Advancing Osteoporosis Opportunistic Screening: Multicenter Validation of a Deep Learning Algorithm Using Abdominal CT Scans

Advancing Osteoporosis Opportunistic Screening: Multicenter Validation of a Deep Learning Algorithm Using Abdominal CT Scans | 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 Advancing Osteoporosis Opportunistic Screening: Multicenter Validation of a Deep Learning Algorithm Using Abdominal CT Scans Augusto Sarquis Serpa, Marcelo Straus Takahashi, Eduardo Moreno Júdice de Mattos Farina, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7347450/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 31 Oct, 2025 Read the published version in Abdominal Radiology → Version 1 posted 12 You are reading this latest preprint version Abstract Purpose To develop and do multicenter validation on an algorithm that screens for osteoporosis from abdominal CTs. Methods This is a diagnostic accuracy study with retrospective data from January 2022 to July 2022 consisting of two steps: a segmentation step of the lumbar vertebral bodies, involving outpatient non-contrast abdominal CTs from [ANONYMIZED], and a multicenter validation step incorporating data from four additional institutions. The segmentation employed a 2D UNet with a ResNet34 backbone. We determined the Pearson correlation coefficient (r) between the mean of the slices’ mean attenuations (MSMA) on CT scans against the bone mineral density (BMD) on recent DEXA scans and calculated performance metrics for osteoporosis prediction, including 95% confidence intervals. Results The multicenter validation included 504 participants (median age, 66 years, interquartile range, 56–72; 388 women). A linear regression analysis showed an r of 0.63 (95% CI, 0.57–0.68) between MSMA (HU) and BMD (g/cm²). The AUCs (95% CI) for distinguishing between normal and osteoporosis were 0.96 (0.89, 1.0) for the internal dataset and 0.82 (0.75, 0.89) for the external dataset, and the performance metrics (95% CI), for a threshold of 202.6 HU, were 100% sensitivity (94, 100) and 91% specificity (84, 95) for internal data and 79% sensitivity (61, 90) and 81% specificity (76, 84) for external sites. Conclusion We developed and performed a multicenter validation of a DL model for osteoporosis prediction on CT. lumbar spine osteoporosis computed tomography deep learning opportunistic screening Figures Figure 1 Figure 2 Figure 3 Introduction Osteoporosis, a systemic skeletal disorder characterized by reduced bone mass and the microarchitectural deterioration of bone tissue [ 1 ], poses a significant public health challenge globally [ 2 , 3 ]. It increasingly leads to a higher risk of fractures, particularly in the hip, spine, and wrist, significantly impacting the quality of life in the aging population worldwide. There has been an increase, from 1990 to 2019, in the total number of deaths attributable to low bone mineral density (LBMD) and LBMD-related fractures [ 4 ]. The silent nature of the disease, often remaining undetected until a fracture occurs, underlines the importance of early detection and intervention [ 5 ]. Traditional methods for diagnosing osteoporosis predominantly rely on DEXA scans to measure bone mineral density (BMD) [ 6 ]. However, the limited availability of DEXA in certain regions, coupled with its cost, poses a barrier to widespread osteoporosis screening [ 7 ]. The development of concept of opportunistic screening stems from efforts to reduce unnecessary diagnostic imaging and the pursuit of innovative strategies to enhance the utility of imaging procedures [ 8 ]. Specifically, abdominal (and thoracic) CT scans, have emerged as a promising tool for opportunistic osteoporosis screening [ 9 – 14 ]. Initially centered on manual measurement techniques, it has progressively evolved to embrace automated methods, particularly deep learning (DL) approaches [ 15 – 25 ]; however, to our knowledge, current algorithms have only undergone internal validation. Accordingly, this paper focuses on the development and multicenter validation of an algorithm designed to predict osteoporosis from abdominal CT scans. Methods This is a diagnostic accuracy study with retrospective data, approved by each local Institutional Review Board (IRB) with a waiver for informed consent. It was carried out through a partnership among multiple institutions centralized by the [ANONYMIZED]. The methodology comprises two main components: a raw segmentation step and a multicenter validation step. Raw Segmentation Step The primary goal of this step was to delineate the vertebral bodies of the lumbar spine in non-contrast abdominal CT scans. Cortical and trabecular bone are included in the segmentations. We utilized a retrospective dataset of outpatient scans acquired between January and March 2022 from [ANONYMIZED], for various clinical indications. A convenience sample was used because there is no universally accepted method for sample size calculation in deep learning models. Inclusion criteria were patients aged 18 years and older. Scans showing movement artifacts were excluded. The data were divided into training, validation, and testing sets at a 70/20/10 ratio. The reference standard of the segmentations were annotated by MST, a senior radiologist with 10 years of experience. A 2D UNet architecture[ 26 ] with a ResNet34 backbone[ 27 ] was used. The model was initialized with weights pre-trained on the ImageNet database. During the training, a batch size of 32 and an initial learning rate of 10⁻³ were employed. Dice loss was used because it is well-suited when class distribution is imbalanced, as it aids in achieving a balance between precision and recall. A key augmentation technique implemented was a vertical flip, ensuring the model's effectiveness for imaging patients in different orientations, including prone positions. Model training incorporated early stopping mechanisms to prevent overfitting, with monitoring of the validation set for improvements in the Dice score. Training was halted if there was no improvement after 10 epochs. Multicenter Validation Step: Our multicenter validation step utilized another retrospective, convenience sample, including new data from [ANONYMIZED], distinct from that used in the initial segmentation step. This dataset was supplemented with data from four additional institutions: [ANONYMIZED], [ANONYMIZED], [ANONYMIZED], and [ANONYMIZED]. Anonymization was done by each institution before sending the data to the internal institution. The scans incorporated into the study ranged from October 2021 and July 2022 and spanned a wide range of clinical scenarios, from outpatient visits and inpatient stays to emergency room presentations. Patients who had undergone an abdominal CT scan, which could also include chest and/or pelvis imagery, were eligible. They should have been performed for any reason in patients aged over 18 years. An inclusion criterion was the presence of a non-contrast phase in the CT scan for algorithm inference. Because of their widespread clinical use and availability, DEXA scans, conducted for the respective patient within a maximum interval of six months before or after the CT scan, were chosen as the reference standard. DEXA scans that did not adequately depict the lumbar spine were excluded. Reasons for exclusion included the presence of artifacts, surgeries, or fractures. If the assessment was based on a single vertebral body they were also excluded, in line with the 2023 official positions from the International Society for Clinical Densitometry (ISCD) [ 28 ]. With the L1 to L5 vertebral bodies segmented, we calculate the mean of the slices’ mean attenuation (MSMA). Expressed in HU, this measurement includes both cortical and trabecular bone. It is measured by first averaging the lumbar spine segmented pixels for each slice and then averaging the result of all slices. The strength of the association between the MSMA from our segmentation and the BMD of the lumbar spine was assessed using linear regression and the Pearson correlation coefficient was calculated. The diagnostic capability of the algorithm was further evaluated by calculating the Area Under the Curve (AUC). Performance metrics were calculated for a specific threshold using the Youden J method [ 29 ], aiming for a good trade-off between sensitivity and specificity. The independent variable was the MSMA obtained from the segmentation, and the dependent variable was the classification of patients based on their spine BMD values. Confidence intervals for AUCs were obtained through bootstrapping techniques and, for the performance metrics, were calculated using the Wilson score interval method. Our diagnostic approach focused exclusively on the lumbar spine due to the absence of segmented CT data for other skeletal sites. The categorization relied on the t-score, with values below or equal to -2.5 being considered LBMD / osteoporosis. The reference values were consistent across all institutions. In our study, the t-score calculation was based on a standardized reference sample of young caucasian women [ 30 ], aligning with the latest official