Research on Assisting X-ray Diagnosis of Osteoporotic Vertebral Compression Fractures Using Interpretable Machine Learning Models and Radiomics Features

Research on Assisting X-ray Diagnosis of Osteoporotic Vertebral Compression Fractures Using Interpretable Machine Learning Models and Radiomics Features | 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 Research on Assisting X-ray Diagnosis of Osteoporotic Vertebral Compression Fractures Using Interpretable Machine Learning Models and Radiomics Features Kangen Han, Hongwen Gu, Yu Li, Junchao LI, Zhihao Zhang, Yin Hu, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6127302/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objective: To improve early diagnosis rates, this study applies a combination of radiomics and machine learning algorithms to aid in the X-ray diagnosis of osteoporotic vertebral compression fractures (OVCF). Methods: Data were collected from 852 patients from January 2016 to December 2023, including lateral X-rays of the L1 vertebra and demographic information. The cohort included 589 patients with lumbar back pain but normal MRI results, and 263 patients diagnosed with various degrees of OVCF by MRI. Patients were randomly divided into training (70%) and validation (30%) groups. X-ray images were annotated to extract radiomics features, which were then selected to finalize the radiomics score, along with meaningful clinical factors. Five machine learning algorithms were utilized to model and compare the diagnostic efficacy of clinical prediction models, radiomics models, and combined models, identifying the optimal model group and machine learning algorithm. The SHAP method was employed for further explanatory analysis. Results: Variables showing significant differences between groups included gender, smoking history, trauma history, history of lumbar surgery, residential area, history of glucocorticoid treatment, age, and VAS score. Through t-tests, intraclass correlation coefficients (ICCs), and LASSO regression analysis (Least Absolute Shrinkage and Selection Operator), eight radiomics features were identified to establish a Radscore. Multifactorial logistic regression analysis identified gender, smoking history, trauma history, lumbar surgery history, residential area, and Radscore as independent risk factors for OVCF. The combined model outperformed the other two. Due to overfitting in the Random Forest algorithm, KNN was determined to be the best machine learning algorithm. SHAP bar graphs displayed the influence factors in descending order of impact: residential area, Radscore, trauma history, gender, smoking, and lumbar surgery history. SHAP swarm plots revealed a broad distribution of Radscore, underscoring its significant predictive influence. Conclusion: The diagnostic model developed through radiomics and machine learning algorithms reached an ideal level of effectiveness, with KNN in the combined model group demonstrating the highest diagnostic efficacy for assisting in the early X-ray diagnosis of OVCF. Radiomics machine learning osteoporotic vertebral compression fractures X-ray Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Introduction Osteoporosis is a metabolic disorder characterized by decreased bone mass and the deterioration of bone structure, with the spine being the most frequently affected area. It has been reported that osteoporotic vertebral compression fractures (OVCF) are on the rise, accounting for approximately 40% of all osteoporotic fractures. These fractures not only compromise quality of life but also impose significant economic burdens on both patients and society, particularly affecting the elderly [ 1 , 2 ] . Early diagnosis of OVCF is crucial for the effective treatment of patients. While X-rays can suggest the presence of vertebral compression fractures by changes in vertebral shape, radiologists often find it challenging to detect subtle fractures, leading to high rates of misdiagnosis and delays in treatment [ 3 ] . In cases where X-ray findings are inconclusive, MRI is often employed; however, its high costs, lengthy wait times, and considerable contraindications [ 4 , 5 ] make it unsuitable for all patients. Thus, enhancing the sensitivity of X-ray diagnosis for vertebral compression fractures remains an urgent clinical need. The advent of radiomics and machine learning offers new approaches. Radiomics involves analyzing a vast amount of high-dimensional data from imaging to extract and quantify features that are imperceptible to the human eye [ 6 , 7 ] . Machine learning leverages large datasets to learn patterns and enhance performance, enabling automatic feature recognition and classification [ 8 – 10 ] . By integrating radiomics with machine learning algorithms, it is possible to develop diagnostic models that allow for the identification and analysis of subtle image features, facilitating disease diagnosis and prognosis. Historically, traditional machine learning has suffered from a lack of interpretability, often being regarded as a "black box," which has led to concerns about its reliability [ 11 , 12 ] . This study aims to employ Shapley Additive exPlanations (SHAP) to provide a global interpretation of machine learning algorithms’ internal operations, developing and validating a robust joint model based on machine learning and radiomics features for early, economical, and efficient OVCF diagnosis. Materials and Methods 2.1 General Information This study analyzed data from 852 patients collected between January 2016 and December 2023, including lateral X-rays of the L1 vertebra and demographic details such as gender, age, BMI, VAS score, symptom duration, smoking, alcohol consumption, diabetes, hypertension, thyroid disorders, trauma history, history of lumbar surgery, residential area, calcium-phosphate metabolism disorders, and glucocorticoid use. Of these, 589 presented with lumbar back pain but had normal MRI findings, while 263 were diagnosed with varying degrees of OVCF by MRI. The patients were randomly allocated into training (70%) and validation (30%) groups; the former was used for model development, and the latter for model validation. Inclusion criteria included: ① Age ≥60 years; ② Admission for lumbar back pain; ③ Complete lumbar spine X-ray and MRI imaging data. Exclusion criteria included: ① Compression of ≥2 vertebral bodies; ② Spinal malignancies, tuberculosis, ankylosing spondylitis, or other specific diseases; ③ Unclear X-ray images; ④ Difficulties in segmenting the Region of Interest (ROI) in compressed vertebral bodies，the flowchart is shown in Figure 1. This retrospective study was conducted without requiring signed informed consent from the participants and adhered to the principles of the Declaration of Helsinki. Fig.1 The patient enrollment process for this study 2.2 X-ray Image Collection and Segmentation DICOM format raw X-ray data were exported from our hospital's A-site imaging system and manually segmented using 3Dslicer version 5.6.1 software (http://www.slicer.org/). Two spinal surgeons performed the segmentation on both normal and compressed L1 vertebrae, without prior knowledge of any fractures, as shown in Figure 2. Fig.2 a: L1 vertebral compression fracture, difficult to recognize by the human eye; b: L1 vertebral hypointensity was seen in T1WI; c: MRI lipid suppression image showed L1 vertebral hyperintensity and compression fracture; d: 3Dslicer segmentation of the compressed vertebrae; e: normal lumbar spine X-ray; f.g: MRI of normal lumbar spine; h: 3Dslicer divides the normal vertebral body 2.3 Radiomics Feature Extraction This study utilized the PyRadiomics package in Python for feature extraction. The images were resampled to 3mm × 3mm × 3mm voxels using the B-Spline interpolation method and normalized by histogram matching to reduce variability across different machines. 2.4 Feature Selection Our feature selection process involved three steps: (1) Initial standardization of features using the Z-score method (X score = (X−μ)/σ, where μ is the mean, and σ is the standard deviation). Preliminary filtering of features between two groups was performed using the t-test and Mann-Whitney U test for those with p < 0.05. (2) Intraclass correlation coefficients (ICCs) were utilized to eliminate individual measurement differences among doctors and for further feature selection. Sixty patients were randomly selected, and their ROIs were manually delineated by two spinal surgeons, A and B. Surgeon B redelineated the ROIs two weeks later; features with ICCs > 0.75, both within and between observers, were included for further analysis. (3) Lastly, LASSO regression was employed to minimize some feature coefficients, using only non-zero coefficients for modeling. Ten-fold cross-validation was used to optimize the regularization parameter (λ). Each patient's radiomics score was calculated using the formula: Radscore = β0 + β1F1 + β2F2 + ... + βnFn, where Fi represents the features and βi are the coefficients from LASSO regression. 2.6 Experimental Design Five machine learning algorithms were used for modeling, including Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (Xgboost), K-Nearest Neighbors (KNN), and LightGBM. Models were established based on the training group: (1) Clinical prediction model group: Variables showing intergroup differences in demographic analysis were incorporated into a multivariate logistic regression to construct the model. (2) Radiomics model group: Meaningful variables identified via final LASSO regression analysis were used for modeling. (3) Combined model group: Meaningful clinical variables and radiomics scores were utilized to develop a combined predictive model. The machine learning classifier with the highest diagnostic performance was selected, and the SHAP algorithm was employed to interpret its internal computational processes. 2.7 Evaluation Criteria The evaluation criteria for the optimal model include: ① Receiver Operating Characteristic (ROC) curves, the Area Under the Curve (AUC), and the plotting of calibration curves to assess how closely the model approximates an ideal model, supplemented by Decision Curve Analysis (DCA) to evaluate the model's clinical utility. The DeLong test is utilized to compare differences among the ROC curves of the five machine learning methods. ② The model's predictive accuracy is assessed using metrics such as Accuracy, Sensitivity, Specificity, Precision, and F1 Score. 2.8 Statistical Methods Feature extraction was conducted using the PyRadiomics package in Python (version 3.9.13), with statistical analyses performed using R statistical software (version 4.4.1). Clinical data from the training and validation groups were analyzed based on the type of variable. Quantitative data adhering to a normal distribution were expressed as mean ± standard deviation and analyzed using the t-test; non-normally distributed quantitative data were presented as medians (Median) and interquartile ranges (IQR), and analyzed using the Mann-Whitney U test. Categorical data were presented as numbers and percentages (N, %). A p-value < 0.05 was considered to indicate statistical significance. Results 3.1 Clinical Demographics The training group included 597 individuals, and the validation group comprised 255. Variables that exhibited significant differences within the training group included gender, smoking history, trauma history, history of lumbar surgery, residential area, history of glucocorticoid treatment, age, and VAS score, as detailed in Table 1. 