Artificial Intelligence vs Human opportunistic Detection of Vertebral Fractures in Routine CT Scans: results of a pilot study

Artificial Intelligence vs Human opportunistic Detection of Vertebral Fractures in Routine CT Scans: results of a pilot study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Artificial Intelligence vs Human opportunistic Detection of Vertebral Fractures in Routine CT Scans: results of a pilot study Polyzois Makras, Eleftherios Chatzellis, Christos Gravvanis, Georgia Kanti, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8461514/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 23 Apr, 2026 Read the published version in Hormones → Version 1 posted 5 You are reading this latest preprint version Abstract Purpose Vertebral fractures (VFs) are the most common osteoporotic fractures and a hallmark of bone fragility, yet most morphometric VFs remain undiagnosed in routine imaging not focused on the spine. This pilot study evaluated the diagnostic performance of an artificial intelligence (AI) system for opportunistic vertebral fracture detection in chest and abdominal computer tomography (CT) scans compared with human readers in a real-world hospital setting. Methods Over two months, all thoracic and abdominal CTs performed for any indication in patients aged ≥ 50 years at a tertiary hospital were analyzed by the Bone Solution HealthOST AI platform (Nanox AI Ltd.). Routine radiology reports and targeted reviews by endocrinologists were compared to AI outputs, using a gold-standard adjudication by an expert panel. Sensitivity, specificity, predictive values, and odds ratios (ORs) were calculated and compared by χ² tests (p < 0.05). Results Among 427 eligible CT scans, AI detected vertebral fractures with significantly higher sensitivity (86.3%) than radiologists (50.0%) or endocrinologists (68.8%) (p ≤ 0.01 for both comparisons). Compared with radiologists, AI had 14.2-fold greater odds of correctly flagging true fractures (OR = 14.23; 95% CI 4.7–43.1; p < 0.000001) while against targeted review by endocrinologists, AI achieved a 4.8-fold advantage (OR = 4.77; 95% CI 1.6–13.6; p = 0.001). Extrapolated annually, AI integration could uncover around 150 new, otherwise undiagnosed, patients eligible for osteoporosis treatment. Conclusion Automated vertebral fracture detection in routine CT scans significantly enhances diagnostic yield versus human interpretation. Its integration into clinical workflows offers a cost-free, high-impact strategy to improve early osteoporosis diagnosis and treatment within the healthcare system. Artificial intelligence Vertebral fractures Computed tomography Opportunistic screening Osteoporosis Diagnostic accuracy Endocrinology Preventive medicine Figures Figure 1 Figure 2 Figure 3 Introduction Osteoporosis is a silent, progressive disease affecting an estimated 200 million people worldwide [ 1 ]. It remains underdiagnosed and undertreated, particularly when fragility fractures are not clinically apparent [ 2 , 3 ]. Vertebral fractures (VFs) are the most common osteoporotic fractures and represent unequivocal evidence of bone fragility. Despite their prognostic importance, they are often incidental findings and are frequently missed in routine imaging performed for non-musculoskeletal indications. Studies have demonstrated that as many as 60–70% of VFs visible on chest or abdominal imaging are not reported by radiologists [ 4 – 6 ]. This diagnostic oversight has profound implications. Each missed VF represents a missed opportunity for intervention, increasing the risk of subsequent hip or VFs and associated mortality [ 7 ]. Recent international and Greek Osteoporosis Guidelines emphasize that even a single low-energy morphometric VF constitutes a definitive diagnosis of osteoporosis, mandating immediate initiation of pharmacological treatment regardless of bone mineral density [ 8 – 10 ]. Identifying such fractures opportunistically could thus shift the paradigm from reactive to preventive osteoporosis care in Greece, a country with a grossly estimated osteoporosis treatment gap of 60% [ 11 ]. The growing volume of imaging studies and the limited availability of expert musculoskeletal radiologists further exacerbate underdiagnosis. Workload fatigue and the absence of targeted clinical queries often lead to incomplete interpretation of vertebral morphology, especially in emergency and general hospital environments. Recent years have witnessed an exponential rise in the number of radiological examinations per capita, amplifying the need for automated tools that can assist human readers and standardize reporting accuracy [ 12 – 15 ]. Artificial intelligence (AI) has emerged as a promising solution to these challenges. AI-based image analysis systems utilize deep learning algorithms capable of recognizing subtle, complex patterns beyond human perception. Initial investigations have demonstrated comparable or superior diagnostic accuracy to that of experienced radiologists in fracture detection, particularly when AI acts as an assistive “second reader” [ 16 – 18 ]. AI offers the additional advantage of continuous, fatigue-free performance and the potential for deployment in any radiology workflow without significant infrastructure costs. Beyond accuracy, AI introduces a paradigm of opportunistic screening, leveraging existing imaging data acquired for other indications to detect conditions such as osteoporosis [ 19 – 21 ], sarcopenia [ 22 – 24 ], or cardiovascular calcification [ 23 , 25 ]. This approach allows clinicians to extract additional diagnostic information from routine studies, enhancing value without increasing radiation exposure or cost. Such a model of preventive and predictive medicine resonates with the vision that artificial intelligence and wellbeing medicine are defining elements of the 21st-century endocrinology and healthcare delivery [ 26 ]. This is the first pilot study in Greece that evaluates the diagnostic performance of an AI system for opportunistic vertebral fracture detection in routine computer tomography (CT) scans, comparing it directly with both general radiologists and endocrinologists. The study also explores the implications of AI integration for early osteoporosis diagnosis and treatment initiation within the Greek national healthcare framework. Methods Study Design and Setting This was a prospective, single-center pilot study conducted at the 251 Hellenic Air Force General Hospital in Athens, Greece, between April 14 and June 16, 2025, by the Department of Medical Research. The institution provides tertiary-level services to both military personnel and civilians and includes an established bone metabolism clinic specializing in osteoporosis. The study assessed the performance of the Bone Solution HealthOST (Nanox AI Ltd., Tel Aviv, Israel), a CE-marked artificial intelligence platform developed for automated detection of vertebral height loss - compression fractures in chest and abdominal CT scans. The software was provided free of charge for the duration of the pilot through a bilateral institutional agreement. The AI model was trained on large-scale pseudoanonymized datasets encompassing over 100,000 CT slices, enabling recognition of vertebral deformities based on morphometric and textural features. Study Objectives The primary objective of the study was to compare the ability of the AI system to “flag” VFs on routine CT examinations performed for other than osteoporosis reasons with the corresponding routine radiology reports produced by general radiologists in their everyday clinical practice. Secondary objectives were to: Compare AI-based fracture flagging with targeted vertebral fracture assessment performed by endocrinologists evaluating and treating patients with osteoporosis in a designated outpatient clinic. Compare the overall ability of AI, radiologists, and endocrinologists to correctly discriminate CT examinations with or without fractures. Evaluate the clinical translation of AI findings by determining how many patients with AI-detected, previously unreported vertebral fractures would qualify—based on the Greek Osteoporosis Guidelines—for immediate osteoporosis treatment irrespective of BMD testing. Study Population and Data Processing All thoracic and abdominal CT scans of patients aged ≥ 50 years performed during the two-month period were retrieved from the hospital picture archiving communication system (PACS). Both contrast and non-contrast scans were eligible. When repeated CT scans were performed during a patient's hospital admission only the first scan was evaluated (i.e., up to one abdominal and/or chest CT per patient). CT scans obtained in the same patient and anatomical site more than once during the study period were considered as separate examinations only if the time interval between them exceeded 30 days. Exclusion criteria included suboptimal image quality, missing radiology reports, or incomplete spine visualization. All DICOM files were anonymized and uploaded to the cloud-based AI platform. Following automated analysis, AI-generated reports were returned and integrated into the local database. Each AI report included a classification of “fracture” or “no fracture” along with localization on sagittal reformats and the density of the vertebrae in Hounsfield units (HU) (Fig. 1); included fractures were either Grade 2 or Grade 3 according to the semi-quantitative visual grading of vertebral fractures [ 27 ]. Human Readers and Reference Standard Three diagnostic modalities were compared: Routine Radiology Reports – generated by general radiologists unaware of the actual study period. Endocrinologist Review – conducted by endocrinologists evaluating and treating osteoporosis patients who systematically examined each scan solely for vertebral deformities. AI System Output – the automated classification produced by the Bone