positions from the ISCD [ 28 ]. All statistical analyses were conducted using Python 3.11, utilizing the latest versions of the SciPy, NumPy, Scikit-learn, Pandas, and Matplotlib libraries. Results Detailed information regarding the selection process of participants, including the inclusion and exclusion criteria for both the segmentation and validation steps, is systematically illustrated in the STARD (Standards for Reporting of Diagnostic Accuracy Studies) diagram (Fig. 1 ). Raw Segmentation Step In the segmentation step, we included 105 participants after applying the exclusion criteria. Two exams were excluded because of motion artifacts. In assessing the raw segmentation performance of our model, the Dice coefficient achieved a score of 0.998 on the validation set. When applied to the unseen test set, the Dice score remained high at 0.978. It also demonstrated robust performance across various patient orientations, including oblique and prone positions (Fig. 2 ). Multicenter validation In the multicenter validation step, the number of participants considered for the final sample was 504. This sample consisted of 388 (80%) female patients, with a median age of 66 years (interquartile range, 56–72) (Table 1 ). For this step, 21 cases were excluded: six due to surgeries, six because of fractures, five due to external artifacts, and four because the DEXA scans depicted only a single vertebral body. Data encompassing diagnostic categories, CT scan vendors, slice thickness for all participants, and segregated details for in-house ([ANONYMIZED] and external datasets are also systematically listed in Table 1 . The CT scans originated from various vendors and were conducted with a range of slice thicknesses, indicating diversity in the imaging data used for this study. In our dataset, 88.5% of the scans were acquired using a voltage of 120 kV. Notably, for the external sites included in our study, the prevalence of this voltage setting was even higher, at 92.3%. As for the DEXA scan diagnosis, 93.8% of the patients did not have osteoporosis. There were 16 exams acquired in the prone position. A linear regression analysis demonstrated a correlation coefficient (r) of 0.63 (95% CI, 0.57–0.68) between the MSMA in HU from CT scans and the BMD from DEXA scans. When selecting only exams acquired with 120 kV, the coefficient did not change significantly. The equation of the best-fit line was: Density (mg/cm³) = 410.52 + (2.56 × MSMA (HU)) The AUC of the Receiver Operating Characteristic (ROC) curves were calculated for various thresholds to differentiate between normal and LBMD, with values exceeding 0.8 for both internal and external datasets (Fig. 3 ). The values were similar when excluding exams acquired with voltages other than 120 kV. Optimal thresholds were also obtained using the Youden J method. The performance metrics derived from the threshold of 202.6 HU, reflecting the algorithm's ability to distinguish between different BMD categories, are detailed in Table 2 . Although these are low positive predictive values (PPVs), it's important to note that below the threshold of 202.6 HU, 73.6% of individuals were identified as having at least osteopenia, defined as a T-score < -1, so they still might benefit from an intervention even if they don’t meet the criteria for osteoporosis. Table 1 Characteristics of the multicenter validation sample. Variables All data (n = 504) Internal data (n = 104) External data (n = 400) Sex, n (%) Feminine 388 (80.0) 91 (87.5) 297 (78.0) Masculine 97 (20.0) 13 (12.5) 84 (22.0) Age (years), median ((IQR) 66 (56–72) 62 (52.75–72.5) 66 (57–72) Institution [ANONYMIZED] (internal site) 104 (20.6) [ANONYMIZED] 182 (36.1) [ANONYMIZED] 38 (7.5) [ANONYMIZED] 90 (17.9) [ANONYMIZED] 90 (17.9) CT Vendor, n (%) General Electric Siemens Healthineers Philips Canon 277 (54.9) 123 (24.4) 94 (18.7) 10 (2.0) 68 (65.4) 25 (24.0) 1 (1.0) 10 (9.6) 209 (52.3) 98 (24.5) 93 (23.2) 0 (0.0) kV 120 446 (88.5) 77 (74.0) 369 (92.3) >120 29 (5.8) 15 (14.4) 14 (3.5) < 120 29 (5.7) 12 (11.6) 17 (4.2) Slice Thickness (mm), n (%) 3.5 68 (13.5) 1 (0.9) 67 (16.7) Patient orientation Supine 488 (96.8) 104 (100.0) 384 (96.0) Prone 16 (3.2) 0 (0.0) 16 (4.0) BMD Category, n (%) t-score >-2.5 t-score <= -2.5 473 (93.8) 31 (6.2) 101 (97.1) 3 (2.9) 372 (93.0) 28 (7.0) IQR, interquartile range. Table 2 Performance metrics [95% CI] for predicting osteoporosis, considering a threshold of 202.6 HU. All data (n = 504) Internal data (n = 104) External data (n = 400) Sensitivity Specificity 81 (25/31) [64–91] 83 (393/474) [79–86] 100 (3/3) [44–100] 91 (92/101) [84–95] 79 (22/28) [61–90] 81 (301/373) [76–84] Positive Predictive Value 24 (25/106) [ 17 – 33 ] 25 (3/12) [9–53] 23 (22/94) [ 16 – 32 ] Negative Predictive Value Accuracy 99 (393/399) [97–99] 83 (418/505) [79–86] 100 (92/92) [96–100] 91 (95/104) [84–95] 98 (301/307) [95–99] 81 (323/401) [76–84] Discussion Our study examines the potential of using abdominal CT scans, a commonly performed imaging test, as an opportunistic screening tool for identifying individuals with LBMD, using a DL technique, enhanced by a multicenter validation. To our knowledge, ours might be the first DL algorithm to be published in the literature that has done this step. The robustness of our algorithm across different patient orientations, including pronated positions, further reinforces its potential for clinical application in diverse settings. By testing our algorithm across a variety of clinical environments and patient populations, we have significantly enhanced the generalizability and credibility of our findings. Furthermore, our algorithm has shown to keep good performance in a dataset comprising multiple CT vendors and varying slice thickness. This suggests that it may work also in a variety of equipment settings. The segmentation performance, as indicated by the Dice coefficient, was consistent in both the validation and hold-out test sets. This high level of precision is crucial, as vertebrae attenuation in routine CT scans has been shown to correlate well with BMD values derived from both Quantitative Computed Tomography and DEXA scans[ 12 , 15 ]. This potential is bolstered by the Pearson correlation coefficient (r) value of 0.63 obtained between MSMA and BMD in our study, consistent with values reported in the literature comparing trabecular bone attenuation and DEXA-based BMD, which range from 0.399 to 0.891[ 31 ], although through studies that utilized internal validation only. A recent single center study by Pickhard et al. (2022)[ 22 ], featuring another DL method, using a seven-slice median approach, with an ellipsis placed on the anterior aspect of the trabecular bone of L1, had an AUC of 0.93 with up to 94% sensitivity and 84% specificity for osteoporosis prediction, similar results to ours’ internal validation performance. An advantage of using the whole lumbar spine is that fractures or other pathological processes that may affect L1 will interfere less with the results. Also, by including cortical bone in the segmentations, we take into account that fragility fractures are often a result of both cortical thinning and loss of trabecular bone[ 32 – 34 ]]. Despite the promising results, our approach has limitations. There was a noted drop in performance during the external testing phase, however, it is important to acknowledge that three out of the four external testing institutions, [ANONYMIZED], [ANONYMIZED] and [ANONYMIZED] are high-complexity healthcare facilities, which handle a lower proportion of normal or near-normal exams, a factor that can impact the performance of a diagnostic tool. We also utilized BMD values obtained from DEXA scans as our primary reference standard. However, it's vital to acknowledge that other reference standards, particularly those based on adverse outcomes like fractures or injurious falls, are also critical. We aim to explore these additional endpoints in our forthcoming research. Another limitation of our study is the employment of a methodology that is not as extensively validated in scientific literature compared to the ellipsis method, and is more challenging to benchmark against manual measurements. Notably, these manual measurements were not incorporated into our analysis. A less critical yet noteworthy limitation of our study is the validation of our algorithm exclusively on non-contrast enhanced scans. This focus may restrict its applicability in certain healthcare settings where scanning protocols exclusively include contrast phases, potentially limiting its broader adoption. Another important thing to note is that the positive predictive values were universally low and the negative predictive values high, partially due to the low prevalence of osteoporosis in our sample. In conclusion, our study advances the field of osteoporosis screening by demonstrating the feasibility and effectiveness of using DL algorithms with abdominal CT scans. The multicenter nature of our validation provides a foundation for future research and potential clinical application. As validation studies continue to evolve, it is anticipated that such