3.2 Radiomics Features A total of 862 features were extracted, of which:(1)First‐order statistics (n = 173); (2) Shape (n = 14); (3) Gray Level Co‐occurrence Matrix (n = 216); (4) Gray Level Run-Length Matrix (n = 144); (5) Gray - Level Size Zone Matrix (n = 144); (6) Gray - Level Dependence Matrix(n = 126); (7) Neighborhood Gray - Tone Difference Matrix (n = 45), and 3 types of filters, including:Laplacian of Gaussian（LoG），wavelets and LBP2D。The t-test and Mann-Whitney U test were employed to initially select 256 features. Applying an ICC threshold of >0.75, 188 features were subsequently selected. Variables were then further refined using LASSO regression, where an increase in λ progressively reduced the coefficients of less contributive features to zero, thus enabling variable selection. The coefficient path graph illustrates that with higher λ values, fewer variables are retained. Ultimately, eight radiomics features were finalized, as depicted in Figure3. Fig.3 Feature screening process for LASSO regression 3.3 Radiomics Score The importance coefficients for the eight variables identified through LASSO regression are displayed in Figure 4. The formula for the radiomics score is established as follows: Radscore=3.4035-0.000003×diagnostics_Mask-original_VoxelNum-8.0844×original_shape_Elongation-0.0031×wavelet-LHL_firstorder_Median-0.8402×wavelet-LHH_glszm_LowGrayLevelZoneEmphasis+3.2912×wavelet-HLL_glrlm_ShortRunLowGrayLevelEmphasis+0.0283×wavelet-HHL_ngtdm_Contrast+3.1581×wavelet-HHH_ngtdm_Strength+1.2240×wavelet-LLL_glcm_MCC Fig.4 Variable importance coefficient 3.4 Research Variables In the training group, variables that showed significant intergroup differences and were meaningful along with the radiomics score were included in a multivariate logistic regression analysis to determine the final variables used for modeling. These included gender, smoking history, trauma history, history of lumbar surgery, residential area, and Radscore, as illustrated in Figure 5. Fig.5 Variables correspond to P-values and ratios to forest plots 3.5 Model Comparison (1) Clinical Prediction Model Group: Using five machine learning algorithms, meaningful clinical variables from the multivariate analysis were employed for modeling. LightGBM exhibited the best performance in the training group with an AUC of 0.862 (95% CI=0.829-0.895), as shown in Figure 6. (2) Radiomics Model Group: Models were built using the final selected radiomics features, among which Xgboost achieved the best result with an AUC of 0.767 (95% CI=0.727-0.808), as displayed in Figure 7. (3) Combined Model Group: A predictive model was constructed combining clinical variables and the radiomics score. Random Forest demonstrated exceptional predictive performance, achieving an AUC of 0.983 (95% CI=0.970-0.996), as depicted in Figure 8. Ultimately, the combined model outperformed the other models. Calibration of the combined model was assessed in the validation group, where Xgboost, KNN, and LightGBM fitted well, as shown in Figure 9. Decision curves indicated that Xgboost, KNN, and SVM were clinically beneficial, potentially enhancing patient outcomes in medical practices, as demonstrated in Figure 10. A comparison of AUC values among each pair of classifiers in the validation group and p-values obtained via the DeLong test revealed no significant differences between Xgboost, KNN, and SVM, but these were significantly different from LightGBM and RF, as noted in Table 2. When comparing the diagnostic performance of the five machine learning algorithms in the combined model between the training and validation groups, Random Forest had an accuracy of 0.998 in the training group but only 0.685 in the validation group, indicating overfitting. KNN, however, exhibited the most stable and balanced performance in the validation group, with an AUC of 0.831 and an accuracy of 0.776, making it the best machine learning method for this study. The model established using KNN showed the highest diagnostic efficacy, as detailed in Table 3. Fig.6 ROC curves of the training group and the validation group of the clinical prediction model group Fig.7 ROC curves for the training and validation groups of the radiomics model group Fig.8 ROC curves for the training and validation groups of the combined model group Fig.9 Calibration curves for the training and validation groups of the combined model group Fig.10 DCA curves for the training group and validation group of combined model group 3.6 SHAP Interpretation of the KNN Algorithm To understand the internal classification mechanisms of the KNN, we calculated the Shapley values for the combined model interpretations using the KNN algorithm. The SHAP bar graphs provide a visual representation of the weights of the model’s six most significant features: residential area, Radscore, trauma history, gender, smoking, and history of lumbar surgery. The residential area had the greatest weight, followed by Radscore. The SHAP bee swarm plot uses purple and yellow to indicate each feature's positive or negative impact on the prediction probability. Positive influences are shown by the residential area, trauma history, smoking, and lumbar surgery history, while gender exhibits a negative influence. Radscore, although not showing a clear trend, has the broadest distribution range, highlighting its significant impact on predictions, as shown in Figure 11. Fig.11 Using SHAP to visualize the model as a whole. (A) SHAP histogram shows the weights of the six features in the model. (B) SHAP bee colony plot showing the overall trend of the data Discussion As global aging trends intensify, osteoporosis diagnoses are increasing, with nearly 200 million new cases annually. Osteoporotic vertebral compression fractures (OVCF) place unprecedented strain on healthcare systems [13] . Currently, X-rays serve as the primary diagnostic tool for OVCF, with radiologists relying on vertebral height and deformation for evaluation [4] . However, minor compression fractures often go undetected when solely relying on visual inspection. For cases where suspicion remains high, further MRI scans are typically recommended to investigate internal changes within the vertebrae. Yet, due to high costs, lengthy procedures, and strict contraindications, many patients are deterred from undergoing MRI scans [14, 15] , leading to delays in diagnosis. Consequently, patients miss the window for timely surgical interventions and face prolonged bed rest, which can lead to severe complications such as pneumonia, bedsores, thrombosis, and even life-threatening conditions [16] . In recent years, the integration of radiomics and machine learning has brought significant advancements in predictive modeling for OVCF [17, 18] . These integrated models enhance diagnostic capabilities by incorporating subtle imaging features undetectable by the human eye. In previous research, Cai et al. utilized MRI-based radiomics to assess new vertebral fractures post-vertebroplasty, achieving an Area Under the Curve (AUC) of over 0.9 in both training and validation groups, showcasing remarkable predictive power [19] . Similarly, Sun et al. applied radiomics features in-depth and utilized a Naive Bayes machine learning model to diagnose osteoporosis, achieving an accuracy of 0.81 and a sensitivity of 0.87 [20] . These findings confirm that applying well-trained radiomics models for clinical image recognition is not only feasible but also surpasses manual recognition in both precision and sensitivity. These results strongly support the potential of combined radiomics and machine learning models to significantly improve early diagnosis rates for OVCF. Notably, prior radiomics diagnostics rarely incorporated clinical factors [7, 21, 22] . This study identified gender, smoking history, trauma history, history of lumbar surgery, and residential area as independent risk factors for OVCF through multivariate logistic regression analysis. Therefore, our research was segmented into clinical prediction, radiomics prediction, and combined prediction groups, employing five cutting-edge machine learning algorithms to explore differences among these groups. The combined model of radiomics emerged with superior performance. Due to overfitting issues in the Random Forest algorithm, the KNN algorithm, with its excellent generalization ability, demonstrated outstanding performance. In machine learning, maintaining a balance between accuracy and model interpretability is crucial [23] . Deeply understanding the complex decision-making mechanisms of models increases their credibility, which is vital for their widespread clinical application. Consequently, this study employed Shapley values to interpret the best-performing KNN algorithm. It was discovered that residential area was the most significant factor, closely followed by the radiomics score, with only a slight difference between the two. Urban residents are more susceptible to OVCF, likely due to lower physical activity levels associated with sedentary work and insufficient outdoor exercise. Concurrently, the prevalence of smoking and excessive drinking in urban populations, which can disrupt bone metabolism, should not be underestimated. Rural residents, who often engage in more physical labor and have more exposure to sunlight, enabling more efficient vitamin D synthesis, are at a lower risk for OVCF [24-26] . In conclusion, the results of this study provide strong justification for the belief that the combined model of radiomics and machine learning possesses impressive diagnostic efficacy, fully capable of being applied in clinical practice. It significantly enhances the detection rate of OVCF in X-rays, drastically reducing the rate of missed diagnoses and offering new alternatives for patients who cannot undergo MRI scans. This study has limitations, including the intentional avoidance of abnormal structures during data collection, which may impact the results. Additionally, being a single-center study without external validation, the variability in X-ray machines, and the relatively small sample size may have led to some overfitting. Future research should involve larger sample sizes to reduce overfitting and enhance model performance. Despite these limitations, our innovative application of radiomics for the early diagnosis of OVCF via X-ray marks a significant advancement. As radiomics continues to evolve, we anticipate that data from radiological examinations worldwide will be transformed into quantitative features, building databases from extensive radiomics data from millions of patients to enhance the accuracy and predictive capabilities of models, thereby providing robust support for clinical diagnosis of OVCF assisted by models. Declarations Human Ethics and Consent to Participate declarations: not applicable Internal Review Board (IRB): Ethical approval was obtained from the Institutional Review Board(IRB) of General Hospital of Northern Theater Command. Clinical trial number: not applicable Clinical trial number: not applicable Ethics approval and consent to