Solution HealthOST platform. The verification of the fracture diagnosis which served as the “gold standard” was established through consensus between a senior radiologist (NK) and an endocrinologist (PM) specialized in bone diseases, both blinded to the AI and human readings, during combined assessment to produce the final ground-truth dataset. The study did not include an absolute quality control of the diagnostic classifications. In particular, it was not assessed whether, for example, the AI and/or the radiologists or endocrinologists correctly graded a fracture as Grade 2 or Grade 3, or whether they accurately identified all existing fractures in each CT scan — such as for example all three present fractures rather than only one or two. Therefore, the study’s design allowed assessment of AI as a platform for opportunistic osteoporosis case-finding in routine clinical practice rather than strictly as a diagnostic tool. Statistical Analysis To estimate the relative probabilities of correct “flagging”, odds ratios (ORs) were computed for the comparisons (AI vs. Radiologists, AI vs. Endocrinologists). ORs were derived from contingency tables of true positives and false positives, applying Woolf’s method (log-transformation of the OR) for the calculation of 95% confidence intervals and two-sided Wald tests for p-values (H₀: log(OR) = 0). Statistical analyses were also conducted to compare the diagnostic performance of the Artificial Intelligence (AI) system with that of human specialists —radiologists and endocrinologists— in detecting or excluding vertebral fractures on CT scans. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated using 2×2 contingency tables derived from the reference standard (gold standard). Although 95% confidence intervals were not directly applied to the proportions, group comparisons were assessed using the chi-square (χ²) test of independence. For each diagnostic metric, χ² tests were performed to compare proportions between the AI system and each corresponding reader group (AI vs. Radiologists, AI vs. Endocrinologists), and p-values are reported. Where appropriate, comparisons were visualized using bar plots. All analyses were performed in Python (SciPy, NumPy, Matplotlib), with statistical significance set at p < 0.05. Ethical Considerations The study was approved by the Institutional Review Board of the 251 Hellenic Air Force General Hospital (IRB No: F.076/6828/S.3128/4-6-24). All patient data were anonymized before analysis, and the IRB waived the requirement of written informed consent. The study was conducted in accordance with the Declaration of Helsinki (1975, revised 2013). No external funding was received for this project. Results Study Population In total, 699 CTs were initially screened; 427 scans met the inclusion criteria and were analyzed corresponding to 331 patients (136 females, mean age: 72.9 ± 13.5 years). Of the 427 included CT scans, 52% were thoracic and 48% abdominal. Indications for imaging included oncologic investigation and/or follow-up, known or suspected pulmonary disease, chronic or acute abdominal pathology. None of the CTs were performed specifically for evaluation of clinically suspected fragility fractures. AI vs. Human Readers The analysis of the study’s primary endpoint was essentially designed to evaluate the ability to detect (flag) the presence of one or more VF in each CT examination that was ultimately confirmed by the gold standard evaluation (case finding). Accordingly, this analysis focused only on the assessment of “positive” findings related to the existence of a fracture and did not include the negative diagnoses. A total of 80 examinations were flagged as having ≥ 1 vertebral fracture either by the AI system and/or by radiologists in their routine clinical reports. In the evaluation of flagging performance that ultimately corresponded to CT examinations harboring true fractures as assessed by the expert panel (case finding), the AI system demonstrated statistically significant superiority over radiologists (OR = 14.23; 95% CI: 4.7–43.1; p < 0.000001) (Fig. 2a). The same comparative approach was applied to the secondary endpoint assessing fracture-flagging performance between the AI system and endocrinologists. In this analysis, the AI system was again statistically superior to endocrinologists (OR = 4.77; 95% CI: 1.6–13.6; p = 0.001) (Fig. 2b). With respect to the secondary objective — comparing the ability to correctly discriminate CT examinations with or without fractures (AI vs. radiologists and AI vs. endocrinologists) — these analyses aimed to assess the sum of both “positive” and “negative” findings regarding the presence of fracture(s). In this per-examination analysis (detection of existing fractures in the whole cohort), the odds that the AI system would identify the true vertebral fractures were 6.3-fold higher than those of radiologists (OR = 6.30; 95% CI: 2.90–13.58; p < 0.000001). The AI system also demonstrated higher sensitivity compared with radiologists (86.3% vs. 50%, p < 0.000001) and a higher negative predictive value (96.9% vs. 89.6%, p = 0.0002) (Fig. 3a). In the analogous per-examination comparison between the AI system and endocrinologists, the odds of the AI system identifying existing vertebral fractures were 2.85-fold higher (OR = 2.85; 95% CI: 1.29–6.30; p = 0.009). The AI system again showed higher sensitivity (86.3% vs. 68.8%, p = 0.01) and a higher negative predictive value (96.9% vs. 93.3%, p = 0.04), although it exhibited slightly lower specificity (98.3% vs. 100%, p = 0.04) (Fig. 3b). Missed Fractures and Clinical Translation AI uniquely detected 33 CT scans with true fractures missed by the routine radiology reports which corresponded to 30 individual patients. Among these patients, 23 had never received osteoporosis therapy, 4 patients were currently under treatment, and 3 had discontinued treatment more than three years earlier. According to the Greek guidelines [ 10 ], these 26 individuals required immediate pharmacological therapy independent of BMD testing. Extrapolating this 2-months detection rate across one year suggested that routine AI deployment in this single hospital setting could identify approximately 156 new treatable osteoporosis cases annually—patients who would otherwise remain undiagnosed and at elevated risk of future fractures. Anatomical Distribution and Error Analysis The majority of fractures were located between Th7 and L4, while the vertebrae with the highest fracture frequency were Th12 and L1 consistent with mechanical stress points. AI missed few true fractures in cases with severe scoliosis while occasionally overcalled mild degenerative deformities in older patients, explaining the small number of false positives (Fig. 2). However, the majority of discordant readings between AI and radiologists represented genuine missed fractures rather than overdiagnosis. Discussion This pilot study demonstrates that a commercially available AI system can markedly outperform both general radiologists and endocrinologists in the opportunistic detection of vertebral fractures. The magnitude of improvement in case finding —over 14-fold higher odds compared with standard reporting— highlights a significant and actionable diagnostic gap in current clinical practice. Every missed vertebral fracture represents a missed opportunity for secondary prevention. In most osteoporosis guidelines, the identification of a low-energy vertebral fracture establishes a diagnosis of osteoporosis requiring treatment initiation [ 8 – 10 ]. Early recognition of non-clinical (morphometric) vertebral fractures allows clinicians to intervene before subsequent fractures occur, improving patient outcomes and reducing healthcare costs. This principle aligns with broader international recommendations advocating for opportunistic case finding through routine imaging [ 28 , 29 ]. Recent reports and meta-analyses demonstrated that AI tools consistently achieve pooled sensitivities around 90% across anatomical regions and yield maximal benefit when combined with human oversight [ 4 , 16 – 18 ]. Additionally, explainable foundation models capable of opportunistic osteoporosis screening from chest X-rays have been recently developed, achieving high area under the curve (AUC) values without reliance on bone density data [ 30 ]. Our findings reinforce these results in a real-world setting, further supporting the reliability of CE-marked AI systems when integrated into clinical workflows. In practice, the integration of AI into PACS could enable automatic flagging of suspected vertebral fractures, prompting radiologists or clinicians to review the flagged images. Such an approach would require minimal human input and no additional imaging. Moreover, automated alerts could be linked to hospital electronic medical records, facilitating direct referral to osteoporosis clinics for therapy initiation transforming the CT imaging into an active screening tool rather than a merely targeted diagnostic test. Future iterations of systems like the one we used in our pilot study could adopt similar visual interpretability modules, displaying heatmaps over detected vertebral deformities to further aid user verification. From a population perspective, opportunistic vertebral fracture detection could dramatically reduce the “osteoporosis care gap” [ 31 ]. In Greece, where adherence to screening and treatment remains suboptimal [ 11 ], automatic identification of high-risk individuals during routine CT imaging offers a sustainable, cost-neutral intervention. Even modest improvements in detection rates could translate into hundreds of prevented fractures annually, reducing morbidity, mortality, and healthcare expenditure. In this regard and specifically for the 251 Hellenic Air Force General Hospital in which the first Greek Fracture Liaison Service (FLS) was