opportunistic screening methods will become increasingly integrated into routine clinical practice, aiding in the early detection and management of osteoporosis. Declarations Competing Interests This is an unfunded research supported by the Bunkerhill Health Consortium and the algorithm is of private use of the company. Augusto Sarquis Serpa, Sean Bennett, Nishith Khandwala, Melissa Major, Jack Paparian, and Felipe Campos Kitamura are (or were) employees or partners of the Bunkerhill consortium. Author Contribution ASS cleaned external datasets, analyzed the results, wrote the main manuscript, created all the figures and is the corresponding author.MST gathered the internal dataset, trained the segmentation model and wrote part of the inference code.Bunkerhill Health promoted the cooperation of the institutions through its consortium. EMJDF, SB, NK, MM, JP and FK helped with the statistical analysis.EMJDF, FV, SR, LKB, RF, JM and FK helped providing data from their respective institutions. SR, LKB, RF, JM and FK made changes to the main manuscript.All authors reviewed the main manuscript. Data Availability If requested, a de-identified dataset containing a table of patient characteristics, algorithm results, and bone mineral density values will be made available. However, the original CT images and segmentation data will not be shared due to privacy and proprietary constraints. Access will be granted to researchers who provide a sound scientific proposal and agree to use the data only for the intended purpose outlined in their proposal, for an unlimited amount of time. References Föger-Samwald U, Dovjak P, Azizi-Semrad U, Kerschan-Schindl K, Pietschmann (2020) Osteoporosis: Pathophysiology and therapeutic options. EXCLI J 19:1017-1037. DOI:10.17179/excli2020-2591 Lane NE (2006) Epidemiology, Etiology, and Diagnosis of Osteoporosis. Am J Obstet Gynecol 194(2 Suppl):S3–11. DOI:10.1016/j.ajog.2005.08.047 Black DM, Rosen CJ (2016) Clinical Practice. Postmenopausal Osteoporosis. New Engl J Med 374(3):254–262. DOI:10.1056/NEJMcp1513724 Shen Y, Huang X, Wu J et al. (2022) The Global Burden of Osteoporosis, Low Bone Mass, and Its Related Fracture in 204 Countries and Territories, 1990-2019. Front Endocrinol 13:882241. DOI:10.3389/fendo.2022.882241 Munch S, Shapiro S (2006) The silent thief: osteoporosis and women's health care across the life span. Health Soc Work 31(1):44-53. DOI:10.1093/hsw/31.1.44 Sheu A, Diamond T (2016) Bone mineral density: testing for osteoporosis. Aust Prescr 39(2):35-39. DOI:10.18773/austprescr.2016.020 Hayes BL, Curtis JR, Laster A et al. (2010) Osteoporosis care in the United States after declines in reimbursements for DXA. J Clin Densitom 13(4):352-360. DOI:10.1016/j.jocd.2010.08.001 Oren O, Kebebew E, Ioannidis JPA (2019) Curbing Unnecessary and Wasted Diagnostic Imaging. JAMA 321(3):245–246. DOI:10.1001/jama.2018.20295 Pickhardt PJ, Graffy PM, Perez AA, Lubner MG, Elton DC, Summers RM (2021) Opportunistic screening at abdominal CT: use of automated body composition biomarkers for added cardiometabolic value. RadioGraphics 41(2):524–542. DOI:10.1148/rg.2021200056 Pickhardt PJ, Summers RM, Garrett JW (2021) Automated CT-based body composition analysis: a golden opportunity. Korean J Radiol 22(12):1934–1937. DOI:10.3348/kjr.2021.0775 Lee SJ, Graffy PM, Zea RD, Ziemlewicz TJ, Pickhardt PJ (2018) Future osteoporotic fracture risk related to lumbar vertebral trabecular attenuation measured at routine body CT. J Bone Miner Res 33(5):860–867. DOI:10.1002/jbmr.3383 Lee SJ, Pickhardt PJ (2017) Opportunistic screening for osteoporosis using body CT scans obtained for other indications: the UW experience. Clin Rev Bone Miner Metab 15(3):128–137. DOI:10.1007/s00198-015-3318-4 Graffy PM, Lee SJ, Ziemlewicz TJ, Pickhardt PJ (2017) Prevalence of vertebral compression fractures on routine CT scans according to L1 trabecular attenuation: determining relevant thresholds for opportunistic osteoporosis screening. AJR Am J Roentgenol 209(3):491–496. DOI:10.2214/AJR.17.17853 Gausden EB, Nwachukwu BU, Schreiber JJ, Lorich DG, Lane JM (2017) Opportunistic use of CT imaging for osteoporosis screening and bone density assessment: a qualitative systematic review. J Bone Joint Surg Am 99(18):1580–1590. DOI:10.2106/JBJS.16.00749 Fang Y, Li W, Chen X et al. (2021) Opportunistic osteoporosis screening in multi-detector CT images using deep convolutional neural networks. Eur Radiol 31(4):1831–1842. DOI:10.1007/s00330-020-07312-8 Cheng X, Zhao K, Zha X et al. (2021) Opportunistic screening using low-dose CT and the prevalence of osteoporosis in China: a nationwide, multicenter study. J Bone Miner Res 36(3):427–435. DOI:10.1002/jbmr.4187 Boutin RD, Lenchik L (2020) Value-added opportunistic CT: insights into osteoporosis and sarcopenia. AJR Am J Roentgenol 215(3):582–594. DOI:10.2214/AJR.20.22874 Pickhardt PJ, Lee SJ, Liu J et al. (2019) Population-based opportunistic osteoporosis screening: validation of a fully automated CT tool for assessing longitudinal BMD changes. Br J Radiol 92(1094):20180726. DOI:10.1259/bjr.20180726 Dagan N, Elnekave E, Barda N et al. (2020) Automated opportunistic osteoporotic fracture risk assessment using computed tomography scans to aid in FRAX underutilization. Nat Med 26(1):77–82. DOI:10.1038/s41591-019-0720-z Roux C, Rozes A, Reizine D et al. (2022) Fully automated opportunistic screening of vertebral fractures and osteoporosis on more than 150,000 routine computed tomography scans. Rheumatology (Oxford) 61(8):3269–3278. DOI:10.1093/rheumatology/keab878 Summers RM, Baecher N, Yao J et al. (2011) Feasibility of simultaneous computed tomographic colonography and fully automated bone mineral densitometry in a single examination. J Comput Assist Tomogr 35(2):212–216. DOI:10.1097/RCT.0b013e3182032537 Pickhardt PJ, Nguyen T, Perez AA et al. (2022) Improved CT-based Osteoporosis Assessment with a Fully Automated Deep Learning Tool. Radiol Artif Intell 4(5):e220042. DOI:10.1148/ryai.220042 Niu X, Huang Y, Li X et al. (2023) Development and validation of a fully automated system using deep learning for opportunistic osteoporosis screening using low-dose computed tomography scans. Quant Imaging Med Surg 13(8):5294-5305. DOI:10.21037/qims-22-1438 Schmidt D, Ulén J, Enqvist O et al. (2022) Deep learning takes the pain out of back-breaking work - Automatic vertebral segmentation and attenuation measurement for osteoporosis. Clin Imaging 81:54-59. DOI:10.1016/j.clinimag.2021.08.009 Oh S, Kang WY, Park H et al. (2024) Evaluation of deep learning-based quantitative computed tomography for opportunistic osteoporosis screening. Sci Rep 14(1):363. DOI:10.1038/s41598-023-45824-7 Ronneberger O, Fischer P, Brox T (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. arXiv. DOI:10.48550/arXiv.1505.04597 He K, Zhang X, Ren S, Sun J (2015) Deep Residual Learning for Image Recognition. arXiv. DOI:10.48550/arXiv.1512.03385 International Society for Clinical Densitometry Addult Official Positions, International Society for Clinical Densitometry website (2023) Available via https://iscd.org/official-positions-2023/. Accessed 21 Jan 2024 Youden WJ (1950) Index for rating diagnostic tests. Cancer 3:32–35. DOI:10.1002/1097-0142(1950)3:13.0.co;2-3 Xue S, Zhang Y, Qiao W et al. (2021) An Updated Reference for Calculating Bone Mineral Density T-Scores. J Clin Endocrinol Metab 106(7):e2613-e2621. DOI:10.1210/clinem/dgab180 Gausden EB, Nwachukwu BU, Schreiber JJ, Lorich DG, Lane JM (2017) Opportunistic Use of CT Imaging for Osteoporosis Screening and Bone Density Assessment. JBJS 99(18):1580–1590. DOI:10.2106/JBJS.16.00749 Seeman E (2002) Pathogenesis of bone fragility in women and men. Lancet 359(9320):1841–1850. DOI:10.1016/S0140-6736(02)08706-8 Cooper DM, Kawalilak CE, Harrison K, Johnston BD, Johnston JD (2016) Cortical Bone Porosity: What Is It, Why Is It Important, and How Can We Detect It? Curr Osteoporos Rep 14(5):187-198. DOI:10.1007/s11914-016-0319-y Ramchand SK, Seeman E (2018) The Influence of Cortical Porosity on the Strength of Bone During Growth and Advancing Age. Curr Osteoporos Rep 16(5):561-572. DOI:10.1007/s11914-018-0478-0 Additional Declarations Competing interest reported. This is an unfunded research supported by the Bunkerhill Health Consortium and the algorithm is of private use of the company. Augusto Sarquis Serpa, Sean Bennett, Nishith Khandwala, Melissa Major, Jack Paparian, and Felipe Campos Kitamura are (or were) employees or partners of the Bunkerhill consortium. Supplementary Files 261visualabstracttemplateupdated1.pptx Cite Share Download PDF Status: Published Journal Publication published 31 Oct, 2025 Read the published version in Abdominal Radiology → Version 1 posted Editorial decision: Revision requested 22 Aug, 2025 Reviews received at journal 21 Aug, 2025 Reviews received at journal 17 Aug, 2025 Reviews received at journal 15 Aug, 2025 Reviewers agreed at journal 14 Aug, 2025 Reviewers agreed at journal 14 Aug, 2025 Reviewers agreed at journal 13 Aug, 2025 Reviewers agreed at journal 13 Aug, 2025 Reviewers invited by journal 13 Aug, 2025 Editor assigned by journal 12 Aug, 