participate: Not applicable Consent for publication: All authors read and approved the final manuscript. Availability of data and material: The data that support the findings of this study are available from the corresponding author on special request Funding: This work was supported by Liaoning Province Applied Basic Research Programme (2023JH2/101700130)；Liaoning Province Applied Basic Research Plan (2022JH2/101300024) Competing interests: All listed authors have made substantial contributions to the manuscript and do not have any conflicts of interest. Authors' contributions: Kangen Han, Yu Li, Hongwen Gu, Yin Hu, Shilei Tang, Zhihao Zhang, Hailong Yu, Hongwei Wang designed and participated in the whole process of the study and drafted the manuscript. All authors read and approved the final manuscript. Informed consent: Ethical approval was obtained from the Institutional Review Board(IRB) of General Hospital of Northern Theater Command. Chinese law does not require individual informed consent from participants in non-invasive observational trials such as the present study. Therefore, the need for informed consent was waived according to the instruction of IRB of General Hospital of Northern Theater Command. Acknowledgements: All listed authors have made substantial contributions to the manuscript and do not have any conflicts of interest. All authors read and approved the final manuscript. This work was supported by Applied Basic Research Project of Liaoning Province (2023JH2/101700130) and Liaoning Province Applied Basic Research Plan (2022JH2/101300024) References Yang XG, Dong YQ, Liu X, et al. Incidence and prognostic factors of residual back pain in patients treated for osteoporotic vertebral compression fractures: a systematic review and meta-analysis [J]. Eur Spine J. 2024;33(12):4521–37. Zheng XQ, Xu L, Huang J, et al. Incidence and cost of vertebral fracture in urban China: a 5-year population-based cohort study [J]. Int J Surg. 2023;109(7):1910–8. Patel D, Liu J, Ebraheim NA. Managements of osteoporotic vertebral compression fractures: A narrative review [J]. World J Orthop. 2022;13(6):564–73. Spiegl U, Bork H, Grüninger S, et al. Osteoporotic Fractures of the Thoracic and Lumbar Vertebrae: Diagnosis and Conservative Treatment [J]. Dtsch Arztebl Int. 2021;118(40):670–7. Yamamoto Y, Iwata E, Shigematsu H, et al. Differential diagnosis between metastatic and osteoporotic vertebral fractures using sagittal T1-weighted magnetic resonance imaging [J]. J Orthop Sci. 2020;25(5):763–9. Guiot J, Vaidyanathan A, Deprez L, et al. A review in radiomics: Making personalized medicine a reality via routine imaging [J]. Med Res Rev. 2022;42(1):426–40. Meng Y, Yang Y, Hu M, et al. Artificial intelligence-based radiomics in bone tumors: Technical advances and clinical application [J]. Semin Cancer Biol. 2023;95:75–87. Choi RY, Coyner AS, Kalpathy-Cramer J, et al. Introduction to Machine Learning, Neural Networks, and Deep Learning [J]. Transl Vis Sci Technol. 2020;9(2):14. Li R, Li L, Xu Y, et al. Machine learning meets omics: applications and perspectives [J]. Brief Bioinform. 2022;23(1). 10.1093/bib/bbab460 . Shelley B, Shaw M. Machine learning and preoperative risk prediction: the machines are coming [J]. Br J Anaesth. 2024;133(5):925–30. Ciobanu-Caraus O, Aicher A, Kernbach JM, et al. A critical moment in machine learning in medicine: on reproducible and interpretable learning [J]. Acta Neurochir (Wien). 2024;166(1):14. Çubukçu HC, Topcu D, Yenice S. Machine learning-based clinical decision support using laboratory data [J]. Clin Chem Lab Med. 2024;62(5):793–823. Liang B, Burley G, Lin S, et al. Osteoporosis pathogenesis and treatment: existing and emerging avenues [J]. Cell Mol Biol Lett. 2022;27(1):72. Caraiani C, Petresc B, Dong Y, et al. Contraindications and adverse effects in abdominal imaging [J]. Med Ultrason. 2019;21(4):456–63. Mittendorff L, Young A, Sim J. A narrative review of current and emerging MRI safety issues: What every MRI technologist (radiographer) needs to know [J]. J Med Radiat Sci. 2022;69(2):250–60. Seah SJ, Yeo MH, Tan JH, et al. Early cement augmentation may be a good treatment option for pain relief for osteoporotic compression fractures: a systematic review and meta-analysis [J]. Eur Spine J. 2023;32(5):1751–62. Sheng L, Yang C, Chen Y et al. Machine Learning Combined with Radiomics Facilitating the Personal Treatment of Malignant Liver Tumors [J]. Biomedicines, 2023, 12(1). Xiao MJ, Pan YT, Tan JH, et al. Computed tomography-based radiomics combined with machine learning allows differentiation between primary intestinal lymphoma and Crohn's disease [J]. World J Gastroenterol. 2024;30(25):3155–65. Cai J, Shen C, Yang T, et al. MRI-based radiomics assessment of the imminent new vertebral fracture after vertebral augmentation [J]. Eur Spine J. 2023;32(11):3892–905. Nian S, Zhao Y, Li C, et al. Development and validation of a radiomics-based model for predicting osteoporosis in patients with lumbar compression fractures [J]. Spine J. 2024;24(9):1625–34. Computer-aided detection. and diagnosis/radiomics/machine learning/deep learning in medical imaging [J]. Med Phys. 2023;50(Suppl 1):50–3. Huang EP, O'Connor JPB, McShane LM, et al. Criteria for the translation of radiomics into clinically useful tests [J]. Nat Rev Clin Oncol. 2023;20(2):69–82. Huang AA, Huang SY. Increasing transparency in machine learning through bootstrap simulation and shapely additive explanations [J]. PLoS ONE. 2023;18(2):e0281922. Charoenngam N, Shirvani A, Holick MF. Vitamin D for skeletal and non-skeletal health: What we should know [J]. J Clin Orthop Trauma. 2019;10(6):1082–93. Reid IR, Bolland MJ. Calcium and/or Vitamin D Supplementation for the Prevention of Fragility Fractures: Who Needs It? [J]. Nutrients, 2020, 12(4). Reid IR, EXTENSIVE EXPERTISE IN ENDOCRINOLOGY. Osteoporosis management [J]. Eur J Endocrinol. 2022;187(4):R65–80. Tables Table 1 Clinical characteristics of the training and validation groups Variables Training set (n=597) P Validation set (n=255) P Non-OVCF group (n = 419) OVCF group (n = 178) Non-OVCF group (n = 170) OVCF group (n = 85) Gender, n (%) < 0.001 0.168 Female 202 (48%) 127 (71%) 87 (51%) 52 (61%) Male 217 (52%) 51 (29%) 83 (49%) 33 (39%) Smoking, n (%) < 0.001 0.002 No 224 (53%) 51 (29%) 96 (56%) 30 (35%) Yes 195 (47%) 127 (71%) 74 (44%) 55 (65%) Drinking, n (%) 0.48 0.209 No 302 (72%) 134 (75%) 130 (76%) 58 (68%) Yes 117 (28%) 44 (25%) 40 (24%) 27 (32%) Diabetes, n (%) 0.112 0.278 No 339 (81%) 133 (75%) 135 (79%) 73 (86%) Yes 80 (19%) 45 (25%) 35 (21%) 12 (14%) Hypertension, n (%) 0.123 0.683 No 263 (63%) 99 (56%) 102 (60%) 54 (64%) Yes 156 (37%) 79 (44%) 68 (40%) 31 (36%) Thyroid disease, n (%) 0.65 0.384 No 363 (87%) 151 (85%) 137 (81%) 73 (86%) Yes 56 (13%) 27 (15%) 33 (19%) 12 (14%) Trauma, n (%) < 0.001 < 0.001 No 302 (72%) 71 (40%) 118 (69%) 30 (35%) Yes 117 (28%) 107 (60%) 52 (31%) 55 (65%) Lumbar surgery, n (%) < 0.001 < 0.001 No 395 (94%) 135 (76%) 159 (94%) 61 (72%) Yes 24 (6%) 43 (24%) 11 (6%) 24 (28%) Living area, n (%) < 0.001 < 0.001 Village 338 (81%) 79 (44%) 145 (85%) 34 (40%) City 81 (19%) 99 (56%) 25 (15%) 51 (60%) Calcium and Phosphate Metabolism, n (%) 0.17 0.305 No 382 (91%) 155 (87%) 158 (93%) 75 (88%) Yes 37 (9%) 23 (13%) 12 (7%) 10 (12%) Glucocorticoid, n (%) 0.006 0.002 No 358 (85%) 167 (94%) 136 (80%) 81 (95%) Yes 61 (15%) 11 (6%) 34 (20%) 4 (5%) Age 68 (65, 71) 70 (66, 76) < 0.001 68 (64, 71) 71 (67, 76) < 0.001 BMI 24.97 (22.75, 26.89) 25.38 (22.89, 27.65) 0.072 25.24 (22.87, 27.37) 24.68 (22.74, 27.55) 0.619 VAS 4.5 (4, 5) 5 (4, 5.5) 0.02 4.55 (4, 5) 5 (4, 5.5) 0.282 Course 10 (4, 20) 10 (4, 20) 0.408 10 (3, 27.75) 10 (3, 31) 0.99 OVCF:osteoporotic vertebral compression fractures; BMI :Body-Mass-Index; VAS:visual analogue scale Table 2 DeLong test for five machine learning algorithms in the validation group of the combined model group Model（AUC Value） Xgboost （0.826） KNN （0.831） SVM （0.804） LightGBM （0.743） RF （0.750） Xgboost（0.826） 1 - - - - KNN（0.831） 0.084 1 - - - SVM（0.804） 0.096 0.843 1 - - LightGBM（0.743） 0.020 0.021 0.003 1 - RF（0.750） 0.027 0.030 0.004 0.390 1 Table 3 Diagnostic performance of five machine learning algorithms in the combined model group Model Training set Validation set Accuracy Sensitivity Specificity Precision F1 score Accuracy Sensitivity Specificity Precision F1 score SVM 0.809 0.659 0.877 0.705 0.682 0.756 0.692 0.784 0.587 0.635 RF 0.988 0.968 0.998 0.994 0.981 0.685 0.808 0.631 0.492 0.612 Xgboost 0.814 0.697 0.867 0.701 0.699 0.754 0.731 0.795 0.613 0.667 KNN 0.888 0.919 0.874 0.766 0.835 0.776 0.756 0.784 0.608 0.674 LightGBM 0.849 0.751 0.893 0.76 0.755 0.762 0.705 0.807 0.618 0.659 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-6127302","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":423890820,"identity":"061a7885-622f-44f2-857b-d15db5e80716","order_by":0,"name":"Kangen Han","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyUlEQVRIiWNgGAWjYBAC+/vvHz7+Y/Bfjp+9gVg9B3KYDXgqmI0lew4Qr4VNgOcMc+KGGwlE6mBsOHuMQbKNzVhy5uONNxhqbKIJamFm7Et7YNjGI8cvnVZswXAsLbeBkBY2ZgZzg8Q2CWPJ2TlmEowNhwlr4WFjMJM42GaQuOHmGSK1SPDwmEk2nEkAep+HSC0GEmzJxgwVB4CBDPRLAjF+MZBgPviYweAAMCoPb7zxocaGsBZU7QmkKIdoIVXHKBgFo2AUjAwAANN7Pg14ianJAAAAAElFTkSuQmCC","orcid":"","institution":"Dalian Medical University","correspondingAuthor":true,"prefix":"","firstName":"Kangen","middleName":"","lastName":"Han","suffix":""},{"id":423890821,"identity":"6210189e-aa34-4a8b-ac1e-29a960e0fdba","order_by":1,"name":"Hongwen Gu","email":"","orcid":"","institution":"Department of Orthopedics, General Hospital of Northern Theater Command of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Hongwen","middleName":"","lastName":"Gu","suffix":""},{"id":423890822,"identity":"a47d9493-baea-4dfe-9136-e3c9f2c195f1","order_by":2,"name":"Yu Li","email":"","orcid":"","institution":"Department of Orthopedics, General Hospital of Northern Theater Command of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Yu","middleName":"","lastName":"Li","suffix":""},{"id":423890824,"identity":"0d59e071-69e4-4141-a998-15b970dfef3f","order_by":3,"name":"Junchao LI","email":"","orcid":"","institution":"Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Junchao","middleName":"","lastName":"LI","suffix":""},{"id":423890825,"identity":"70afa6ae-f5fb-4583-a650-7a2ca2ac6c14","order_by":4,"name":"Zhihao Zhang","email":"","orcid":"","institution":"Department of Orthopedics, General Hospital of Northern Theater Command of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Zhihao","middleName":"","lastName":"Zhang","suffix":""},{"id":423890826,"identity":"3a6f4088-7d38-4dfe-8a7f-6dc309108f62","order_by":5,"name":"Yin Hu","email":"","orcid":"","institution":"Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yin","middleName":"","lastName":"Hu","suffix":""},{"id":423890828,"identity":"705e5aa7-5a0f-4fea-b2ef-70c83c44f28e","order_by":6,"name":"Le Xing","email":"","orcid":"","institution":"Department of Orthopedics, General Hospital of Northern Theater Command of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Le","middleName":"","lastName":"Xing","suffix":""},{"id":423890830,"identity":"4a6ee9d6-6fad-4f89-ad4f-b6eaef5c31ad","order_by":7,"name":"Hailong Yu","email":"","orcid":"","institution":"Department of