implemented, a total of 213 incident fragility fractures were identified during the 1st year period, comprising approximately 26 vertebral, 51 hip, and 136 non-vertebral fractures [ 32 ]. In this context, the projected annual detection of approximately 156 previously unrecognized, treatment-eligible patients through routine AI-assisted CT analysis represents a clinically meaningful expansion of case-finding capacity. Within an established FLS framework, such an increase could nearly double the number of vertebral and other major osteoporotic fractures captured and appropriately referred for secondary fracture prevention, thereby significantly amplifying the clinical impact and effectiveness of the FLS model. Beyond diagnostic efficiency, the implications of AI extend to the evolving concept of precision and wellbeing medicine. Artificial intelligence appears to be not only a technological novelty but a fundamental enabler of predictive endocrinology, where data-driven insights enhance disease prevention and health maintenance [ 26 ]. In this context, the integration of AI into bone health management might reflect a transition toward holistic, anticipatory care. The study has inherent limitations. It was single center, with a relatively short observation period and a moderate sample size while the reference standard was based on expert consensus. Furthermore, the AI model’s training dataset was international, and its performance might vary with different scanners or population demographics. Future research should prioritize multicenter validation of AI-assisted vertebral fracture detection across diverse patient populations to ensure generalizability and robustness. Equally important is the integration of AI outputs with established clinical tools, such as FRAX scores and BMD measurements, to develop hybrid risk-stratification models with superior predictive performance. Rigorous health-economic analyses are also needed to quantify the potential cost savings arising from earlier diagnosis and subsequent fracture prevention. In parallel, the development of explainable AI frameworks will be essential to enhance algorithmic transparency and strengthen clinician trust. Finally, evidence generated from these efforts should inform national-level policy frameworks that support the widespread implementation of AI-assisted opportunistic screening within routine clinical practice. In conclusion, in this first Greek pilot study, AI achieved significantly higher sensitivity and diagnostic yield for vertebral fracture detection than both radiologists and endocrinologists. Given the central role of radiologists in the diagnostic procedure, AI integration into radiology workflows could convert routine CT scans into powerful, cost-free screening instruments for osteoporosis thereby facilitating early detection and treatment. Declarations Funding declaration : No funding was received for this study. Compliance with ethical standards: Conflict of Interest : The authors have no competing interests to declare that are relevant to the content of this article. Ethical Approval: The study was approved by the Institutional Review Board of the 251 Hellenic Air Force General Hospital (IRB No: F.076/6828/S.3128/4-6-24). Informed Consent: All patient data were anonymized before analysis, and the IRB waived the requirement of written informed consent. Author Contribution declaration Conceptualization: Polyzois Makras; Methodology: Polyzois Makras, Nikolaos Kyriakopoulos; Formal analysis and investigation: Polyzois Makras, Eleftherios Chatzellis, Christos Gravvanis, Georgia Kanti, Nikolaos Kyriakopoulos; Data analysis and statistics: Polyzois Makras, Maria P. Yavropoulou; Writing - original draft preparation: Polyzois Makras; Writing - review and editing: Polyzois Makras, Eleftherios Chatzellis, Christos Gravvanis, Georgia Kanti, Maria P. Yavropoulou, Konstantinos Papadimitropoulos, Nikolaos Kyriakopoulos; Supervision: Polyzois Makras, Konstantinos Papadimitropoulos CT reviews in routine clinical practice: Radiology Department reporting group References Clynes MA, Harvey NC, Curtis EM, Fuggle NR, Dennison EM, Cooper C (2020) The epidemiology of osteoporosis. Br Med Bull 133(1):105–117 Curtis EM, Dennison EM, Cooper C, Harvey NC (2022) Osteoporosis in 2022: Care gaps to screening and personalised medicine. Best Pract Res Clin Rheumatol 36(3):101754 Kanis JA, Norton N, Harvey NC, Jacobson T, Johansson H, Lorentzon M, McCloskey EV, Willers C, Borgstrom F (2021) SCOPE 2021: a new scorecard for osteoporosis in Europe. Arch Osteoporos 16(1):82 Husarek J, Hess S, Razaeian S, Ruder TD, Sehmisch S, Muller M, Liodakis E (2024) Artificial intelligence in commercial fracture detection products: a systematic review and meta-analysis of diagnostic test accuracy. Sci Rep 14(1):23053 Muller M, Schechter CB, Hautz WE, Sauter TC, Exadaktylos AK, Stock S, Birrenbach T (2021) The development and validation of a resource consumption score of an emergency department consultation. PLoS ONE 16(2):e0247244 Bruls RJM, Kwee RM (2020) Workload for radiologists during on-call hours: dramatic increase in the past 15 years. Insights Imaging 11(1):121 Toth E, Banefelt J, Akesson K, Spangeus A, Ortsater G, Libanati C (2020) History of Previous Fracture and Imminent Fracture Risk in Swedish Women Aged 55 to 90 Years Presenting With a Fragility Fracture. J Bone Min Res 35(5):861–868 Kanis JA, Cooper C, Rizzoli R, Reginster JY, Scientific Advisory Board of the European Society for Clinical and Economic Aspects of Osteoporosis, Committees of Scientific Advisors and National Societies of the International Osteoporosis Foundation (2019) European guidance for the diagnosis and management of osteoporosis in postmenopausal women. Osteoporos Int 30(1):3–44 Shoback D, Rosen CJ, Black DM, Cheung AM, Murad MH, Eastell R (2020) Pharmacological Management of Osteoporosis in Postmenopausal Women: An Endocrine Society Guideline Update. J Clin Endocrinol Metab 105(3) Makras P, Anastasilakis AD, Antypas G, Chronopoulos E, Kaskani EG, Matsouka A, Patrikos DK, Stathopoulos KD, Tournis S, Trovas G, Kosmidis C (2019) The 2018 Guidelines for the diagnosis and treatment of osteoporosis in Greece. Arch Osteoporos 14(1):39 Yavropoulou MP, Makras P, Athanasakis K, Bournia VK, Mathioudakis K, Tsolakidis A, Kassi E, Kaltsas G, Mitrou P, Sfikakis PP (2022) Prevalence and patterns of anti-osteoporotic drug use based on 2019 real-world nationwide data in Greece. Arch Osteoporos 17(1):86 Di Somma S, Paladino L, Vaughan L, Lalle I, Magrini L, Magnanti M (2015) Overcrowding in emergency department: an international issue. Intern Emerg Med 10(2):171–175 Lantsman CD, Barash Y, Klang E, Guranda L, Konen E, Tau N (2022) Trend in radiologist workload compared to number of admissions in the emergency department. Eur J Radiol 149:110195 Smith E, Dasan S (2018) A system under pressure. Br J Hosp Med (Lond) 79(9):495–499 Mdodo R, Frazier EL, Dube SR, Mattson CL, Sutton MY, Brooks JT, Skarbinski J (2015) Cigarette smoking prevalence among adults with HIV compared with the general adult population in the United States: cross-sectional surveys. Ann Intern Med 162(5):335–344 Canoni-Meynet L, Verdot P, Danner A, Calame P, Aubry S (2022) Added value of an artificial intelligence solution for fracture detection in the radiologist's daily trauma emergencies workflow. Diagn Interv Imaging 103(12):594–600 Yang S, Yin B, Cao W, Feng C, Fan G, He S (2020) Diagnostic accuracy of deep learning in orthopaedic fractures: a systematic review and meta-analysis. Clin Radiol 75(9):713e17–713e28 Kuo RYL, Harrison C, Curran TA, Jones B, Freethy A, Cussons D, Stewart M, Collins GS, Furniss D (2022) Artificial Intelligence in Fracture Detection: A Systematic Review and Meta-Analysis. Radiology 304(1):50–62 Yang J, Liao M, Wang Y, Chen L, He L, Ji Y, Xiao Y, Lu Y, Fan W, Nie Z, Wang R, Qi B, Yang F (2022) Opportunistic osteoporosis screening using chest CT with artificial intelligence. Osteoporos Int 33(12):2547–2561 Wu Y, Yang X, Wang M, Lian Y, Hou P, Chai X, Dai Q, Qian B, Jiang Y, Gao J (2025) Artificial intelligence assisted automatic screening of opportunistic osteoporosis in computed tomography images from different scanners. Eur Radiol 35(4):2287–2295 Behanova M, Sokhan A, Haschka J, Zandieh S, Salzlechner C, Ljuhar R, Zwerina J, Kocijan R (2026) AI-supported opportunistic detection of vertebral fractures on routine CT scans: Diagnostic performance and clinical relevance. Bone 203:117735 Urooj B, Ko Y, Na S, Kim IO, Lee EH, Cho S, Jeong H, Khang S, Lee J, Kim KW (2025) Implementation of Fully Automated AI-Integrated System for Body Composition Assessment on Computed Tomography for Opportunistic Sarcopenia Screening: Multicenter Prospective Study. JMIR Form Res 9:e69940 Pickhardt PJ, Correale L, Hassan C (2023) AI-based opportunistic CT screening of incidental cardiovascular disease, osteoporosis, and sarcopenia: cost-effectiveness analysis. Abdom Radiol (NY) 48(3):1181–1198 Ryu J, Eom S, Kim HC, Kim CO, Rhee Y, You SC, Hong N (2023) Chest X-ray-based opportunistic screening of sarcopenia using deep learning. J Cachexia Sarcopenia Muscle 14(1):418–428 Jeong J, Chao CJ, Arsanjani R, Ayoub C, Lester SJ, Pereyra M, Said EF, Roarke M, Tagle-Cornell C, Koepke LM, Tsai YL, Jung-Hsuan C, Chang CC, Farina JM, Trivedi H, Patel BN, Banerjee I (2025) Artificial Intelligence Chest X-Ray Opportunistic Screening Model for Coronary Artery Calcium Deposition: A Multi-Objective Model With Multimodal Data Fusion. Mayo Clin Proc Digit Health 3(4):100300 Stratakis CA (2025) Artificial intelligence and quality of life are two leading themes of contemporary medicine. Horm (Athens) 24(3):589–591 Genant HK, Wu CY, van Kuijk C, Nevitt MC (1993) Vertebral fracture assessment using a semiquantitative technique. J Bone Min Res 8(9):1137–1148 Aggarwal V, Maslen C, Abel RL, Bhattacharya P, Bromiley PA, Clark EM, Compston JE, Crabtree N, Gregory JS, Kariki EP, Harvey NC, Ward KA, Poole KES (2021) Opportunistic diagnosis of osteoporosis, fragile bone strength and vertebral fractures from routine CT scans; a review of approved technology systems and pathways to implementation. Ther Adv Musculoskelet Dis 13:1759720X211024029. 