2025 Submission checks completed at journal 12 Aug, 2025 First submitted to journal 11 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7347450","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":501396070,"identity":"536f8370-67da-4cc6-8613-78e574554528","order_by":0,"name":"Augusto Sarquis Serpa","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9UlEQVRIiWNgGAWjYJAD8w8fgCQbO1GKE8CkGeMMkBZmUrQw84AoQlr4Z6Rfk/z5w4aBn//wtsc2v7bJ8zEzMH74mINbi8SNnDJpnoQ0BskZaeXGuX23DduYGZglZ27DY82NnDRphoTDDAY3eAykc3tuMwK1sDHz4tEiD9Qi+QOoxf78GQNpy57b9gS1GNxIPybBA7KFIcdMmuHH7USCWgzPvGG25klL45G4kVZs2NtwO7mNmbEZr1/kjqc/vPnDxkaOv//wxgc//ty2nd/efPDDR3zeZ+AxAJNgNmMbmGzApx4I2B8gcf4QUDwKRsEoGAUjEgAAdydM0oNh1+AAAAAASUVORK5CYII=","orcid":"","institution":"Federal University of São Paulo","correspondingAuthor":true,"prefix":"","firstName":"Augusto","middleName":"Sarquis","lastName":"Serpa","suffix":""},{"id":501396071,"identity":"d43ceeb6-2f70-4bd9-b9a7-a0369576da96","order_by":1,"name":"Marcelo Straus Takahashi","email":"","orcid":"","institution":"Dasa","correspondingAuthor":false,"prefix":"","firstName":"Marcelo","middleName":"Straus","lastName":"Takahashi","suffix":""},{"id":501396072,"identity":"9b8cac7d-9b9e-4aaa-86bc-5a2c80aace73","order_by":2,"name":"Eduardo Moreno Júdice de Mattos Farina","email":"","orcid":"","institution":"Federal University of São Paulo","correspondingAuthor":false,"prefix":"","firstName":"Eduardo","middleName":"Moreno Júdice de Mattos","lastName":"Farina","suffix":""},{"id":501396073,"identity":"a0386eb1-2229-4135-9242-fbee7777c4a8","order_by":3,"name":"Sean Bennett","email":"","orcid":"","institution":"Bunkerhill Health","correspondingAuthor":false,"prefix":"","firstName":"Sean","middleName":"","lastName":"Bennett","suffix":""},{"id":501396075,"identity":"8e726cad-8402-4345-8742-2944074c1744","order_by":4,"name":"Nishith Khandwala","email":"","orcid":"","institution":"Bunkerhill Health","correspondingAuthor":false,"prefix":"","firstName":"Nishith","middleName":"","lastName":"Khandwala","suffix":""},{"id":501396077,"identity":"018a4e27-fb49-48ff-9cf0-526b7f3b63a7","order_by":5,"name":"Melissa Major","email":"","orcid":"","institution":"Bunkerhill Health","correspondingAuthor":false,"prefix":"","firstName":"Melissa","middleName":"","lastName":"Major","suffix":""},{"id":501396079,"identity":"ec6d08e8-a90f-4fcd-a78e-3bb4b58f9b64","order_by":6,"name":"Jack Paparian","email":"","orcid":"","institution":"Bunkerhill Health","correspondingAuthor":false,"prefix":"","firstName":"Jack","middleName":"","lastName":"Paparian","suffix":""},{"id":501396081,"identity":"46a4b34a-bcd5-4851-ad57-9bb4c0359d43","order_by":7,"name":"Fernanda Veloni","email":"","orcid":"","institution":"Dasa","correspondingAuthor":false,"prefix":"","firstName":"Fernanda","middleName":"","lastName":"Veloni","suffix":""},{"id":501396083,"identity":"51a3e462-3410-404d-bdc8-436c8c125c9d","order_by":8,"name":"Steven Rothenberg","email":"","orcid":"","institution":"University of Alabama at Birmingham","correspondingAuthor":false,"prefix":"","firstName":"Steven","middleName":"","lastName":"Rothenberg","suffix":""},{"id":501396084,"identity":"48193dda-92f6-4fcf-b060-5f8eadf7dceb","order_by":9,"name":"Leonardo Kayat Bittencourt","email":"","orcid":"","institution":"University Hospitals Cleveland Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Leonardo","middleName":"Kayat","lastName":"Bittencourt","suffix":""},{"id":501396085,"identity":"0984e53a-c5a2-46d3-9d37-d7e727429f0b","order_by":10,"name":"Ross Filice","email":"","orcid":"","institution":"MedStar Georgetown University Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ross","middleName":"","lastName":"Filice","suffix":""},{"id":501396086,"identity":"8701b607-42fd-404c-8286-684fcd1c93b8","order_by":11,"name":"John Mongan","email":"","orcid":"","institution":"University of California, San Francisco","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Mongan","suffix":""},{"id":501396087,"identity":"57a05f09-0604-4516-acc9-377bff79e0b7","order_by":12,"name":"Felipe Kitamura","email":"","orcid":"","institution":"Bunkerhill Health","correspondingAuthor":false,"prefix":"","firstName":"Felipe","middleName":"","lastName":"Kitamura","suffix":""}],"badges":[],"createdAt":"2025-08-11 14:38:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7347450/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7347450/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s00261-025-05213-2","type":"published","date":"2025-10-31T15:57:53+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":89562853,"identity":"f496841f-ad79-4c4f-9631-c6179513bd6f","added_by":"auto","created_at":"2025-08-21 10:26:37","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":217309,"visible":true,"origin":"","legend":"\u003cp\u003eSTARD (Standards for Reporting of Diagnostic Accuracy Studies) diagram illustrating each phase of the study.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7347450/v1/eed358ce2bd05505e59757b4.png"},{"id":89562855,"identity":"9449a8c9-679b-4fb6-8eb8-5907a2ac81fa","added_by":"auto","created_at":"2025-08-21 10:26:37","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":411499,"visible":true,"origin":"","legend":"\u003cp\u003eExamples of segmentations in pronated (Figures 2a and 2b) and oblique (Figures 2c and 2d) patients.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7347450/v1/6f31a8e8d29d86e92a2df4ca.png"},{"id":89565395,"identity":"32fa0eb0-79aa-40ef-9af5-b76daa70f813","added_by":"auto","created_at":"2025-08-21 10:42:37","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":154783,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve for each dataset for predicting osteoporosis. Red markings correspond to a threshold of 202.6 HU.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7347450/v1/74c904ddbf197b6a5634705f.png"},{"id":95040428,"identity":"bf1738b8-1744-4533-8d8b-7e258f6dcc47","added_by":"auto","created_at":"2025-11-03 16:08:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1494477,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7347450/v1/60225725-b133-4cd0-b46c-dea6e129ed74.pdf"},{"id":89562852,"identity":"7549abe8-b3c2-4f18-81e8-652848818326","added_by":"auto","created_at":"2025-08-21 10:26:37","extension":"pptx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":159679,"visible":true,"origin":"","legend":"","description":"","filename":"261visualabstracttemplateupdated1.pptx","url":"https://assets-eu.researchsquare.com/files/rs-7347450/v1/b291c02351de509ce9b785e0.pptx"}],"financialInterests":"Competing interest reported. This is an unfunded research supported by the Bunkerhill Health Consortium and the algorithm is of private use of the company. Augusto Sarquis Serpa, Sean Bennett, Nishith Khandwala, Melissa Major, Jack Paparian, and Felipe Campos Kitamura are (or were) employees or partners of the Bunkerhill consortium.","formattedTitle":"Advancing Osteoporosis Opportunistic Screening: Multicenter Validation of a Deep Learning Algorithm Using Abdominal CT Scans","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOsteoporosis, a systemic skeletal disorder characterized by reduced bone mass and the microarchitectural deterioration of bone tissue [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], poses a significant public health challenge globally [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. It increasingly leads to a higher risk of fractures, particularly in the hip, spine, and wrist, significantly impacting the quality of life in the aging population worldwide. There has been an increase, from 1990 to 2019, in the total number of deaths attributable to low bone mineral density (LBMD) and LBMD-related fractures [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. The silent nature of the disease, often remaining undetected until a fracture occurs, underlines the importance of early detection and intervention [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTraditional methods for diagnosing osteoporosis predominantly rely on DEXA scans to measure bone mineral density (BMD) [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. However, the limited availability of DEXA in certain regions, coupled with its cost, poses a barrier to widespread osteoporosis screening [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe development of concept of opportunistic screening stems from efforts to reduce unnecessary diagnostic imaging and the pursuit of innovative strategies to enhance the utility of imaging procedures [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Specifically, abdominal (and thoracic) CT scans, have emerged as a promising tool for opportunistic osteoporosis screening [\u003cspan additionalcitationids=\"CR10 CR11 CR12 CR13\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Initially centered on manual measurement techniques, it has progressively evolved to embrace automated methods, particularly deep learning (DL) approaches [\u003cspan additionalcitationids=\"CR16 CR17 CR18 CR19 CR20 CR21 CR22 CR23 CR24\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]; however, to our knowledge, current algorithms have only undergone internal validation.