0026thinsp;=\u0026thinsp;0.946, AUPRC\u0026thinsp;=\u0026thinsp;0.793) and the LinearSVC model (AUC\u0026thinsp;=\u0026thinsp;0.966, AUPRC\u0026thinsp;=\u0026thinsp;0.813).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eRadiomics models based on lumbar X-rays showed robust performance and hold promise as a non-invasive, accessible tool for early and accurate identification of FJOA, potentially enabling timely intervention.\u003c/p\u003e","manuscriptTitle":"Radiomics analysis of lumbar spine X-ray images for diagnosing facet joint osteoarthritis: a two-center study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-22 10:06:37","doi":"10.21203/rs.3.rs-7515717/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2025-09-12T08:34:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"56251127507712142054338636484966140471","date":"2025-09-12T07:03:13+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-11T22:46:31+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-09T12:25:41+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-09T12:24:18+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2025-09-02T08:58:56+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9236ed9b-77bf-4896-8f32-eca9b1889a04","owner":[],"postedDate":"September 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-09-22T10:06:38+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-22 10:06:37","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7515717","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7515717","identity":"rs-7515717","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}