10.1177/1759720X211024029 Reginster JY, Schmidmaier R, Alokail M, Hiligsmann M (2025) Cost-effectiveness of opportunistic osteoporosis screening using chest radiographs with deep learning in Germany. Aging Clin Exp Res 37(1):149 Kim J, Kwak S, Lee H, Chang J, Park SM (2025) Explainable opportunistic osteoporosis screening from chest X-rays: a retrospective comparison of foundation models. Osteoporos Int. 10.1007/s00198-025-07727-3 Willers C, Norton N, Harvey NC, Jacobson T, Johansson H, Lorentzon M, McCloskey EV, Borgstrom F, Kanis JA Scientific Report Panel of the IOF (2022) Osteoporosis in Europe: a compendium of country-specific reports. Arch Osteoporos 17(1):23. 10.1007/s11657-021-00969-8 Makras P, Panagoulia M, Mari A, Rizou S, Lyritis GP (2017) Evaluation of the first fracture liaison service in the Greek healthcare setting. Arch Osteoporos 12(1):3. 10.1007/s11657-016-0299-7 Cite Share Download PDF Status: Published Journal Publication published 23 Apr, 2026 Read the published version in Hormones → Version 1 posted Editorial decision: Major revisions 03 Mar, 2026 Reviewers agreed at journal 08 Jan, 2026 Reviewers invited by journal 08 Jan, 2026 Editor assigned by journal 30 Dec, 2025 First submitted to journal 29 Dec, 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-8461514","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":571478321,"identity":"3156ed12-ca75-4a4c-b113-f40bc7473042","order_by":0,"name":"Polyzois Makras","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvUlEQVRIiWNgGAWjYHACZhAhByIOPCBFizFYSwIpWhIbQCRRWszFDj82rqipS58fdvgh0BY7Od0GAlosZ6cZJ545djh34+00A6CWZGOzAwS0GNxOMD7YwHYgd+PsBJCWA4nbCGtJ/3yw4V9duuHs9A/EaskxTmxsY06Ql84h0hbL2TnFho19hw03SOcUHEgwIMIv5tLpmyUbvtXJy89O3/zhQ4WdHGHvwxkHULjEaJFvIEL1KBgFo2AUjEwAAE7NRyRB2H5SAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-3231-3357","institution":"251o Genikó Nosokomeío Aeroporías: 251 Geniko Nosokomeio Aeroporias","correspondingAuthor":true,"prefix":"","firstName":"Polyzois","middleName":"","lastName":"Makras","suffix":""},{"id":571478324,"identity":"40de8cb9-069d-44e4-bd3e-4c5ed55f9981","order_by":1,"name":"Eleftherios Chatzellis","email":"","orcid":"","institution":"251o Genikó Nosokomeío Aeroporías: 251 Geniko Nosokomeio Aeroporias","correspondingAuthor":false,"prefix":"","firstName":"Eleftherios","middleName":"","lastName":"Chatzellis","suffix":""},{"id":571478326,"identity":"9c240797-3ad5-4e42-b379-dba8d5b68bf7","order_by":2,"name":"Christos Gravvanis","email":"","orcid":"","institution":"251o Genikó Nosokomeío Aeroporías: 251 Geniko Nosokomeio Aeroporias","correspondingAuthor":false,"prefix":"","firstName":"Christos","middleName":"","lastName":"Gravvanis","suffix":""},{"id":571478328,"identity":"9a803765-6c26-4d36-84df-80bcc9c6f8d8","order_by":3,"name":"Georgia Kanti","email":"","orcid":"","institution":"251o Genikó Nosokomeío Aeroporías: 251 Geniko Nosokomeio Aeroporias","correspondingAuthor":false,"prefix":"","firstName":"Georgia","middleName":"","lastName":"Kanti","suffix":""},{"id":571478331,"identity":"15a4be2c-9688-4c78-86b1-7758df83374e","order_by":4,"name":"Maria P. Yavropoulou","email":"","orcid":"","institution":"Laiko Geniko Nosokomeio: Laiko Geniko Nosokomeio Athenon","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"P.","lastName":"Yavropoulou","suffix":""},{"id":571478333,"identity":"442fb5fc-c5b0-4c36-a58c-16a35de0e204","order_by":5,"name":"Konstantinos Papadimitropoulos","email":"","orcid":"","institution":"251o Genikó Nosokomeío Aeroporías: 251 Geniko Nosokomeio Aeroporias","correspondingAuthor":false,"prefix":"","firstName":"Konstantinos","middleName":"","lastName":"Papadimitropoulos","suffix":""},{"id":571478336,"identity":"e31d731c-136b-421e-ae4f-584c37dc299e","order_by":6,"name":"Nikolaos Kyriakopoulos","email":"","orcid":"","institution":"Laiko Geniko Nosokomeio: Laiko Geniko Nosokomeio Athenon","correspondingAuthor":false,"prefix":"","firstName":"Nikolaos","middleName":"","lastName":"Kyriakopoulos","suffix":""}],"badges":[],"createdAt":"2025-12-27 14:07:01","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8461514/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8461514/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s42000-026-00784-1","type":"published","date":"2026-04-23T15:59:07+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":100126617,"identity":"742b4a3e-532c-47aa-ba53-57523e16fc0e","added_by":"auto","created_at":"2026-01-13 09:25:22","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10361,"visible":true,"origin":"","legend":"","description":"","filename":"hormHORMD2500631.xml","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/888764dc0da74c4f136eba06.xml"},{"id":100126600,"identity":"6e18d07a-8088-4ec2-9a23-03227608b22f","added_by":"auto","created_at":"2026-01-13 09:25:18","extension":"xml","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1041,"visible":true,"origin":"","legend":"","description":"","filename":"HORMD25006318601.go.xml","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/d539690a718f02450de477db.xml"},{"id":100126605,"identity":"ef41eff3-5f7a-45d1-a8fd-6666a0bc184c","added_by":"auto","created_at":"2026-01-13 09:25:20","extension":"xml","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":963,"visible":true,"origin":"","legend":"","description":"","filename":"HORMD2500631Import.xml","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/24c87d7a1bc7018550a2fc1a.xml"},{"id":100126588,"identity":"4f94e206-391c-4599-915f-036059416a5a","added_by":"auto","created_at":"2026-01-13 09:25:15","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":92054,"visible":true,"origin":"","legend":"","description":"","filename":"HORMD25006310enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/a7306e1aa805ebd2f5e133c9.xml"},{"id":100126594,"identity":"2152cccc-676e-4e2b-8262-888fb6c839df","added_by":"auto","created_at":"2026-01-13 09:25:17","extension":"pptx","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":459482,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1.pptx","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/ebb6e906806555fc73321a1d.pptx"},{"id":100126611,"identity":"8370c023-2d51-4a4c-aac9-3025cd6416b2","added_by":"auto","created_at":"2026-01-13 09:25:21","extension":"pptx","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1057675,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2.pptx","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/c94085382abee55b77aae59e.pptx"},{"id":100126623,"identity":"2587334a-693f-4a51-ba7f-5debe6f90478","added_by":"auto","created_at":"2026-01-13 09:25:22","extension":"pptx","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1353079,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3.pptx","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/53de0153fd48f0b1b391f5e2.pptx"},{"id":100126624,"identity":"ba945af5-b43f-4749-8553-f94ef75493de","added_by":"auto","created_at":"2026-01-13 09:25:22","extension":"xml","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":88371,"visible":true,"origin":"","legend":"","description":"","filename":"HORMD25006310structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/a6fde0869d67e6e093a77071.xml"},{"id":100366008,"identity":"f58fe8e1-5a0e-45b0-b2fa-34a3241926f9","added_by":"auto","created_at":"2026-01-16 07:55:51","extension":"html","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":102526,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/11e9d8cb2bfd6c133b1afa54.html"},{"id":100126630,"identity":"923ef8df-ce7d-497f-b18c-7c9a060cf6a2","added_by":"auto","created_at":"2026-01-13 09:25:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":343911,"visible":true,"origin":"","legend":"\u003cp\u003eSnapshots from AI output. Panel A: Identification of a Th12 moderate fracture. Panel B: identification of a Th12 severe fracture in a patient also hosting a Th11 fracture that was also evident in the specific AI output (but not in this snapshot)\u003c/p\u003e","description":"","filename":"Fig11.png","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/5cea5925ffa3abbb7edc74fb.png"},{"id":100126626,"identity":"3395a177-1e5b-448c-ba27-d585579fbf26","added_by":"auto","created_at":"2026-01-13 09:25:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":64822,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation of flagging performance: comparison AI Vs Radiologists (Panel A) and comparison AI Vs Endocrinologists (Panel B)\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNote: The scans flagged for fracture(s) from radiologists were not identical with those flagged by the endocrinologists. The number of 80 scans in both instances is simply a coincidence\u003c/em\u003e.