\u003c/p\u003e\u003cp\u003eAccordingly, this paper focuses on the development and multicenter validation of an algorithm designed to predict osteoporosis from abdominal CT scans.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e This is a diagnostic accuracy study with retrospective data, approved by each local Institutional Review Board (IRB) with a waiver for informed consent. It was carried out through a partnership among multiple institutions centralized by the [ANONYMIZED]. The methodology comprises two main components: a raw segmentation step and a multicenter validation step.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eRaw Segmentation Step\u003c/h2\u003e\u003cp\u003eThe primary goal of this step was to delineate the vertebral bodies of the lumbar spine in non-contrast abdominal CT scans. Cortical and trabecular bone are included in the segmentations.\u003c/p\u003e\u003cp\u003eWe utilized a retrospective dataset of outpatient scans acquired between January and March 2022 from [ANONYMIZED], for various clinical indications. A convenience sample was used because there is no universally accepted method for sample size calculation in deep learning models. Inclusion criteria were patients aged 18 years and older. Scans showing movement artifacts were excluded. The data were divided into training, validation, and testing sets at a 70/20/10 ratio. The reference standard of the segmentations were annotated by MST, a senior radiologist with 10 years of experience.\u003c/p\u003e\u003cp\u003eA 2D UNet architecture[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] with a ResNet34 backbone[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] was used. The model was initialized with weights pre-trained on the ImageNet database. During the training, a batch size of 32 and an initial learning rate of 10⁻\u0026sup3; were employed. Dice loss was used because it is well-suited when class distribution is imbalanced, as it aids in achieving a balance between precision and recall.\u003c/p\u003e\u003cp\u003eA key augmentation technique implemented was a vertical flip, ensuring the model's effectiveness for imaging patients in different orientations, including prone positions. Model training incorporated early stopping mechanisms to prevent overfitting, with monitoring of the validation set for improvements in the Dice score. Training was halted if there was no improvement after 10 epochs.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eMulticenter Validation Step:\u003c/h3\u003e\n\u003cp\u003eOur multicenter validation step utilized another retrospective, convenience sample, including new data from [ANONYMIZED], distinct from that used in the initial segmentation step. This dataset was supplemented with data from four additional institutions: [ANONYMIZED], [ANONYMIZED], [ANONYMIZED], and [ANONYMIZED]. Anonymization was done by each institution before sending the data to the internal institution. The scans incorporated into the study ranged from October 2021 and July 2022 and spanned a wide range of clinical scenarios, from outpatient visits and inpatient stays to emergency room presentations.\u003c/p\u003e\u003cp\u003ePatients who had undergone an abdominal CT scan, which could also include chest and/or pelvis imagery, were eligible. They should have been performed for any reason in patients aged over 18 years. An inclusion criterion was the presence of a non-contrast phase in the CT scan for algorithm inference. Because of their widespread clinical use and availability, DEXA scans, conducted for the respective patient within a maximum interval of six months before or after the CT scan, were chosen as the reference standard. DEXA scans that did not adequately depict the lumbar spine were excluded. Reasons for exclusion included the presence of artifacts, surgeries, or fractures. If the assessment was based on a single vertebral body they were also excluded, in line with the 2023 official positions from the International Society for Clinical Densitometry (ISCD) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eWith the L1 to L5 vertebral bodies segmented, we calculate the mean of the slices\u0026rsquo; mean attenuation (MSMA). Expressed in HU, this measurement includes both cortical and trabecular bone. It is measured by first averaging the lumbar spine segmented pixels for each slice and then averaging the result of all slices.\u003c/p\u003e\u003cp\u003eThe strength of the association between the MSMA from our segmentation and the BMD of the lumbar spine was assessed using linear regression and the Pearson correlation coefficient was calculated. The diagnostic capability of the algorithm was further evaluated by calculating the Area Under the Curve (AUC). Performance metrics were calculated for a specific threshold using the Youden J method [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], aiming for a good trade-off between sensitivity and specificity. The independent variable was the MSMA obtained from the segmentation, and the dependent variable was the classification of patients based on their spine BMD values. Confidence intervals for AUCs were obtained through bootstrapping techniques and, for the performance metrics, were calculated using the Wilson score interval method.\u003c/p\u003e\u003cp\u003eOur diagnostic approach focused exclusively on the lumbar spine due to the absence of segmented CT data for other skeletal sites. The categorization relied on the t-score, with values below or equal to -2.5 being considered LBMD / osteoporosis. The reference values were consistent across all institutions. In our study, the t-score calculation was based on a standardized reference sample of young caucasian women [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], aligning with the latest official positions from the ISCD [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAll statistical analyses were conducted using Python 3.11, utilizing the latest versions of the SciPy, NumPy, Scikit-learn, Pandas, and Matplotlib libraries.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eDetailed information regarding the selection process of participants, including the inclusion and exclusion criteria for both the segmentation and validation steps, is systematically illustrated in the STARD (Standards for Reporting of Diagnostic Accuracy Studies) diagram (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eRaw Segmentation Step\u003c/h3\u003e\n\u003cp\u003eIn the segmentation step, we included 105 participants after applying the exclusion criteria. Two exams were excluded because of motion artifacts.\u003c/p\u003e\u003cp\u003eIn assessing the raw segmentation performance of our model, the Dice coefficient achieved a score of 0.998 on the validation set. When applied to the unseen test set, the Dice score remained high at 0.978. It also demonstrated robust performance across various patient orientations, including oblique and prone positions (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eMulticenter validation\u003c/h3\u003e\n\u003cp\u003eIn the multicenter validation step, the number of participants considered for the final sample was 504. This sample consisted of 388 (80%) female patients, with a median age of 66 years (interquartile range, 56\u0026ndash;72) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). For this step, 21 cases were excluded: six due to surgeries, six because of fractures, five due to external artifacts, and four because the DEXA scans depicted only a single vertebral body.\u003c/p\u003e\u003cp\u003eData encompassing diagnostic categories, CT scan vendors, slice thickness for all participants, and segregated details for in-house ([ANONYMIZED] and external datasets are also systematically listed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The CT scans originated from various vendors and were conducted with a range of slice thicknesses, indicating diversity in the imaging data used for this study. In our dataset, 88.5% of the scans were acquired using a voltage of 120 kV. Notably, for the external sites included in our study, the prevalence of this voltage setting was even higher, at 92.3%. As for the DEXA scan diagnosis, 93.8% of the patients did not have osteoporosis. There were 16 exams acquired in the prone position.\u003c/p\u003e\u003cp\u003eA linear regression analysis demonstrated a correlation coefficient (r) of 0.63 (95% CI, 0.57\u0026ndash;0.68) between the MSMA in HU from CT scans and the BMD from DEXA scans. When selecting only exams acquired with 120 kV, the coefficient did not change significantly. The equation of the best-fit line was:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eDensity (mg/cm\u0026sup3;)\u0026thinsp;=\u0026thinsp;410.52 + (2.56 \u0026times; MSMA (HU))\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe AUC of the Receiver Operating Characteristic (ROC) curves were calculated for various thresholds to differentiate between normal and LBMD, with values exceeding 0.8 for both internal and external datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The values were similar when excluding exams acquired with voltages other than 120 kV. Optimal thresholds were also obtained using the Youden J method. The performance metrics derived from the threshold of 202.6 HU, reflecting the algorithm's ability to distinguish between different BMD categories, are detailed in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Although these are low positive predictive values (PPVs), it's important to note that below the threshold of 202.6 HU, 73.6% of individuals were identified as having at least osteopenia, defined as a T-score \u0026lt; -1, so they still might benefit from an intervention even if they don\u0026rsquo;t meet the criteria for osteoporosis.