Orthopedics, General Hospital of Northern Theater Command of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Hailong","middleName":"","lastName":"Yu","suffix":""},{"id":423890832,"identity":"4367abef-ef42-437b-bc0b-8996e17f219b","order_by":8,"name":"Hongwei Wang","email":"","orcid":"","institution":"Department of Orthopedics, General Hospital of Northern Theater Command of Chinese PLA","correspondingAuthor":false,"prefix":"","firstName":"Hongwei","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2025-02-28 09:38:37","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6127302/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6127302/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":78240531,"identity":"023bd8cc-3840-4692-9310-28648d61bf69","added_by":"auto","created_at":"2025-03-11 08:57:14","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":182562,"visible":true,"origin":"","legend":"\u003cp\u003eThe patient enrollment process for this study\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/62a60df2b2f05b5e798e45de.jpg"},{"id":78241504,"identity":"4b4e2f0b-944d-4ff6-8624-694760473d17","added_by":"auto","created_at":"2025-03-11 09:05:14","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":422254,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea:\u003c/strong\u003e L1 vertebral compression fracture, difficult to recognize by the human eye; \u003cstrong\u003eb:\u003c/strong\u003e L1 vertebral hypointensity was seen in T1WI; \u003cstrong\u003ec:\u003c/strong\u003e MRI lipid suppression image showed L1 vertebral hyperintensity and compression fracture; \u003cstrong\u003ed:\u003c/strong\u003e3Dslicer segmentation of the compressed vertebrae; \u003cstrong\u003ee:\u003c/strong\u003e normal lumbar spine X-ray;\u003cstrong\u003e f.g:\u003c/strong\u003e MRI of normal lumbar spine; \u003cstrong\u003eh:\u003c/strong\u003e 3Dslicer divides the normal vertebral body\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/6996821713aaf3f59b376577.jpg"},{"id":78243722,"identity":"ca7223b9-b986-4dd5-847b-d6fa81f31775","added_by":"auto","created_at":"2025-03-11 09:13:14","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":210323,"visible":true,"origin":"","legend":"\u003cp\u003eFeature screening process for LASSO regression\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/ef1fe6e10a1606a1a1d2867e.jpg"},{"id":78241505,"identity":"33e596d0-3132-474c-a16d-bed47c40922a","added_by":"auto","created_at":"2025-03-11 09:05:14","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":249456,"visible":true,"origin":"","legend":"\u003cp\u003eVariable importance coefficient\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/c1ed65021a45145c3a5b45e7.jpg"},{"id":78243718,"identity":"3aee0ae3-121c-4d91-9d93-b66b503803b4","added_by":"auto","created_at":"2025-03-11 09:13:14","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":254980,"visible":true,"origin":"","legend":"\u003cp\u003eVariables correspond to P-values and ratios to forest plots\u003c/p\u003e","description":"","filename":"Figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/1416b5c68a897e49dd415804.jpg"},{"id":78246386,"identity":"94781e0c-ac3c-4b69-a774-d1c3dc7d4ae0","added_by":"auto","created_at":"2025-03-11 09:29:14","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":296271,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of the training group and the validation group of the clinical prediction model group\u003c/p\u003e","description":"","filename":"Figure6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/427a8648b29a39a5652fc276.jpg"},{"id":78241509,"identity":"ddb87367-c0a2-464a-878a-62f02daa02af","added_by":"auto","created_at":"2025-03-11 09:05:14","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":296155,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves for the training and validation groups of the radiomics model group\u003c/p\u003e","description":"","filename":"Figure7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/afb2dd41232bb94b2c7a11ba.jpg"},{"id":78243721,"identity":"f4aea41f-9a64-419c-86c7-f0f29cd791a6","added_by":"auto","created_at":"2025-03-11 09:13:14","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":196145,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves for the training and validation groups of the combined model group\u003c/p\u003e","description":"","filename":"Figure8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/39849ddbde863c32e35603ae.jpg"},{"id":78241510,"identity":"e27e283e-b1dd-41b8-8e29-51e093555e9b","added_by":"auto","created_at":"2025-03-11 09:05:14","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":157617,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration curves for the training and validation groups of the combined model group\u003c/p\u003e","description":"","filename":"Figure9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/4a52c55e0f6ad4a6cf519728.jpg"},{"id":78244770,"identity":"21725c3c-87d3-4a76-90ee-842888999440","added_by":"auto","created_at":"2025-03-11 09:21:14","extension":"jpg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":162948,"visible":true,"origin":"","legend":"\u003cp\u003eDCA curves for the training group and validation group of combined model group\u003c/p\u003e","description":"","filename":"Figure10.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/1baa75e3df1a79cb3ab25d22.jpg"},{"id":78240542,"identity":"e6923866-acbd-4a12-9fff-2b6372781a38","added_by":"auto","created_at":"2025-03-11 08:57:14","extension":"jpg","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":108977,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eUsing SHAP to visualize the model as a whole.\u003c/strong\u003e (A) SHAP histogram shows the weights of the six features in the model. (B) SHAP bee colony plot showing the overall trend of the data\u003c/p\u003e","description":"","filename":"Figure11.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/84520ba4ee54f2410b6e957c.jpg"},{"id":78720958,"identity":"ffa2de44-ceb5-4c6f-baaf-f4910465baa9","added_by":"auto","created_at":"2025-03-18 04:38:33","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3785329,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6127302/v1/29a30431-b9f8-42b9-b9b9-3b6e6b9b0cd5.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Research on Assisting X-ray Diagnosis of Osteoporotic Vertebral Compression Fractures Using Interpretable Machine Learning Models and Radiomics Features","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOsteoporosis is a metabolic disorder characterized by decreased bone mass and the deterioration of bone structure, with the spine being the most frequently affected area. It has been reported that osteoporotic vertebral compression fractures (OVCF) are on the rise, accounting for approximately 40% of all osteoporotic fractures. These fractures not only compromise quality of life but also impose significant economic burdens on both patients and society, particularly affecting the elderly \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. Early diagnosis of OVCF is crucial for the effective treatment of patients. While X-rays can suggest the presence of vertebral compression fractures by changes in vertebral shape, radiologists often find it challenging to detect subtle fractures, leading to high rates of misdiagnosis and delays in treatment \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. In cases where X-ray findings are inconclusive, MRI is often employed; however, its high costs, lengthy wait times, and considerable contraindications \u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e] make it unsuitable for all patients. Thus, enhancing the sensitivity of X-ray diagnosis for vertebral compression fractures remains an urgent clinical need. The advent of radiomics and machine learning offers new approaches. Radiomics involves analyzing a vast amount of high-dimensional data from imaging to extract and quantify features that are imperceptible to the human eye \u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. Machine learning leverages large datasets to learn patterns and enhance performance, enabling automatic feature recognition and classification \u003csup\u003e[\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. By integrating radiomics with machine learning algorithms, it is possible to develop diagnostic models that allow for the identification and analysis of subtle image features, facilitating disease diagnosis and prognosis. Historically, traditional machine learning has suffered from a lack of interpretability, often being regarded as a \"black box,\" which has led to concerns about its reliability \u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. This study aims to employ Shapley Additive exPlanations (SHAP) to provide a global interpretation of machine learning algorithms\u0026rsquo; internal operations, developing and validating a robust joint model based on machine learning and radiomics features for early, economical, and efficient OVCF diagnosis.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e\u003cstrong\u003e2.1 General Information\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study analyzed data from 852 patients collected between January 2016 and December 2023, including lateral X-rays of the L1 vertebra and demographic details such as gender, age, BMI, VAS score, symptom duration, smoking, alcohol consumption, diabetes, hypertension, thyroid disorders, trauma history, history of lumbar surgery, residential area, calcium-phosphate metabolism disorders, and glucocorticoid use. Of these, 589 presented with lumbar back pain but had normal MRI findings, while 263 were diagnosed with varying degrees of OVCF by MRI. The patients were randomly allocated into training (70%) and validation (30%) groups; the former was used for model development, and the latter for model validation. Inclusion criteria included: ① Age \u0026ge;60 years; ② Admission for lumbar back pain; ③ Complete lumbar spine X-ray and MRI imaging data. Exclusion criteria included: ① Compression of \u0026ge;2 vertebral bodies; ② Spinal malignancies, tuberculosis, ankylosing spondylitis, or other specific diseases; ③ Unclear X-ray images; ④ Difficulties in segmenting the Region of Interest (ROI) in compressed vertebral bodies，the flowchart is shown in Figure 1. This retrospective study was conducted without requiring signed informed consent from the participants and adhered to the principles of the Declaration of Helsinki.