\u003c/p\u003e","description":"","filename":"Fig21.png","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/85ebaef8561d57c9f2f15937.png"},{"id":100126632,"identity":"5b762113-4ec0-47aa-b556-41efbd7c68e2","added_by":"auto","created_at":"2026-01-13 09:25:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":104973,"visible":true,"origin":"","legend":"\u003cp\u003eDetection of existing fractures: comparison AI Vs Radiologists (Panel A) and comparison AI Vs Endocrinologists (Panel B)\u003c/p\u003e","description":"","filename":"Fig31.png","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/ac33872222f295c567aaa951.png"},{"id":107927829,"identity":"4e4686ce-9688-4483-ada8-c6931f2ae382","added_by":"auto","created_at":"2026-04-27 16:05:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":726932,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8461514/v1/bd22177a-e9bc-43c0-bba6-9ce3095f0c02.pdf"}],"financialInterests":"","formattedTitle":"Artificial Intelligence vs Human opportunistic Detection of Vertebral Fractures in Routine CT Scans: results of a pilot study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOsteoporosis is a silent, progressive disease affecting an estimated 200\u0026nbsp;million people worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. It remains underdiagnosed and undertreated, particularly when fragility fractures are not clinically apparent [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Vertebral fractures (VFs) are the most common osteoporotic fractures and represent unequivocal evidence of bone fragility. Despite their prognostic importance, they are often incidental findings and are frequently missed in routine imaging performed for non-musculoskeletal indications. Studies have demonstrated that as many as 60\u0026ndash;70% of VFs visible on chest or abdominal imaging are not reported by radiologists [\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis diagnostic oversight has profound implications. Each missed VF represents a missed opportunity for intervention, increasing the risk of subsequent hip or VFs and associated mortality [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Recent international and Greek Osteoporosis Guidelines emphasize that even a single low-energy morphometric VF constitutes a definitive diagnosis of osteoporosis, mandating immediate initiation of pharmacological treatment regardless of bone mineral density [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Identifying such fractures opportunistically could thus shift the paradigm from reactive to preventive osteoporosis care in Greece, a country with a grossly estimated osteoporosis treatment gap of 60% [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe growing volume of imaging studies and the limited availability of expert musculoskeletal radiologists further exacerbate underdiagnosis. Workload fatigue and the absence of targeted clinical queries often lead to incomplete interpretation of vertebral morphology, especially in emergency and general hospital environments. Recent years have witnessed an exponential rise in the number of radiological examinations per capita, amplifying the need for automated tools that can assist human readers and standardize reporting accuracy [\u003cspan additionalcitationids=\"CR13 CR14\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eArtificial intelligence (AI) has emerged as a promising solution to these challenges. AI-based image analysis systems utilize deep learning algorithms capable of recognizing subtle, complex patterns beyond human perception. Initial investigations have demonstrated comparable or superior diagnostic accuracy to that of experienced radiologists in fracture detection, particularly when AI acts as an assistive \u0026ldquo;second reader\u0026rdquo; [\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. AI offers the additional advantage of continuous, fatigue-free performance and the potential for deployment in any radiology workflow without significant infrastructure costs.\u003c/p\u003e \u003cp\u003eBeyond accuracy, AI introduces a paradigm of opportunistic screening, leveraging existing imaging data acquired for other indications to detect conditions such as osteoporosis [\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], sarcopenia [\u003cspan additionalcitationids=\"CR23\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], or cardiovascular calcification [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. This approach allows clinicians to extract additional diagnostic information from routine studies, enhancing value without increasing radiation exposure or cost. Such a model of preventive and predictive medicine resonates with the vision that artificial intelligence and wellbeing medicine are defining elements of the 21st-century endocrinology and healthcare delivery [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis is the first pilot study in Greece that evaluates the diagnostic performance of an AI system for opportunistic vertebral fracture detection in routine computer tomography (CT) scans, comparing it directly with both general radiologists and endocrinologists. The study also explores the implications of AI integration for early osteoporosis diagnosis and treatment initiation within the Greek national healthcare framework.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design and Setting\u003c/h2\u003e \u003cp\u003eThis was a prospective, single-center pilot study conducted at the 251 Hellenic Air Force General Hospital in Athens, Greece, between April 14 and June 16, 2025, by the Department of Medical Research. The institution provides tertiary-level services to both military personnel and civilians and includes an established bone metabolism clinic specializing in osteoporosis.\u003c/p\u003e \u003cp\u003eThe study assessed the performance of the Bone Solution HealthOST (Nanox AI Ltd., Tel Aviv, Israel), a CE-marked artificial intelligence platform developed for automated detection of vertebral height loss - compression fractures in chest and abdominal CT scans. The software was provided free of charge for the duration of the pilot through a bilateral institutional agreement. The AI model was trained on large-scale pseudoanonymized datasets encompassing over 100,000 CT slices, enabling recognition of vertebral deformities based on morphometric and textural features.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy Objectives\u003c/h3\u003e\n\u003cp\u003eThe primary objective of the study was to compare the ability of the AI system to \u0026ldquo;flag\u0026rdquo; VFs on routine CT examinations performed for other than osteoporosis reasons with the corresponding routine radiology reports produced by general radiologists in their everyday clinical practice.\u003c/p\u003e \u003cp\u003eSecondary objectives were to:\u003c/p\u003e \u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eCompare AI-based fracture flagging with targeted vertebral fracture assessment performed by endocrinologists evaluating and treating patients with osteoporosis in a designated outpatient clinic.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eCompare the overall ability of AI, radiologists, and endocrinologists to correctly discriminate CT examinations with or without fractures.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e Evaluate the clinical translation of AI findings by determining how many patients with AI-detected, previously unreported vertebral fractures would qualify\u0026mdash;based on the Greek Osteoporosis Guidelines\u0026mdash;for immediate osteoporosis treatment irrespective of BMD testing.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e\n\u003ch3\u003eStudy Population and Data Processing\u003c/h3\u003e\n\u003cp\u003eAll thoracic and abdominal CT scans of patients aged\u0026thinsp;\u0026ge;\u0026thinsp;50 years performed during the two-month period were retrieved from the hospital picture archiving communication system (PACS). Both contrast and non-contrast scans were eligible. When repeated CT scans were performed during a patient's hospital admission only the first scan was evaluated (i.e., up to one abdominal and/or chest CT per patient). CT scans obtained in the same patient and anatomical site more than once during the study period were considered as separate examinations only if the time interval between them exceeded 30 days. Exclusion criteria included suboptimal image quality, missing radiology reports, or incomplete spine visualization.\u003c/p\u003e \u003cp\u003eAll DICOM files were anonymized and uploaded to the cloud-based AI platform. Following automated analysis, AI-generated reports were returned and integrated into the local database. Each AI report included a classification of \u0026ldquo;fracture\u0026rdquo; or \u0026ldquo;no fracture\u0026rdquo; along with localization on sagittal reformats and the density of the vertebrae in Hounsfield units (HU) (Fig.\u0026nbsp;1); included fractures were either Grade 2 or Grade 3 according to the semi-quantitative visual grading of vertebral fractures [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003eHuman Readers and Reference Standard\u003c/h3\u003e\n\u003cp\u003eThree diagnostic modalities were compared:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eRoutine Radiology Reports \u0026ndash; generated by general radiologists unaware of the actual study period.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eEndocrinologist Review \u0026ndash; conducted by endocrinologists evaluating and treating osteoporosis patients who systematically examined each scan solely for vertebral deformities.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eAI System Output \u0026ndash; the automated classification produced by the Bone Solution HealthOST platform.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eThe verification of the fracture diagnosis which served as the \u0026ldquo;gold standard\u0026rdquo; was established through consensus between a senior radiologist (NK) and an endocrinologist (PM) specialized in bone diseases, both blinded to the AI and human readings, during combined assessment to produce the final ground-truth dataset.\u003c/p\u003e \u003cp\u003eThe study did not include an absolute quality control of the diagnostic classifications. In particular, it was not assessed whether, for example, the AI and/or the radiologists or endocrinologists correctly graded a fracture as Grade 2 or Grade 3, or whether they accurately identified all existing fractures in each CT scan \u0026mdash; such as for example all three present fractures rather than only one or two. Therefore, the study\u0026rsquo;s design allowed assessment of AI as a platform for opportunistic osteoporosis case-finding in routine clinical practice rather than strictly as a diagnostic tool.