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eCharacteristics of the multicenter validation sample.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariables\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAll data (n\u0026thinsp;=\u0026thinsp;504)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eInternal data (n\u0026thinsp;=\u0026thinsp;104)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eExternal data (n\u0026thinsp;=\u0026thinsp;400)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSex, n (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFeminine\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e388 (80.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e91 (87.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e297 (78.0)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMasculine\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e97 (20.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e13 (12.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e84 (22.0)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAge (years), median ((IQR)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e66 (56\u0026ndash;72)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e62 (52.75\u0026ndash;72.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e66 (57\u0026ndash;72)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eInstitution\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e[ANONYMIZED] (internal site)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e104 (20.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e[ANONYMIZED]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e182 (36.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e[ANONYMIZED]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e38 (7.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e[ANONYMIZED]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e90 (17.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e[ANONYMIZED]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e90 (17.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eCT Vendor, n (%)\u003c/b\u003e\u003c/p\u003e\u003cp\u003eGeneral Electric\u003c/p\u003e\u003cp\u003eSiemens Healthineers\u003c/p\u003e\u003cp\u003ePhilips\u003c/p\u003e\u003cp\u003eCanon\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\u003cp\u003e277 (54.9)\u003c/p\u003e\u003cp\u003e123 (24.4)\u003c/p\u003e\u003cp\u003e94 (18.7)\u003c/p\u003e\u003cp\u003e10 (2.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\u003cp\u003e68 (65.4)\u003c/p\u003e\u003cp\u003e25 (24.0)\u003c/p\u003e\u003cp\u003e1 (1.0)\u003c/p\u003e\u003cp\u003e10 (9.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\u003cp\u003e209 (52.3)\u003c/p\u003e\u003cp\u003e98 (24.5)\u003c/p\u003e\u003cp\u003e93 (23.2)\u003c/p\u003e\u003cp\u003e0 (0.0)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ekV\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e120\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e446 (88.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e77 (74.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e369 (92.3)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u0026gt;120\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e29 (5.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e15 (14.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e14 (3.5)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u0026lt; 120\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e29 (5.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e12 (11.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e17 (4.2)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eSlice Thickness (mm), n (%)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u0026lt;= 1.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e153 (30.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66 (63.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e87 (21.8)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1.5\u0026ndash;2.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e141 (28.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e37 (35.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e104 (26.0)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e2.5\u0026ndash;3.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e142 (28.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0 (0.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e142 (35.5)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u0026gt;3.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e68 (13.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1 (0.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e67 (16.7)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePatient orientation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSupine\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e488 (96.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e104 (100.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e384 (96.0)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eProne\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e16 (3.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0 (0.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e16 (4.0)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBMD Category, n (%)\u003c/b\u003e\u003c/p\u003e\u003cp\u003et-score \u0026gt;-2.5\u003c/p\u003e\u003cp\u003et-score \u0026lt;= -2.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\u003cp\u003e473 (93.8)\u003c/p\u003e\u003cp\u003e31 (6.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\u003cp\u003e101 (97.1)\u003c/p\u003e\u003cp\u003e3 (2.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\u003cp\u003e372 (93.0)\u003c/p\u003e\u003cp\u003e28 (7.0)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eIQR, interquartile range.\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance metrics [95% CI] for predicting osteoporosis, considering a threshold of 202.6 HU.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAll data (n\u0026thinsp;=\u0026thinsp;504)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eInternal data (n\u0026thinsp;=\u0026thinsp;104)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eExternal data (n\u0026thinsp;=\u0026thinsp;400)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e81 (25/31) [64\u0026ndash;91]\u003c/p\u003e\u003cp\u003e83 (393/474) [79\u0026ndash;86]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e100 (3/3) [44\u0026ndash;100]\u003c/p\u003e\u003cp\u003e91 (92/101) [84\u0026ndash;95]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e79 (22/28) [61\u0026ndash;90]\u003c/p\u003e\u003cp\u003e81 (301/373) [76\u0026ndash;84]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePositive Predictive Value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e24 (25/106) [\u003cspan additionalcitationids=\"CR18 CR19 CR20 CR21 CR22 CR23 CR24 CR25 CR26 CR27 CR28 CR29 CR30 CR31 CR32\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e25 (3/12) [9\u0026ndash;53]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e23 (22/94) [\u003cspan additionalcitationids=\"CR17 CR18 CR19 CR20 CR21 CR22 CR23 CR24 CR25 CR26 CR27 CR28 CR29 CR30 CR31\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNegative Predictive Value\u003c/p\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e99 (393/399) [97\u0026ndash;99]\u003c/p\u003e\u003cp\u003e83 (418/505) [79\u0026ndash;86]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e100 (92/92) [96\u0026ndash;100]\u003c/p\u003e\u003cp\u003e91 (95/104) [84\u0026ndash;95]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e98 (301/307) [95\u0026ndash;99]\u003c/p\u003e\u003cp\u003e81 (323/401) [76\u0026ndash;84]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur study examines the potential of using abdominal CT scans, a commonly performed imaging test, as an opportunistic screening tool for identifying individuals with LBMD, using a DL technique, enhanced by a multicenter validation. To our knowledge, ours might be the first DL algorithm to be published in the literature that has done this step. The robustness of our algorithm across different patient orientations, including pronated positions, further reinforces its potential for clinical application in diverse settings. By testing our algorithm across a variety of clinical environments and patient populations, we have significantly enhanced the generalizability and credibility of our findings. Furthermore, our algorithm has shown to keep good performance in a dataset comprising multiple CT vendors and varying slice thickness. This suggests that it may work also in a variety of equipment settings.