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.1\u003c/strong\u003e The patient enrollment process for this study\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 X-ray Image Collection and Segmentation\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDICOM format raw X-ray data were exported from our hospital\u0026apos;s A-site imaging system and manually segmented using 3Dslicer version 5.6.1 software (http://www.slicer.org/). Two spinal surgeons performed the segmentation on both normal and compressed L1 vertebrae, without prior knowledge of any fractures, as shown in Figure 2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.2\u003c/strong\u003e \u003cstrong\u003ea:\u003c/strong\u003e L1 vertebral compression fracture, difficult to recognize by the human eye; \u003cstrong\u003eb:\u003c/strong\u003e L1 vertebral hypointensity was seen in T1WI; \u003cstrong\u003ec:\u003c/strong\u003e MRI lipid suppression image showed L1 vertebral hyperintensity and compression fracture; \u003cstrong\u003ed:\u003c/strong\u003e 3Dslicer segmentation of the compressed vertebrae; \u003cstrong\u003ee:\u003c/strong\u003e normal lumbar spine X-ray;\u003cstrong\u003e\u0026nbsp;f.g:\u003c/strong\u003e MRI of normal lumbar spine; \u003cstrong\u003eh:\u003c/strong\u003e 3Dslicer divides the normal vertebral body\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Radiomics Feature Extraction\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study utilized the PyRadiomics package in Python for feature extraction. The images were resampled to 3mm \u0026times; 3mm \u0026times; 3mm voxels using the B-Spline interpolation method and normalized by histogram matching to reduce variability across different machines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4 Feature Selection\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOur feature selection process involved three steps: (1) Initial standardization of features using the Z-score method (X score = (X\u0026minus;\u0026mu;)/\u0026sigma;, where \u0026mu; is the mean, and \u0026sigma; is the standard deviation). Preliminary filtering of features between two groups was performed using the t-test and Mann-Whitney U test for those with p \u0026lt; 0.05. (2) Intraclass correlation coefficients (ICCs) were utilized to eliminate individual measurement differences among doctors and for further feature selection. Sixty patients were randomly selected, and their ROIs were manually delineated by two spinal surgeons, A and B. Surgeon B redelineated the ROIs two weeks later; features with ICCs \u0026gt; 0.75, both within and between observers, were included for further analysis. (3) Lastly, LASSO regression was employed to minimize some feature coefficients, using only non-zero coefficients for modeling. Ten-fold cross-validation was used to optimize the regularization parameter (\u0026lambda;). Each patient\u0026apos;s radiomics score was calculated using the formula: Radscore = \u0026beta;0 + \u0026beta;1F1 + \u0026beta;2F2 + ... + \u0026beta;nFn, where Fi represents the features and \u0026beta;i are the coefficients from LASSO regression.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.6 Experimental Design\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFive machine learning algorithms were used for modeling, including Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (Xgboost), K-Nearest Neighbors (KNN), and LightGBM. Models were established based on the training group: (1) Clinical prediction model group: Variables showing intergroup differences in demographic analysis were incorporated into a multivariate logistic regression to construct the model. (2) Radiomics model group: Meaningful variables identified via final LASSO regression analysis were used for modeling. (3) Combined model group: Meaningful clinical variables and radiomics scores were utilized to develop a combined predictive model. The machine learning classifier with the highest diagnostic performance was selected, and the SHAP algorithm was employed to interpret its internal computational processes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.7 Evaluation Criteria\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe evaluation criteria for the optimal model include: ① Receiver Operating Characteristic (ROC) curves, the Area Under the Curve (AUC), and the plotting of calibration curves to assess how closely the model approximates an ideal model, supplemented by Decision Curve Analysis (DCA) to evaluate the model\u0026apos;s clinical utility. The DeLong test is utilized to compare differences among the ROC curves of the five machine learning methods. ② The model\u0026apos;s predictive accuracy is assessed using metrics such as Accuracy, Sensitivity, Specificity, Precision, and F1 Score.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.8 Statistical Methods\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFeature extraction was conducted using the PyRadiomics package in Python (version 3.9.13), with statistical analyses performed using R statistical software (version 4.4.1). Clinical data from the training and validation groups were analyzed based on the type of variable. Quantitative data adhering to a normal distribution were expressed as mean \u0026plusmn; standard deviation and analyzed using the t-test; non-normally distributed quantitative data were presented as medians (Median) and interquartile ranges (IQR), and analyzed using the Mann-Whitney U test. Categorical data were presented as numbers and percentages (N, %). A p-value \u0026lt; 0.05 was considered to indicate statistical significance.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003e3.1 Clinical Demographics\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe training group included 597 individuals, and the validation group comprised 255. Variables that exhibited significant differences within the training group included gender, smoking history, trauma history, history of lumbar surgery, residential area, history of glucocorticoid treatment, age, and VAS score, as detailed in Table 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Radiomics Features\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eA total of 862 features were extracted, of which:(1)First‐order statistics (n = 173); (2) Shape (n = 14); (3) Gray Level Co‐occurrence Matrix (n = 216); (4) Gray Level Run-Length Matrix (n = 144); (5) Gray - Level Size Zone Matrix (n = 144); (6) Gray - Level Dependence Matrix(n = 126); (7) Neighborhood Gray - Tone Difference Matrix (n = 45), and 3 types of filters, including:Laplacian of Gaussian（LoG），wavelets and LBP2D。The t-test and Mann-Whitney U test were employed to initially select 256 features. Applying an ICC threshold of \u0026gt;0.75, 188 features were subsequently selected. Variables were then further refined using LASSO regression, where an increase in \u0026lambda; progressively reduced the coefficients of less contributive features to zero, thus enabling variable selection. The coefficient path graph illustrates that with higher \u0026lambda; values, fewer variables are retained. Ultimately, eight radiomics features were finalized, as depicted in Figure3.\u003c/p\u003e\n\u003cp\u003eFig.3 Feature screening process for LASSO regression\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Radiomics Score\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe importance coefficients for the eight variables identified through LASSO regression are displayed in Figure 4. The formula for the radiomics score is established as follows:\u003c/p\u003e\n\u003cp\u003eRadscore=3.4035-0.000003\u0026times;diagnostics_Mask-original_VoxelNum-8.0844\u0026times;original_shape_Elongation-0.0031\u0026times;wavelet-LHL_firstorder_Median-0.8402\u0026times;wavelet-LHH_glszm_LowGrayLevelZoneEmphasis+3.2912\u0026times;wavelet-HLL_glrlm_ShortRunLowGrayLevelEmphasis+0.0283\u0026times;wavelet-HHL_ngtdm_Contrast+3.1581\u0026times;wavelet-HHH_ngtdm_Strength+1.2240\u0026times;wavelet-LLL_glcm_MCC\u003c/p\u003e\n\u003cp\u003eFig.4 Variable importance coefficient\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Research Variables\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the training group, variables that showed significant intergroup differences and were meaningful along with the radiomics score were included in a multivariate logistic regression analysis to determine the final variables used for modeling. These included gender, smoking history, trauma history, history of lumbar surgery, residential area, and Radscore, as illustrated in Figure 5.\u003c/p\u003e\n\u003cp\u003eFig.5 Variables correspond to P-values and ratios to forest plots\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Model Comparison\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(1) Clinical Prediction Model Group: Using five machine learning algorithms, meaningful clinical variables from the multivariate analysis were employed for modeling. LightGBM exhibited the best performance in the training group with an AUC of 0.862 (95% CI=0.829-0.895), as shown in Figure 6. (2) Radiomics Model Group: Models were built using the final selected radiomics features, among which Xgboost achieved the best result with an AUC of 0.767 (95% CI=0.727-0.808), as displayed in Figure 7. (3) Combined Model Group: A predictive model was constructed combining clinical variables and the radiomics score. Random Forest demonstrated exceptional predictive performance, achieving an AUC of 0.983 (95% CI=0.970-0.996), as depicted in Figure 8. Ultimately, the combined model outperformed the other models. Calibration of the combined model was assessed in the validation group, where Xgboost, KNN, and LightGBM fitted well, as shown in Figure 9. Decision curves indicated that Xgboost, KNN, and SVM were clinically beneficial, potentially enhancing patient outcomes in medical practices, as demonstrated in Figure 10. A comparison of AUC values among each pair of classifiers in the validation group and p-values obtained via the DeLong test revealed no significant differences between Xgboost, KNN, and SVM, but these were significantly different from LightGBM and RF, as noted in Table 2. When comparing the diagnostic performance of the five machine learning algorithms in the combined model between the training and validation groups, Random Forest had an accuracy of 0.998 in the training group but only 0.685 in the validation group, indicating overfitting. KNN, however, exhibited the most stable and balanced performance in the validation group, with an AUC of 0.831 and an accuracy of 0.776, making it the best machine learning method for this study. The model established using KNN showed the highest diagnostic efficacy, as detailed in Table 3.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.6 ROC curves of the training group and the validation group of the clinical prediction model group\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.7 ROC curves for the training and validation groups of the radiomics model group\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.8 ROC curves for the training and validation groups of the combined model group\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.9 Calibration curves for the training and validation groups of the combined model group\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.10 DCA curves for the training group and validation group\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eof\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003ecombined model group\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.6 SHAP Interpretation of the KNN Algorithm\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo understand the internal classification mechanisms of the KNN, we calculated the Shapley values for the combined model interpretations using the KNN algorithm. The SHAP bar graphs provide a visual representation of the weights of the model\u0026rsquo;s six most significant features: residential area, Radscore, trauma history, gender, smoking, and history of lumbar surgery. The residential area had the greatest weight, followed by Radscore. The SHAP bee swarm plot uses purple and yellow to indicate each feature\u0026apos;s positive or negative impact on the prediction probability. Positive influences are shown by the residential area, trauma history, smoking, and lumbar surgery history, while gender exhibits a negative influence. Radscore, although not showing a clear trend, has the broadest distribution range, highlighting its significant impact on predictions, as shown in Figure 11.