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eTo estimate the relative probabilities of correct \u0026ldquo;flagging\u0026rdquo;, odds ratios (ORs) were computed for the comparisons (AI vs. Radiologists, AI vs. Endocrinologists). ORs were derived from contingency tables of true positives and false positives, applying Woolf\u0026rsquo;s method (log-transformation of the OR) for the calculation of 95% confidence intervals and two-sided Wald tests for p-values (H₀: log(OR)\u0026thinsp;=\u0026thinsp;0). Statistical analyses were also conducted to compare the diagnostic performance of the Artificial Intelligence (AI) system with that of human specialists \u0026mdash;radiologists and endocrinologists\u0026mdash; in detecting or excluding vertebral fractures on CT scans. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated using 2\u0026times;2 contingency tables derived from the reference standard (gold standard). Although 95% confidence intervals were not directly applied to the proportions, group comparisons were assessed using the chi-square (χ\u0026sup2;) test of independence. For each diagnostic metric, χ\u0026sup2; tests were performed to compare proportions between the AI system and each corresponding reader group (AI vs. Radiologists, AI vs. Endocrinologists), and p-values are reported. Where appropriate, comparisons were visualized using bar plots. All analyses were performed in Python (SciPy, NumPy, Matplotlib), with statistical significance set at p\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eEthical Considerations\u003c/h2\u003e \u003cp\u003e The study was approved by the Institutional Review Board of the 251 Hellenic Air Force General Hospital (IRB No: F.076/6828/S.3128/4-6-24). All patient data were anonymized before analysis, and the IRB waived the requirement of written informed consent. The study was conducted in accordance with the Declaration of Helsinki (1975, revised 2013). No external funding was received for this project.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStudy Population\u003c/h2\u003e \u003cp\u003eIn total, 699 CTs were initially screened; 427 scans met the inclusion criteria and were analyzed corresponding to 331 patients (136 females, mean age: 72.9\u0026thinsp;\u0026plusmn;\u0026thinsp;13.5 years). Of the 427 included CT scans, 52% were thoracic and 48% abdominal. Indications for imaging included oncologic investigation and/or follow-up, known or suspected pulmonary disease, chronic or acute abdominal pathology. None of the CTs were performed specifically for evaluation of clinically suspected fragility fractures.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eAI vs. Human Readers\u003c/h2\u003e \u003cp\u003eThe analysis of the study\u0026rsquo;s primary endpoint was essentially designed to evaluate the ability to detect (flag) the presence of one or more VF in each CT examination that was ultimately confirmed by the gold standard evaluation (case finding). Accordingly, this analysis focused only on the assessment of \u0026ldquo;positive\u0026rdquo; findings related to the existence of a fracture and did not include the negative diagnoses.\u003c/p\u003e \u003cp\u003eA total of 80 examinations were flagged as having\u0026thinsp;\u0026ge;\u0026thinsp;1 vertebral fracture either by the AI system and/or by radiologists in their routine clinical reports. In the evaluation of flagging performance that ultimately corresponded to CT examinations harboring true fractures as assessed by the expert panel (case finding), the AI system demonstrated statistically significant superiority over radiologists (OR\u0026thinsp;=\u0026thinsp;14.23; 95% CI: 4.7\u0026ndash;43.1; p\u0026thinsp;\u0026lt;\u0026thinsp;0.000001) (Fig.\u0026nbsp;2a).\u003c/p\u003e \u003cp\u003eThe same comparative approach was applied to the secondary endpoint assessing fracture-flagging performance between the AI system and endocrinologists. In this analysis, the AI system was again statistically superior to endocrinologists (OR\u0026thinsp;=\u0026thinsp;4.77; 95% CI: 1.6\u0026ndash;13.6; p\u0026thinsp;=\u0026thinsp;0.001) (Fig.\u0026nbsp;2b).\u003c/p\u003e \u003cp\u003eWith respect to the secondary objective \u0026mdash; comparing the ability to correctly discriminate CT examinations with or without fractures (AI vs. radiologists and AI vs. endocrinologists) \u0026mdash; these analyses aimed to assess the sum of both \u0026ldquo;positive\u0026rdquo; and \u0026ldquo;negative\u0026rdquo; findings regarding the presence of fracture(s). In this per-examination analysis (detection of existing fractures in the whole cohort), the odds that the AI system would identify the true vertebral fractures were 6.3-fold higher than those of radiologists (OR\u0026thinsp;=\u0026thinsp;6.30; 95% CI: 2.90\u0026ndash;13.58; p\u0026thinsp;\u0026lt;\u0026thinsp;0.000001). The AI system also demonstrated higher sensitivity compared with radiologists (86.3% vs. 50%, p\u0026thinsp;\u0026lt;\u0026thinsp;0.000001) and a higher negative predictive value (96.9% vs. 89.6%, p\u0026thinsp;=\u0026thinsp;0.0002) (Fig.\u0026nbsp;3a).\u003c/p\u003e \u003cp\u003eIn the analogous per-examination comparison between the AI system and endocrinologists, the odds of the AI system identifying existing vertebral fractures were 2.85-fold higher (OR\u0026thinsp;=\u0026thinsp;2.85; 95% CI: 1.29\u0026ndash;6.30; p\u0026thinsp;=\u0026thinsp;0.009). The AI system again showed higher sensitivity (86.3% vs. 68.8%, p\u0026thinsp;=\u0026thinsp;0.01) and a higher negative predictive value (96.9% vs. 93.3%, p\u0026thinsp;=\u0026thinsp;0.04), although it exhibited slightly lower specificity (98.3% vs. 100%, p\u0026thinsp;=\u0026thinsp;0.04) (Fig.\u0026nbsp;3b).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eMissed Fractures and Clinical Translation\u003c/h2\u003e \u003cp\u003eAI uniquely detected 33 CT scans with true fractures missed by the routine radiology reports which corresponded to 30 individual patients. Among these patients, 23 had never received osteoporosis therapy, 4 patients were currently under treatment, and 3 had discontinued treatment more than three years earlier. According to the Greek guidelines [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], these 26 individuals required immediate pharmacological therapy independent of BMD testing.\u003c/p\u003e \u003cp\u003eExtrapolating this 2-months detection rate across one year suggested that routine AI deployment in this single hospital setting could identify approximately 156 new treatable osteoporosis cases annually\u0026mdash;patients who would otherwise remain undiagnosed and at elevated risk of future fractures.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eAnatomical Distribution and Error Analysis\u003c/h2\u003e \u003cp\u003eThe majority of fractures were located between Th7 and L4, while the vertebrae with the highest fracture frequency were Th12 and L1 consistent with mechanical stress points.\u003c/p\u003e \u003cp\u003eAI missed few true fractures in cases with severe scoliosis while occasionally overcalled mild degenerative deformities in older patients, explaining the small number of false positives (Fig.\u0026nbsp;2). However, the majority of discordant readings between AI and radiologists represented genuine missed fractures rather than overdiagnosis.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis pilot study demonstrates that a commercially available AI system can markedly outperform both general radiologists and endocrinologists in the opportunistic detection of vertebral fractures. The magnitude of improvement in case finding \u0026mdash;over 14-fold higher odds compared with standard reporting\u0026mdash; highlights a significant and actionable diagnostic gap in current clinical practice.\u003c/p\u003e \u003cp\u003eEvery missed vertebral fracture represents a missed opportunity for secondary prevention. In most osteoporosis guidelines, the identification of a low-energy vertebral fracture establishes a diagnosis of osteoporosis requiring treatment initiation [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Early recognition of non-clinical (morphometric) vertebral fractures allows clinicians to intervene before subsequent fractures occur, improving patient outcomes and reducing healthcare costs. This principle aligns with broader international recommendations advocating for opportunistic case finding through routine imaging [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Recent reports and meta-analyses demonstrated that AI tools consistently achieve pooled sensitivities around 90% across anatomical regions and yield maximal benefit when combined with human oversight [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Additionally, explainable foundation models capable of opportunistic osteoporosis screening from chest X-rays have been recently developed, achieving high area under the curve (AUC) values without reliance on bone density data [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Our findings reinforce these results in a real-world setting, further supporting the reliability of CE-marked AI systems when integrated into clinical workflows.\u003c/p\u003e \u003cp\u003eIn practice, the integration of AI into PACS could enable automatic flagging of suspected vertebral fractures, prompting radiologists or clinicians to review the flagged images. Such an approach would require minimal human input and no additional imaging. Moreover, automated alerts could be linked to hospital electronic medical records, facilitating direct referral to osteoporosis clinics for therapy initiation transforming the CT imaging into an active screening tool rather than a merely targeted diagnostic test. Future iterations of systems like the one we used in our pilot study could adopt similar visual interpretability modules, displaying heatmaps over detected vertebral deformities to further aid user verification.