\u003c/p\u003e\u003cp\u003eThe segmentation performance, as indicated by the Dice coefficient, was consistent in both the validation and hold-out test sets. This high level of precision is crucial, as vertebrae attenuation in routine CT scans has been shown to correlate well with BMD values derived from both Quantitative Computed Tomography and DEXA scans[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. This potential is bolstered by the Pearson correlation coefficient (r) value of 0.63 obtained between MSMA and BMD in our study, consistent with values reported in the literature comparing trabecular bone attenuation and DEXA-based BMD, which range from 0.399 to 0.891[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], although through studies that utilized internal validation only. A recent single center study by Pickhard et al. (2022)[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], featuring another DL method, using a seven-slice median approach, with an ellipsis placed on the anterior aspect of the trabecular bone of L1, had an AUC of 0.93 with up to 94% sensitivity and 84% specificity for osteoporosis prediction, similar results to ours\u0026rsquo; internal validation performance. An advantage of using the whole lumbar spine is that fractures or other pathological processes that may affect L1 will interfere less with the results. Also, by including cortical bone in the segmentations, we take into account that fragility fractures are often a result of both cortical thinning and loss of trabecular bone[\u003cspan additionalcitationids=\"CR33\" citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]].\u003c/p\u003e\u003cp\u003eDespite the promising results, our approach has limitations. There was a noted drop in performance during the external testing phase, however, it is important to acknowledge that three out of the four external testing institutions, [ANONYMIZED], [ANONYMIZED] and [ANONYMIZED] are high-complexity healthcare facilities, which handle a lower proportion of normal or near-normal exams, a factor that can impact the performance of a diagnostic tool. We also utilized BMD values obtained from DEXA scans as our primary reference standard. However, it's vital to acknowledge that other reference standards, particularly those based on adverse outcomes like fractures or injurious falls, are also critical. We aim to explore these additional endpoints in our forthcoming research. Another limitation of our study is the employment of a methodology that is not as extensively validated in scientific literature compared to the ellipsis method, and is more challenging to benchmark against manual measurements. Notably, these manual measurements were not incorporated into our analysis. A less critical yet noteworthy limitation of our study is the validation of our algorithm exclusively on non-contrast enhanced scans. This focus may restrict its applicability in certain healthcare settings where scanning protocols exclusively include contrast phases, potentially limiting its broader adoption. Another important thing to note is that the positive predictive values were universally low and the negative predictive values high, partially due to the low prevalence of osteoporosis in our sample.\u003c/p\u003e\u003cp\u003eIn conclusion, our study advances the field of osteoporosis screening by demonstrating the feasibility and effectiveness of using DL algorithms with abdominal CT scans. The multicenter nature of our validation provides a foundation for future research and potential clinical application. As validation studies continue to evolve, it is anticipated that such opportunistic screening methods will become increasingly integrated into routine clinical practice, aiding in the early detection and management of osteoporosis.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003cp\u003eThis is an unfunded research supported by the Bunkerhill Health Consortium and the algorithm is of private use of the company. Augusto Sarquis Serpa, Sean Bennett, Nishith Khandwala, Melissa Major, Jack Paparian, and Felipe Campos Kitamura are (or were) employees or partners of the Bunkerhill consortium.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eASS cleaned external datasets, analyzed the results, wrote the main manuscript, created all the figures and is the corresponding author.MST gathered the internal dataset, trained the segmentation model and wrote part of the inference code.Bunkerhill Health promoted the cooperation of the institutions through its consortium. EMJDF, SB, NK, MM, JP and FK helped with the statistical analysis.EMJDF, FV, SR, LKB, RF, JM and FK helped providing data from their respective institutions. SR, LKB, RF, JM and FK made changes to the main manuscript.All authors reviewed the main manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eIf requested, a de-identified dataset containing a table of patient characteristics, algorithm results, and bone mineral density values will be made available. However, the original CT images and segmentation data will not be shared due to privacy and proprietary constraints. Access will be granted to researchers who provide a sound scientific proposal and agree to use the data only for the intended purpose outlined in their proposal, for an unlimited amount of time.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eF\u0026ouml;ger-Samwald U, Dovjak P, Azizi-Semrad U, Kerschan-Schindl K, Pietschmann (2020) Osteoporosis: Pathophysiology and therapeutic options. EXCLI J 19:1017-1037. DOI:10.17179/excli2020-2591\u003c/li\u003e\n\u003cli\u003eLane NE (2006) Epidemiology, Etiology, and Diagnosis of Osteoporosis. Am J Obstet Gynecol 194(2 Suppl):S3\u0026ndash;11. DOI:10.1016/j.ajog.2005.08.047\u003c/li\u003e\n\u003cli\u003eBlack DM, Rosen CJ (2016) Clinical Practice. Postmenopausal Osteoporosis. New Engl J Med 374(3):254\u0026ndash;262. DOI:10.1056/NEJMcp1513724\u003c/li\u003e\n\u003cli\u003eShen Y, Huang X, Wu J et al. (2022) The Global Burden of Osteoporosis, Low Bone Mass, and Its Related Fracture in 204 Countries and Territories, 1990-2019. Front Endocrinol 13:882241. DOI:10.3389/fendo.2022.882241\u003c/li\u003e\n\u003cli\u003eMunch S, Shapiro S (2006) The silent thief: osteoporosis and women\u0026apos;s health care across the life span. Health Soc Work 31(1):44-53. DOI:10.1093/hsw/31.1.44\u003c/li\u003e\n\u003cli\u003eSheu A, Diamond T (2016) Bone mineral density: testing for osteoporosis. Aust Prescr 39(2):35-39. DOI:10.18773/austprescr.2016.020\u003c/li\u003e\n\u003cli\u003eHayes BL, Curtis JR, Laster A et al. (2010) Osteoporosis care in the United States after declines in reimbursements for DXA. J Clin Densitom 13(4):352-360. DOI:10.1016/j.jocd.2010.08.001\u003c/li\u003e\n\u003cli\u003eOren O, Kebebew E, Ioannidis JPA (2019) Curbing Unnecessary and Wasted Diagnostic Imaging. JAMA 321(3):245\u0026ndash;246. DOI:10.1001/jama.2018.20295\u003c/li\u003e\n\u003cli\u003ePickhardt PJ, Graffy PM, Perez AA, Lubner MG, Elton DC, Summers RM (2021) Opportunistic screening at abdominal CT: use of automated body composition biomarkers for added cardiometabolic value. RadioGraphics 41(2):524\u0026ndash;542. DOI:10.1148/rg.2021200056\u003c/li\u003e\n\u003cli\u003ePickhardt PJ, Summers RM, Garrett JW (2021) Automated CT-based body composition analysis: a golden opportunity. Korean J Radiol 22(12):1934\u0026ndash;1937. DOI:10.3348/kjr.2021.0775\u003c/li\u003e\n\u003cli\u003eLee SJ, Graffy PM, Zea RD, Ziemlewicz TJ, Pickhardt PJ (2018) Future osteoporotic fracture risk related to lumbar vertebral trabecular attenuation measured at routine body CT. J Bone Miner Res 33(5):860\u0026ndash;867. DOI:10.1002/jbmr.3383\u003c/li\u003e\n\u003cli\u003eLee SJ, Pickhardt PJ (2017) Opportunistic screening for osteoporosis using body CT scans obtained for other indications: the UW experience. Clin Rev Bone Miner Metab 15(3):128\u0026ndash;137. DOI:10.1007/s00198-015-3318-4\u003c/li\u003e\n\u003cli\u003eGraffy PM, Lee SJ, Ziemlewicz TJ, Pickhardt PJ (2017) Prevalence of vertebral compression fractures on routine CT scans according to L1 trabecular attenuation: determining relevant thresholds for opportunistic osteoporosis screening. AJR Am J Roentgenol 209(3):491\u0026ndash;496. DOI:10.2214/AJR.17.17853\u003c/li\u003e\n\u003cli\u003eGausden EB, Nwachukwu BU, Schreiber JJ, Lorich DG, Lane JM (2017) Opportunistic use of CT imaging for osteoporosis screening and bone density assessment: a qualitative systematic review. J Bone Joint Surg Am 99(18):1580\u0026ndash;1590. DOI:10.2106/JBJS.16.00749\u003c/li\u003e\n\u003cli\u003eFang Y, Li W, Chen X et al. (2021) Opportunistic osteoporosis screening in multi-detector CT images using deep convolutional neural networks. Eur Radiol 31(4):1831\u0026ndash;1842. DOI:10.1007/s00330-020-07312-8\u003c/li\u003e\n\u003cli\u003eCheng X, Zhao K, Zha X et al. (2021) Opportunistic