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig.11 Using SHAP to visualize the model as a whole.\u003c/strong\u003e (A) SHAP histogram shows the weights of the six features in the model. (B) SHAP bee colony plot showing the overall trend of the data\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eAs global aging trends intensify, osteoporosis diagnoses are increasing, with nearly 200 million new cases annually. Osteoporotic vertebral compression fractures (OVCF) place unprecedented strain on healthcare systems \u003csup\u003e[13]\u003c/sup\u003e. Currently, X-rays serve as the primary diagnostic tool for OVCF, with radiologists relying on vertebral height and deformation for evaluation\u003csup\u003e[4]\u003c/sup\u003e. However, minor compression fractures often go undetected when solely relying on visual inspection. For cases where suspicion remains high, further MRI scans are typically recommended to investigate internal changes within the vertebrae. Yet, due to high costs, lengthy procedures, and strict contraindications, many patients are deterred from undergoing MRI scans \u003csup\u003e[14, 15]\u003c/sup\u003e, leading to delays in diagnosis. Consequently, patients miss the window for timely surgical interventions and face prolonged bed rest, which can lead to severe complications such as pneumonia, bedsores, thrombosis, and even life-threatening conditions \u003csup\u003e[16]\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eIn recent years, the integration of radiomics and machine learning has brought significant advancements in predictive modeling for OVCF\u003csup\u003e\u0026nbsp;[17, 18]\u003c/sup\u003e. These integrated models enhance diagnostic capabilities by incorporating subtle imaging features undetectable by the human eye. In previous research, Cai et al. utilized MRI-based radiomics to assess new vertebral fractures post-vertebroplasty, achieving an Area Under the Curve (AUC) of over 0.9 in both training and validation groups, showcasing remarkable predictive power \u003csup\u003e[19]\u003c/sup\u003e. Similarly, Sun et al. applied radiomics features in-depth and utilized a Naive Bayes machine learning model to diagnose osteoporosis, achieving an accuracy of 0.81 and a sensitivity of 0.87 \u003csup\u003e[20]\u003c/sup\u003e. These findings confirm that applying well-trained radiomics models for clinical image recognition is not only feasible but also surpasses manual recognition in both precision and sensitivity.\u003c/p\u003e\n\u003cp\u003eThese results strongly support the potential of combined radiomics and machine learning models to significantly improve early diagnosis rates for OVCF. Notably, prior radiomics diagnostics rarely incorporated clinical factors \u003csup\u003e[7, 21, 22]\u003c/sup\u003e. This study identified gender, smoking history, trauma history, history of lumbar surgery, and residential area as independent risk factors for OVCF through multivariate logistic regression analysis. Therefore, our research was segmented into clinical prediction, radiomics prediction, and combined prediction groups, employing five cutting-edge machine learning algorithms to explore differences among these groups. The combined model of radiomics emerged with superior performance. Due to overfitting issues in the Random Forest algorithm, the KNN algorithm, with its excellent generalization ability, demonstrated outstanding performance.\u003c/p\u003e\n\u003cp\u003eIn machine learning, maintaining a balance between accuracy and model interpretability is crucial \u003csup\u003e[23]\u003c/sup\u003e. Deeply understanding the complex decision-making mechanisms of models increases their credibility, which is vital for their widespread clinical application. Consequently, this study employed Shapley values to interpret the best-performing KNN algorithm. It was discovered that residential area was the most significant factor, closely followed by the radiomics score, with only a slight difference between the two. Urban residents are more susceptible to OVCF, likely due to lower physical activity levels associated with sedentary work and insufficient outdoor exercise. Concurrently, the prevalence of smoking and excessive drinking in urban populations, which can disrupt bone metabolism, should not be underestimated. Rural residents, who often engage in more physical labor and have more exposure to sunlight, enabling more efficient vitamin D synthesis, are at a lower risk for OVCF \u003csup\u003e[24-26]\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eIn conclusion, the results of this study provide strong justification for the belief that the combined model of radiomics and machine learning possesses impressive diagnostic efficacy, fully capable of being applied in clinical practice. It significantly enhances the detection rate of OVCF in X-rays, drastically reducing the rate of missed diagnoses and offering new alternatives for patients who cannot undergo MRI scans. This study has limitations, including the intentional avoidance of abnormal structures during data collection, which may impact the results. Additionally, being a single-center study without external validation, the variability in X-ray machines, and the relatively small sample size may have led to some overfitting. Future research should involve larger sample sizes to reduce overfitting and enhance model performance. Despite these limitations, our innovative application of radiomics for the early diagnosis of OVCF via X-ray marks a significant advancement. As radiomics continues to evolve, we anticipate that data from radiological examinations worldwide will be transformed into quantitative features, building databases from extensive radiomics data from millions of patients to enhance the accuracy and predictive capabilities of models, thereby providing robust support for clinical diagnosis of OVCF assisted by models.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eHuman Ethics and Consent to Participate declarations:\u003c/strong\u003e not applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInternal Review Board (IRB):\u0026nbsp;\u003c/strong\u003eEthical approval was obtained from the Institutional Review Board(IRB) of General Hospital of Northern Theater Command.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial number:\u0026nbsp;\u003c/strong\u003enot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial number:\u0026nbsp;\u003c/strong\u003enot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate:\u0026nbsp;\u003c/strong\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u003c/strong\u003e All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and material:\u003c/strong\u003e The data that support the findings of this study are available from the corresponding author on special request\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e This work was supported by Liaoning Province Applied Basic Research Programme (2023JH2/101700130)；Liaoning Province Applied Basic Research Plan (2022JH2/101300024)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u0026nbsp;\u003c/strong\u003eAll listed authors have made substantial contributions to the manuscript and do not have any conflicts of interest.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions:\u003c/strong\u003e Kangen Han, Yu Li, Hongwen Gu, Yin Hu, Shilei Tang, Zhihao Zhang, Hailong Yu, Hongwei Wang designed and participated in the whole process of the \u0026nbsp;study and drafted the manuscript. All authors read and approved the final manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed consent:\u0026nbsp;\u003c/strong\u003eEthical approval was obtained from the Institutional Review Board(IRB) of General Hospital of Northern Theater Command. Chinese law does not require individual informed consent from participants in non-invasive observational trials such as the present study. Therefore, the need for informed consent was waived according to the instruction of IRB of General Hospital of Northern Theater Command.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u003c/strong\u003e All listed authors have made substantial contributions to the manuscript and do not have any conflicts of interest. All authors read and approved the final manuscript. This work was supported by Applied Basic Research Project of Liaoning Province\u003c/p\u003e\n\u003cp\u003e(2023JH2/101700130) and Liaoning Province Applied Basic Research Plan (2022JH2/101300024)\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eYang XG, Dong YQ, Liu X, et al. Incidence and prognostic factors of residual back pain in patients treated for osteoporotic vertebral compression fractures: a systematic review and meta-analysis [J]. Eur Spine J. 2024;33(12):4521\u0026ndash;37.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng XQ, Xu L, Huang J, et al. Incidence and cost of vertebral fracture in urban China: a 5-year population-based cohort study [J]. Int J Surg. 2023;109(7):1910\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePatel D, Liu J, Ebraheim NA. Managements of osteoporotic vertebral compression fractures: A narrative review [J]. World J Orthop. 2022;13(6):564\u0026ndash;73.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpiegl U, Bork H, Gr\u0026uuml;ninger S, et al. Osteoporotic Fractures of the Thoracic and Lumbar Vertebrae: Diagnosis and Conservative Treatment [J]. Dtsch Arztebl Int. 2021;118(40):670\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYamamoto Y, Iwata E, Shigematsu H, et al. Differential diagnosis between metastatic and osteoporotic vertebral fractures using sagittal T1-weighted magnetic resonance imaging [J]. J Orthop Sci. 2020;25(5):763\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuiot J, Vaidyanathan A, Deprez L, et al. A review in radiomics: Making personalized medicine a reality via routine imaging [J]. Med Res Rev. 2022;42(1):426\u0026ndash;40.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeng Y, Yang Y, Hu M, et al. Artificial intelligence-based radiomics in bone tumors: Technical advances and clinical application [J]. Semin Cancer Biol. 2023;95:75\u0026ndash;87.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi RY, Coyner AS, Kalpathy-Cramer J, et al. Introduction to Machine Learning, Neural Networks, and Deep Learning [J]. Transl Vis Sci Technol. 2020;9(2):14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi R, Li L, Xu Y, et al. Machine learning meets omics: applications and perspectives [J]. Brief Bioinform. 2022;23(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/bib/bbab460\u003c/span\u003e\u003cspan address=\"10.1093/bib/bbab460\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShelley B, Shaw M. Machine learning and preoperative risk prediction: the machines are coming [J]. Br J Anaesth. 