\u003c/p\u003e \u003cp\u003eFrom a population perspective, opportunistic vertebral fracture detection could dramatically reduce the \u0026ldquo;osteoporosis care gap\u0026rdquo; [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. In Greece, where adherence to screening and treatment remains suboptimal [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], automatic identification of high-risk individuals during routine CT imaging offers a sustainable, cost-neutral intervention. Even modest improvements in detection rates could translate into hundreds of prevented fractures annually, reducing morbidity, mortality, and healthcare expenditure. In this regard and specifically for the 251 Hellenic Air Force General Hospital in which the first Greek Fracture Liaison Service (FLS) was implemented, a total of 213 incident fragility fractures were identified during the 1st year period, comprising approximately 26 vertebral, 51 hip, and 136 non-vertebral fractures [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. In this context, the projected annual detection of approximately 156 previously unrecognized, treatment-eligible patients through routine AI-assisted CT analysis represents a clinically meaningful expansion of case-finding capacity. Within an established FLS framework, such an increase could nearly double the number of vertebral and other major osteoporotic fractures captured and appropriately referred for secondary fracture prevention, thereby significantly amplifying the clinical impact and effectiveness of the FLS model.\u003c/p\u003e \u003cp\u003eBeyond diagnostic efficiency, the implications of AI extend to the evolving concept of precision and wellbeing medicine. Artificial intelligence appears to be not only a technological novelty but a fundamental enabler of predictive endocrinology, where data-driven insights enhance disease prevention and health maintenance [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. In this context, the integration of AI into bone health management might reflect a transition toward holistic, anticipatory care.\u003c/p\u003e \u003cp\u003eThe study has inherent limitations. It was single center, with a relatively short observation period and a moderate sample size while the reference standard was based on expert consensus. Furthermore, the AI model\u0026rsquo;s training dataset was international, and its performance might vary with different scanners or population demographics.\u003c/p\u003e \u003cp\u003eFuture research should prioritize multicenter validation of AI-assisted vertebral fracture detection across diverse patient populations to ensure generalizability and robustness. Equally important is the integration of AI outputs with established clinical tools, such as FRAX scores and BMD measurements, to develop hybrid risk-stratification models with superior predictive performance. Rigorous health-economic analyses are also needed to quantify the potential cost savings arising from earlier diagnosis and subsequent fracture prevention. In parallel, the development of explainable AI frameworks will be essential to enhance algorithmic transparency and strengthen clinician trust. Finally, evidence generated from these efforts should inform national-level policy frameworks that support the widespread implementation of AI-assisted opportunistic screening within routine clinical practice.\u003c/p\u003e \u003cp\u003eIn conclusion, in this first Greek pilot study, AI achieved significantly higher sensitivity and diagnostic yield for vertebral fracture detection than both radiologists and endocrinologists. Given the central role of radiologists in the diagnostic procedure, AI integration into radiology workflows could convert routine CT scans into powerful, cost-free screening instruments for osteoporosis thereby facilitating early detection and treatment.\u003c/p\u003e"},{"header":"Declarations","content":" \u003cp\u003e \u003cb\u003eFunding declaration\u003c/b\u003e:\u003c/p\u003e \u003cp\u003eNo funding was received for this study.\u003c/p\u003e\u003ch2\u003eCompliance with ethical standards:\u003c/h2\u003e \u003cp\u003e \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eConflict of Interest\u003c/span\u003e: The authors have no competing interests to declare that are relevant to the content of this article.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eEthical Approval:\u003c/strong\u003e \u003cp\u003eThe study was approved by the Institutional Review Board of the 251 Hellenic Air Force General Hospital (IRB No: F.076/6828/S.3128/4-6-24).\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eInformed Consent:\u003c/strong\u003e \u003cp\u003eAll patient data were anonymized before analysis, and the IRB waived the requirement of written informed consent.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eAuthor Contribution declaration\u003c/h2\u003e \u003cp\u003eConceptualization: Polyzois Makras; Methodology: Polyzois Makras, Nikolaos Kyriakopoulos; Formal analysis and investigation: Polyzois Makras, Eleftherios Chatzellis, Christos Gravvanis, Georgia Kanti, Nikolaos Kyriakopoulos; Data analysis and statistics: Polyzois Makras, Maria P. Yavropoulou; Writing - original draft preparation: Polyzois Makras; Writing - review and editing: Polyzois Makras, Eleftherios Chatzellis, Christos Gravvanis, Georgia Kanti, Maria P. Yavropoulou, Konstantinos Papadimitropoulos, Nikolaos Kyriakopoulos; Supervision: Polyzois Makras, Konstantinos Papadimitropoulos\u003c/p\u003e \u003cp\u003eCT reviews in routine clinical practice: Radiology Department reporting group\u003c/p\u003e \u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eClynes MA, Harvey NC, Curtis EM, Fuggle NR, Dennison EM, Cooper C (2020) The epidemiology of osteoporosis. Br Med Bull 133(1):105\u0026ndash;117\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCurtis EM, Dennison EM, Cooper C, Harvey NC (2022) Osteoporosis in 2022: Care gaps to screening and personalised medicine. Best Pract Res Clin Rheumatol 36(3):101754\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanis JA, Norton N, Harvey NC, Jacobson T, Johansson H, Lorentzon M, McCloskey EV, Willers C, Borgstrom F (2021) SCOPE 2021: a new scorecard for osteoporosis in Europe. Arch Osteoporos 16(1):82\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHusarek J, Hess S, Razaeian S, Ruder TD, Sehmisch S, Muller M, Liodakis E (2024) Artificial intelligence in commercial fracture detection products: a systematic review and meta-analysis of diagnostic test accuracy. Sci Rep 14(1):23053\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMuller M, Schechter CB, Hautz WE, Sauter TC, Exadaktylos AK, Stock S, Birrenbach T (2021) The development and validation of a resource consumption score of an emergency department consultation. PLoS ONE 16(2):e0247244\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBruls RJM, Kwee RM (2020) Workload for radiologists during on-call hours: dramatic increase in the past 15 years. Insights Imaging 11(1):121\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eToth E, Banefelt J, Akesson K, Spangeus A, Ortsater G, Libanati C (2020) History of Previous Fracture and Imminent Fracture Risk in Swedish Women Aged 55 to 90 Years Presenting With a Fragility Fracture. J Bone Min Res 35(5):861\u0026ndash;868\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanis JA, Cooper C, Rizzoli R, Reginster JY, Scientific Advisory Board of the European Society for Clinical and Economic Aspects of Osteoporosis, Committees of Scientific Advisors and National Societies of the International Osteoporosis Foundation (2019) European guidance for the diagnosis and management of osteoporosis in postmenopausal women. Osteoporos Int 30(1):3\u0026ndash;44\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShoback D, Rosen CJ, Black DM, Cheung AM, Murad MH, Eastell R (2020) Pharmacological Management of Osteoporosis in Postmenopausal Women: An Endocrine Society Guideline Update. J Clin Endocrinol Metab 105(3)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMakras P, Anastasilakis AD, Antypas G, Chronopoulos E, Kaskani EG, Matsouka A, Patrikos DK, Stathopoulos KD, Tournis S, Trovas G, Kosmidis C (2019) The 2018 Guidelines for the diagnosis and treatment of osteoporosis in Greece. Arch Osteoporos 14(1):39\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYavropoulou MP, Makras P, Athanasakis K, Bournia VK, Mathioudakis K, Tsolakidis A, Kassi E, Kaltsas G, Mitrou P, Sfikakis PP (2022) Prevalence and patterns of anti-osteoporotic drug use based on 2019 real-world nationwide data in Greece. Arch Osteoporos 17(1):86\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDi Somma S, Paladino L, Vaughan L, Lalle I, Magrini L, Magnanti M (2015) Overcrowding in emergency department: an international issue. Intern Emerg Med 10(2):171\u0026ndash;175\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLantsman CD, Barash Y, Klang E, Guranda L, Konen E, Tau N (2022) Trend in radiologist workload compared to number of admissions in the emergency department. Eur J Radiol 149:110195\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmith E, Dasan S (2018) A system under pressure. Br J Hosp Med (Lond) 79(9):495\u0026ndash;499\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMdodo R, Frazier EL, Dube SR, Mattson CL, Sutton MY, Brooks JT, Skarbinski J (2015) Cigarette smoking prevalence among adults with HIV compared with the general adult population in the United States: cross-sectional surveys. Ann Intern Med 162(5):335\u0026ndash;344\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCanoni-Meynet L, Verdot P, Danner A, Calame P, Aubry S (2022) Added value of an artificial intelligence solution for fracture detection in the