screening using low-dose CT and the prevalence of osteoporosis in China: a nationwide, multicenter study. J Bone Miner Res 36(3):427\u0026ndash;435. DOI:10.1002/jbmr.4187\u003c/li\u003e\n\u003cli\u003eBoutin RD, Lenchik L (2020) Value-added opportunistic CT: insights into osteoporosis and sarcopenia. AJR Am J Roentgenol 215(3):582\u0026ndash;594. DOI:10.2214/AJR.20.22874\u003c/li\u003e\n\u003cli\u003ePickhardt PJ, Lee SJ, Liu J et al. (2019) Population-based opportunistic osteoporosis screening: validation of a fully automated CT tool for assessing longitudinal BMD changes. Br J Radiol 92(1094):20180726. DOI:10.1259/bjr.20180726\u003c/li\u003e\n\u003cli\u003eDagan N, Elnekave E, Barda N et al. (2020) Automated opportunistic osteoporotic fracture risk assessment using computed tomography scans to aid in FRAX underutilization. Nat Med 26(1):77\u0026ndash;82. DOI:10.1038/s41591-019-0720-z\u003c/li\u003e\n\u003cli\u003eRoux C, Rozes A, Reizine D et al. (2022) Fully automated opportunistic screening of vertebral fractures and osteoporosis on more than 150,000 routine computed tomography scans. Rheumatology (Oxford) 61(8):3269\u0026ndash;3278. DOI:10.1093/rheumatology/keab878\u003c/li\u003e\n\u003cli\u003eSummers RM, Baecher N, Yao J et al. (2011) Feasibility of simultaneous computed tomographic colonography and fully automated bone mineral densitometry in a single examination. J Comput Assist Tomogr 35(2):212\u0026ndash;216. DOI:10.1097/RCT.0b013e3182032537\u003c/li\u003e\n\u003cli\u003ePickhardt PJ, Nguyen T, Perez AA et al. (2022) Improved CT-based Osteoporosis Assessment with a Fully Automated Deep Learning Tool. Radiol Artif Intell 4(5):e220042. DOI:10.1148/ryai.220042\u003c/li\u003e\n\u003cli\u003eNiu X, Huang Y, Li X et al. (2023) Development and validation of a fully automated system using deep learning for opportunistic osteoporosis screening using low-dose computed tomography scans. Quant Imaging Med Surg 13(8):5294-5305. DOI:10.21037/qims-22-1438\u003c/li\u003e\n\u003cli\u003eSchmidt D, Ul\u0026eacute;n J, Enqvist O et al. (2022) Deep learning takes the pain out of back-breaking work - Automatic vertebral segmentation and attenuation measurement for osteoporosis. Clin Imaging 81:54-59. DOI:10.1016/j.clinimag.2021.08.009\u003c/li\u003e\n\u003cli\u003eOh S, Kang WY, Park H et al. (2024) Evaluation of deep learning-based quantitative computed tomography for opportunistic osteoporosis screening. Sci Rep 14(1):363. DOI:10.1038/s41598-023-45824-7\u003c/li\u003e\n\u003cli\u003eRonneberger O, Fischer P, Brox T (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. arXiv. DOI:10.48550/arXiv.1505.04597\u003c/li\u003e\n\u003cli\u003eHe K, Zhang X, Ren S, Sun J (2015) Deep Residual Learning for Image Recognition. arXiv. DOI:10.48550/arXiv.1512.03385\u003c/li\u003e\n\u003cli\u003eInternational Society for Clinical Densitometry Addult Official Positions, International Society for Clinical Densitometry website (2023) Available via https://iscd.org/official-positions-2023/. Accessed 21 Jan 2024\u003c/li\u003e\n\u003cli\u003eYouden WJ (1950) Index for rating diagnostic tests. Cancer 3:32\u0026ndash;35. DOI:10.1002/1097-0142(1950)3:1\u0026lt;32::aid-cncr2820030106\u0026gt;3.0.co;2-3\u003c/li\u003e\n\u003cli\u003eXue S, Zhang Y, Qiao W et al. (2021) An Updated Reference for Calculating Bone Mineral Density T-Scores. J Clin Endocrinol Metab 106(7):e2613-e2621. DOI:10.1210/clinem/dgab180\u003c/li\u003e\n\u003cli\u003eGausden EB, Nwachukwu BU, Schreiber JJ, Lorich DG, Lane JM (2017) Opportunistic Use of CT Imaging for Osteoporosis Screening and Bone Density Assessment. JBJS 99(18):1580\u0026ndash;1590. DOI:10.2106/JBJS.16.00749\u003c/li\u003e\n\u003cli\u003eSeeman E (2002) Pathogenesis of bone fragility in women and men. Lancet 359(9320):1841\u0026ndash;1850. DOI:10.1016/S0140-6736(02)08706-8\u003c/li\u003e\n\u003cli\u003eCooper DM, Kawalilak CE, Harrison K, Johnston BD, Johnston JD (2016) Cortical Bone Porosity: What Is It, Why Is It Important, and How Can We Detect It? Curr Osteoporos Rep 14(5):187-198. DOI:10.1007/s11914-016-0319-y\u003c/li\u003e\n\u003cli\u003eRamchand SK, Seeman E (2018) The Influence of Cortical Porosity on the Strength of Bone During Growth and Advancing Age. Curr Osteoporos Rep 16(5):561-572. DOI:10.1007/s11914-018-0478-0\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"abdominal-radiology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"aima","sideBox":"Learn more about [Abdominal Radiology](http://link.springer.com/journal/261)","snPcode":"261","submissionUrl":"https://submission.springernature.com/new-submission/261/3","title":"Abdominal Radiology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"lumbar spine, osteoporosis, computed tomography, deep learning, opportunistic screening","lastPublishedDoi":"10.21203/rs.3.rs-7347450/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7347450/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e\u003cp\u003eTo develop and do multicenter validation on an algorithm that screens for osteoporosis from abdominal CTs.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eThis is a diagnostic accuracy study with retrospective data from January 2022 to July 2022 consisting of two steps: a segmentation step of the lumbar vertebral bodies, involving outpatient non-contrast abdominal CTs from [ANONYMIZED], and a multicenter validation step incorporating data from four additional institutions. The segmentation employed a 2D UNet with a ResNet34 backbone. We determined the Pearson correlation coefficient (r) between the mean of the slices\u0026rsquo; mean attenuations (MSMA) on CT scans against the bone mineral density (BMD) on recent DEXA scans and calculated performance metrics for osteoporosis prediction, including 95% confidence intervals.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003e The multicenter validation included 504 participants (median age, 66 years, interquartile range, 56\u0026ndash;72; 388 women). A linear regression analysis showed an r of 0.63 (95% CI, 0.57\u0026ndash;0.68) between MSMA (HU) and BMD (g/cm\u0026sup2;). The AUCs (95% CI) for distinguishing between normal and osteoporosis were 0.96 (0.89, 1.0) for the internal dataset and 0.82 (0.75, 0.89) for the external dataset, and the performance metrics (95% CI), for a threshold of 202.6 HU, were 100% sensitivity (94, 100) and 91% specificity (84, 95) for internal data and 79% sensitivity (61, 90) and 81% specificity (76, 84) for external sites.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eWe developed and performed a multicenter validation of a DL model for osteoporosis prediction on CT.\u003c/p\u003e","manuscriptTitle":"Advancing Osteoporosis Opportunistic Screening: Multicenter Validation of a Deep Learning Algorithm Using Abdominal CT Scans","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-21 10:26:32","doi":"10.21203/rs.3.rs-7347450/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-08-23T00:30:23+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-21T09:13:37+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-17T04:02:11+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-16T03:02:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"238069855842331457072286766248694868297","date":"2025-08-14T08:13:47+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"247817726968175088339176054081713928362","date":"2025-08-14T04:53:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"52552632325999842538709371467153730308","date":"2025-08-13T12:49:34+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"102285992457259616399855503388347566539","date":"2025-08-13T12:33:58+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-08-13T12:16:46+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-12T12:16:02+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-12T06:06:47+00:00","index":"","fulltext":""},{"type":"submitted","content":"Abdominal Radiology","date":"2025-08-11T14:31:00+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"abdominal-radiology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"aima","sideBox":"Learn more about [Abdominal Radiology](http://link.springer.com/journal/261)","snPcode":"261","submissionUrl":"https://submission.springernature.com/new-submission/261/3","title":"Abdominal Radiology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"82cb9126-1874-4230-9e22-b76b297b9b4e","owner":[],"postedDate":"August 21st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-03T16:04:33+00:00","versionOfRecord":{"articleIdentity":"rs-7347450","link":"https://doi.org/10.1007/s00261-025-05213-2","journal":{"identity":"abdominal-radiology","isVorOnly":false,"title":"Abdominal Radiology"},"publishedOn":"2025-10-31 15:57:53","publishedOnDateReadable":"October 31st, 2025"},"versionCreatedAt":"2025-08-21 10:26:32","video":"","vorDoi":"10.1007/s00261-025-05213-2","vorDoiUrl":"https://doi.org/10.1007/s00261-025-05213-2","workflowStages":[]},"version":"v1","identity":"rs-7347450","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7347450","identity":"rs-7347450","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}