2024;133(5):925\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCiobanu-Caraus O, Aicher A, Kernbach JM, et al. A critical moment in machine learning in medicine: on reproducible and interpretable learning [J]. Acta Neurochir (Wien). 2024;166(1):14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e\u0026Ccedil;ubuk\u0026ccedil;u HC, Topcu D, Yenice S. Machine learning-based clinical decision support using laboratory data [J]. Clin Chem Lab Med. 2024;62(5):793\u0026ndash;823.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiang B, Burley G, Lin S, et al. Osteoporosis pathogenesis and treatment: existing and emerging avenues [J]. Cell Mol Biol Lett. 2022;27(1):72.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaraiani C, Petresc B, Dong Y, et al. Contraindications and adverse effects in abdominal imaging [J]. Med Ultrason. 2019;21(4):456\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMittendorff L, Young A, Sim J. A narrative review of current and emerging MRI safety issues: What every MRI technologist (radiographer) needs to know [J]. J Med Radiat Sci. 2022;69(2):250\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeah SJ, Yeo MH, Tan JH, et al. Early cement augmentation may be a good treatment option for pain relief for osteoporotic compression fractures: a systematic review and meta-analysis [J]. Eur Spine J. 2023;32(5):1751\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSheng L, Yang C, Chen Y et al. Machine Learning Combined with Radiomics Facilitating the Personal Treatment of Malignant Liver Tumors [J]. Biomedicines, 2023, 12(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiao MJ, Pan YT, Tan JH, et al. Computed tomography-based radiomics combined with machine learning allows differentiation between primary intestinal lymphoma and Crohn's disease [J]. World J Gastroenterol. 2024;30(25):3155\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCai J, Shen C, Yang T, et al. MRI-based radiomics assessment of the imminent new vertebral fracture after vertebral augmentation [J]. Eur Spine J. 2023;32(11):3892\u0026ndash;905.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNian S, Zhao Y, Li C, et al. Development and validation of a radiomics-based model for predicting osteoporosis in patients with lumbar compression fractures [J]. Spine J. 2024;24(9):1625\u0026ndash;34.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eComputer-aided detection. and diagnosis/radiomics/machine learning/deep learning in medical imaging [J]. Med Phys. 2023;50(Suppl 1):50\u0026ndash;3.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang EP, O'Connor JPB, McShane LM, et al. Criteria for the translation of radiomics into clinically useful tests [J]. Nat Rev Clin Oncol. 2023;20(2):69\u0026ndash;82.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang AA, Huang SY. Increasing transparency in machine learning through bootstrap simulation and shapely additive explanations [J]. PLoS ONE. 2023;18(2):e0281922.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCharoenngam N, Shirvani A, Holick MF. Vitamin D for skeletal and non-skeletal health: What we should know [J]. J Clin Orthop Trauma. 2019;10(6):1082\u0026ndash;93.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReid IR, Bolland MJ. Calcium and/or Vitamin D Supplementation for the Prevention of Fragility Fractures: Who Needs It? [J]. Nutrients, 2020, 12(4).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReid IR, EXTENSIVE EXPERTISE IN ENDOCRINOLOGY. Osteoporosis management [J]. Eur J Endocrinol. 2022;187(4):R65\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e \u003cstrong\u003eClinical characteristics of the training and validation groups \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"102%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 18px;\"\u003e\n \u003cp\u003eVariables\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 33px;\"\u003e\n \u003cp\u003eTraining set\u003cbr\u003e\u0026nbsp;(n=597)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width: 10px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 28px;\"\u003e\n \u003cp\u003eValidation set\u003cbr\u003e\u0026nbsp;(n=255)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width: 9px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 21px;\"\u003e\n \u003cp\u003eNon-OVCF\u0026nbsp;\u003c/p\u003e\n \u003cp\u003egroup\u003cbr\u003e\u0026nbsp;(n = 419)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003eOVCF group\u003cbr\u003e\u0026nbsp;(n = 178)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003eNon-OVCF group\u003cbr\u003e\u0026nbsp;(n = 170)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003eOVCF group\u003cbr\u003e\u0026nbsp;(n = 85)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eGender, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.168\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e202 (48%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e127 (71%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e87 (51%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e52 (61%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e217 (52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e51 (29%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e83 (49%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e33 (39%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eSmoking, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e224 (53%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e51 (29%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e96 (56%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e30 (35%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e195 (47%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e127 (71%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e74 (44%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e55 (65%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eDrinking, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.209\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e302 (72%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e134 (75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e130 (76%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e58 (68%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e117 (28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e44 (25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e40 (24%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e27 (32%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eDiabetes, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.112\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.278\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e339 (81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e133 (75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e135 (79%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e73 (86%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e80 (19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e45 (25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e35 (21%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e12 (14%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eHypertension, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.123\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.683\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e263 (63%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e99 (56%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e102 (60%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e54 (64%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e156 (37%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e79 (44%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e68 (40%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e31 (36%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eThyroid disease, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.384\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e363 (87%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e151 (85%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e137 (81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e73 (86%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e56 (13%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e27 (15%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e33 (19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e12 (14%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eTrauma, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e302 (72%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e71 (40%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e118 (69%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e30 (35%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e117 (28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e107 (60%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e52 (31%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e55 (65%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eLumbar surgery, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e395 (94%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e135 (76%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e159 (94%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e61 (72%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e24 (6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e43 (24%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e11 (6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e24 (28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eLiving area, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eVillage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e338 (81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e79 (44%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e145 (85%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e34 (40%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eCity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e81 (19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e99 (56%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e25 (15%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e51 (60%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eCalcium and Phosphate Metabolism, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.305\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e382 (91%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e155 (87%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e158 (93%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e75 (88%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e37 (9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e23 (13%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e12 (7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e10 (12%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eGlucocorticoid, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e358 (85%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e167 (94%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e136 (80%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e81 (95%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e61 (15%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e11 (6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e34 (20%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e4 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e68 (65, 71)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e70 (66, 76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e68 (64, 71)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e71 (67, 76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e24.97\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(22.75, 26.89)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e25.38\u003c/p\u003e\n \u003cp\u003e(22.89, 27.