radiologist's daily trauma emergencies workflow. Diagn Interv Imaging 103(12):594\u0026ndash;600\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang S, Yin B, Cao W, Feng C, Fan G, He S (2020) Diagnostic accuracy of deep learning in orthopaedic fractures: a systematic review and meta-analysis. Clin Radiol 75(9):713e17\u0026ndash;713e28\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuo RYL, Harrison C, Curran TA, Jones B, Freethy A, Cussons D, Stewart M, Collins GS, Furniss D (2022) Artificial Intelligence in Fracture Detection: A Systematic Review and Meta-Analysis. Radiology 304(1):50\u0026ndash;62\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang J, Liao M, Wang Y, Chen L, He L, Ji Y, Xiao Y, Lu Y, Fan W, Nie Z, Wang R, Qi B, Yang F (2022) Opportunistic osteoporosis screening using chest CT with artificial intelligence. Osteoporos Int 33(12):2547\u0026ndash;2561\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu Y, Yang X, Wang M, Lian Y, Hou P, Chai X, Dai Q, Qian B, Jiang Y, Gao J (2025) Artificial intelligence assisted automatic screening of opportunistic osteoporosis in computed tomography images from different scanners. Eur Radiol 35(4):2287\u0026ndash;2295\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBehanova M, Sokhan A, Haschka J, Zandieh S, Salzlechner C, Ljuhar R, Zwerina J, Kocijan R (2026) AI-supported opportunistic detection of vertebral fractures on routine CT scans: Diagnostic performance and clinical relevance. Bone 203:117735\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUrooj B, Ko Y, Na S, Kim IO, Lee EH, Cho S, Jeong H, Khang S, Lee J, Kim KW (2025) Implementation of Fully Automated AI-Integrated System for Body Composition Assessment on Computed Tomography for Opportunistic Sarcopenia Screening: Multicenter Prospective Study. JMIR Form Res 9:e69940\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePickhardt PJ, Correale L, Hassan C (2023) AI-based opportunistic CT screening of incidental cardiovascular disease, osteoporosis, and sarcopenia: cost-effectiveness analysis. Abdom Radiol (NY) 48(3):1181\u0026ndash;1198\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRyu J, Eom S, Kim HC, Kim CO, Rhee Y, You SC, Hong N (2023) Chest X-ray-based opportunistic screening of sarcopenia using deep learning. J Cachexia Sarcopenia Muscle 14(1):418\u0026ndash;428\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJeong J, Chao CJ, Arsanjani R, Ayoub C, Lester SJ, Pereyra M, Said EF, Roarke M, Tagle-Cornell C, Koepke LM, Tsai YL, Jung-Hsuan C, Chang CC, Farina JM, Trivedi H, Patel BN, Banerjee I (2025) Artificial Intelligence Chest X-Ray Opportunistic Screening Model for Coronary Artery Calcium Deposition: A Multi-Objective Model With Multimodal Data Fusion. Mayo Clin Proc Digit Health 3(4):100300\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStratakis CA (2025) Artificial intelligence and quality of life are two leading themes of contemporary medicine. Horm (Athens) 24(3):589\u0026ndash;591\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGenant HK, Wu CY, van Kuijk C, Nevitt MC (1993) Vertebral fracture assessment using a semiquantitative technique. J Bone Min Res 8(9):1137\u0026ndash;1148\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAggarwal V, Maslen C, Abel RL, Bhattacharya P, Bromiley PA, Clark EM, Compston JE, Crabtree N, Gregory JS, Kariki EP, Harvey NC, Ward KA, Poole KES (2021) Opportunistic diagnosis of osteoporosis, fragile bone strength and vertebral fractures from routine CT scans; a review of approved technology systems and pathways to implementation. Ther Adv Musculoskelet Dis 13:1759720X211024029. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/1759720X211024029\u003c/span\u003e\u003cspan address=\"10.1177/1759720X211024029\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReginster JY, Schmidmaier R, Alokail M, Hiligsmann M (2025) Cost-effectiveness of opportunistic osteoporosis screening using chest radiographs with deep learning in Germany. Aging Clin Exp Res 37(1):149\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim J, Kwak S, Lee H, Chang J, Park SM (2025) Explainable opportunistic osteoporosis screening from chest X-rays: a retrospective comparison of foundation models. Osteoporos Int. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00198-025-07727-3\u003c/span\u003e\u003cspan address=\"10.1007/s00198-025-07727-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWillers C, Norton N, Harvey NC, Jacobson T, Johansson H, Lorentzon M, McCloskey EV, Borgstrom F, Kanis JA Scientific Report Panel of the IOF (2022) Osteoporosis in Europe: a compendium of country-specific reports. Arch Osteoporos 17(1):23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11657-021-00969-8\u003c/span\u003e\u003cspan address=\"10.1007/s11657-021-00969-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMakras P, Panagoulia M, Mari A, Rizou S, Lyritis GP (2017) Evaluation of the first fracture liaison service in the Greek healthcare setting. Arch Osteoporos 12(1):3. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11657-016-0299-7\u003c/span\u003e\u003cspan address=\"10.1007/s11657-016-0299-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\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":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"hormones","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"HORM","sideBox":"Learn more about [Hormones](https://www.springer.com/journal/42000)","snPcode":"42000","submissionUrl":"https://www.editorialmanager.com/horm/default2.aspx","title":"Hormones","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Artificial intelligence, Vertebral fractures, Computed tomography, Opportunistic screening, Osteoporosis, Diagnostic accuracy, Endocrinology, Preventive medicine","lastPublishedDoi":"10.21203/rs.3.rs-8461514/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8461514/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eVertebral fractures (VFs) are the most common osteoporotic fractures and a hallmark of bone fragility, yet most morphometric VFs remain undiagnosed in routine imaging not focused on the spine. This pilot study evaluated the diagnostic performance of an artificial intelligence (AI) system for opportunistic vertebral fracture detection in chest and abdominal computer tomography (CT) scans compared with human readers in a real-world hospital setting.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eOver two months, all thoracic and abdominal CTs performed for any indication in patients aged\u0026thinsp;\u0026ge;\u0026thinsp;50 years at a tertiary hospital were analyzed by the Bone Solution HealthOST AI platform (Nanox AI Ltd.). Routine radiology reports and targeted reviews by endocrinologists were compared to AI outputs, using a gold-standard adjudication by an expert panel. Sensitivity, specificity, predictive values, and odds ratios (ORs) were calculated and compared by χ\u0026sup2; tests (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eAmong 427 eligible CT scans, AI detected vertebral fractures with significantly higher sensitivity (86.3%) than radiologists (50.0%) or endocrinologists (68.8%) (p\u0026thinsp;\u0026le;\u0026thinsp;0.01 for both comparisons). Compared with radiologists, AI had 14.2-fold greater odds of correctly flagging true fractures (OR\u0026thinsp;=\u0026thinsp;14.23; 95% CI 4.7\u0026ndash;43.1; p\u0026thinsp;\u0026lt;\u0026thinsp;0.000001) while against targeted review by endocrinologists, AI achieved a 4.8-fold advantage (OR\u0026thinsp;=\u0026thinsp;4.77; 95% CI 1.6\u0026ndash;13.6; p\u0026thinsp;=\u0026thinsp;0.001). Extrapolated annually, AI integration could uncover around 150 new, otherwise undiagnosed, patients eligible for osteoporosis treatment.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eAutomated vertebral fracture detection in routine CT scans significantly enhances diagnostic yield versus human interpretation. Its integration into clinical workflows offers a cost-free, high-impact strategy to improve early osteoporosis diagnosis and treatment within the healthcare system.\u003c/p\u003e","manuscriptTitle":"Artificial Intelligence vs Human opportunistic Detection of Vertebral Fractures in Routine CT Scans: results of a pilot study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-13 09:24:10","doi":"10.21203/rs.3.rs-8461514/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revisions","date":"2026-03-03T08:10:40+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2026-01-08T13:28:58+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-01-08T12:24:08+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-12-31T00:41:32+00:00","index":"","fulltext":""},{"type":"submitted","content":"Hormones","date":"2025-12-29T06:44:24+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"hormones","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"HORM","sideBox":"Learn more about [Hormones](https://www.springer.com/journal/42000)","snPcode":"42000","submissionUrl":"https://www.editorialmanager.com/horm/default2.aspx","title":"Hormones","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"47cc7c82-dc22-42ee-9414-5627b8e61aba","owner":[],"postedDate":"January 13th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-04-27T16:02:56+00:00","versionOfRecord":{"articleIdentity":"rs-8461514","link":"https://doi.org/10.1007/s42000-026-00784-1","journal":{"identity":"hormones","isVorOnly":false,"title":"Hormones"},"publishedOn":"2026-04-23 15:59:07","publishedOnDateReadable":"April 23rd, 2026"},"versionCreatedAt":"2026-01-13 09:24:10","video":"","vorDoi":"10.1007/s42000-026-00784-1","vorDoiUrl":"https://doi.org/10.1007/s42000-026-00784-1","workflowStages":[]},"version":"v1","identity":"rs-8461514","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8461514","identity":"rs-8461514","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}