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.072\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e25.24 (22.87, 27.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e24.68 (22.74, 27.55)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.619\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eVAS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e4.5\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e5\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(4, 5.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e4.55\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(4, 5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e5\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(4, 5.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.282\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eCourse\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e10\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(4, 20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 16px;\"\u003e\n \u003cp\u003e10\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(4, 20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0.408\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;(3, 27.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e10\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(3, 31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 221px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 23px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 51px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 112px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 112px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 51px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eOVCF:osteoporotic vertebral compression fractures; BMI :Body-Mass-Index; VAS:visual analogue scale\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2 DeLong test for five machine learning algorithms in the validation group of the combined model group\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"562\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eModel（AUC Value）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003eXgboost\u003cbr\u003e\u0026nbsp;（0.826）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003eKNN\u003cbr\u003e\u0026nbsp;（0.831）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 99px;\"\u003e\n \u003cp\u003eSVM\u003cbr\u003e\u0026nbsp;（0.804）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003eLightGBM\u003cbr\u003e\u0026nbsp;（0.743）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003eRF\u003cbr\u003e\u0026nbsp;（0.750）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eXgboost（0.826）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 99px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eKNN（0.831）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.084\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 99px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eSVM（0.804）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.096\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e0.843\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 99px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eLightGBM（0.743）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.020\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e0.021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 99px;\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 138px;\"\u003e\n \u003cp\u003eRF（0.750）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.027\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e0.030\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 99px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e0.390\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 78px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3 Diagnostic performance of five machine learning algorithms in the combined model group\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"108%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 9px;\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"5\" style=\"width: 45px;\"\u003e\n \u003cp\u003eTraining set\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"5\" style=\"width: 45px;\"\u003e\n \u003cp\u003eValidation set\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.809\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.659\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.877\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.705\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.682\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.756\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.692\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.587\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.635\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.988\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.968\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.998\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.994\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.981\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.685\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.808\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.631\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.492\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.612\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eXgboost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.814\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.697\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.867\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.701\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.699\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.754\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.731\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.795\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.613\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.667\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.888\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.919\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.874\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.766\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.835\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.776\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.756\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.608\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.674\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003eLightGBM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.849\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.751\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.893\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.755\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.762\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.705\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.807\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.618\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9px;\"\u003e\n \u003cp\u003e0.659\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Radiomics, machine learning, osteoporotic vertebral compression fractures, X-ray","lastPublishedDoi":"10.21203/rs.3.rs-6127302/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6127302/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eObjective:\u003c/b\u003e\u003c/p\u003e \u003cp\u003eTo improve early diagnosis rates, this study applies a combination of radiomics and machine learning algorithms to aid in the X-ray diagnosis of osteoporotic vertebral compression fractures (OVCF).\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods:\u003c/b\u003e\u003c/p\u003e \u003cp\u003eData were collected from 852 patients from January 2016 to December 2023, including lateral X-rays of the L1 vertebra and demographic information. The cohort included 589 patients with lumbar back pain but normal MRI results, and 263 patients diagnosed with various degrees of OVCF by MRI. Patients were randomly divided into training (70%) and validation (30%) groups. X-ray images were annotated to extract radiomics features, which were then selected to finalize the radiomics score, along with meaningful clinical factors. Five machine learning algorithms were utilized to model and compare the diagnostic efficacy of clinical prediction models, radiomics models, and combined models, identifying the optimal model group and machine learning algorithm. The SHAP method was employed for further explanatory analysis.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults:\u003c/b\u003e\u003c/p\u003e \u003cp\u003eVariables showing significant differences between groups included gender, smoking history, trauma history, history of lumbar surgery, residential area, history of glucocorticoid treatment, age, and VAS score. Through t-tests, intraclass correlation coefficients (ICCs), and LASSO regression analysis (Least Absolute Shrinkage and Selection Operator), eight radiomics features were identified to establish a Radscore. Multifactorial logistic regression analysis identified gender, smoking history, trauma history, lumbar surgery history, residential area, and Radscore as independent risk factors for OVCF. The combined model outperformed the other two. Due to overfitting in the Random Forest algorithm, KNN was determined to be the best machine learning algorithm. SHAP bar graphs displayed the influence factors in descending order of impact: residential area, Radscore, trauma history, gender, smoking, and lumbar surgery history. SHAP swarm plots revealed a broad distribution of Radscore, underscoring its significant predictive influence.\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusion:\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe diagnostic model developed through radiomics and machine learning algorithms reached an ideal level of effectiveness, with KNN in the combined model group demonstrating the highest diagnostic efficacy for assisting in the early X-ray diagnosis of OVCF.\u003c/p\u003e","manuscriptTitle":"Research on Assisting X-ray Diagnosis of Osteoporotic Vertebral Compression Fractures Using Interpretable Machine Learning Models and Radiomics Features","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-11 08:57:09","doi":"10.21203/rs.3.rs-6127302/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4394b64c-b02d-4c8a-9c10-bfea79caa495","owner":[],"postedDate":"March 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-03-18T04:38:12+00:00","versionOfRecord":[],"versionCreatedAt":"2025-03-11 08:57:09","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6127302","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6127302","identity":"rs-6127302","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}