KI zur automatisierten vBMD- und Fragilitätsanalyse des proximalen Femurs an CT-Scans

Zusammenfassung

Hintergrund

Osteoporotische Hüftfrakturen sind mit hoher Morbidität und Mortalität verbunden. Die nebenbefundliche Routineanalyse klinischer CT-Scans im Hinblick auf Frakturrisiko (opportunistisches Screening) könnte Präventionsbedarf frühzeitig aufdecken, bislang fehlt jedoch ein frei verfügbares vollautomatisches Verfahren zur Bestimmung der volumetrischen Knochendichte (vBMD) des proximalen Femurs.

Methoden

Es wurde die open source KI TotalSegmentator mit zwei eigenen KI-Modellen kombiniert, um sowohl das proximale Femur als auch ein Kalibrierphantom zu segmentieren und daraus einen vBMD vollautomatisch zu berechnen. Die Güte der KI vBMD Messungen wurde an 1070 Hüft QCT-Scans der AGES Studie durch den Vergleich mit dem semi-automatischen Goldstandard MIAF ermittelt. Zur ersten Prüfung der Eignung wurden 289 klinische CT-Scans (ARTEMIS Studie) bzgl. der Vorhersage inzidenter Hüftfrakturen analysiert.

Ergebnisse

Die KI HU vBMD Werte korrelierten eng mit den MIAF vBMD Werten (r=0,88–0,97); nach Kalibrierung betrug die Korrelation r=0,96 bei einem Bias von 1,6 mg/cm³ (integral) und 21,9 mg/cm³ (trabekuläre) und RMS-Fehlern von 15,1 mg/cm³ (integral) und 9,8 mg/cm³ (trabekulär). Die prädiktive Güte für Hüftfrakturen (AUC 0,771–0,836) lag signifikant (p<0,031) über dem Basismodell aus Alter und Geschlecht (AUC=0,641).

Schlussfolgerungen

Die entwickelte KI ermöglicht eine vollautomatische, schnelle und kalibrierte Bestimmung der vBMD am proximalen Femur direkt aus klinischen CT-Scans und erlaubt die Vorhersage des Hüftfrakturrisikos. Die positiven Ergebnisse aus dieser ersten Prädiktionsstudie müssen jedoch in einem unabhängigen und größeren Datensatz überprüft werden. Damit eröffnet sich die Möglichkeit, Risikopatienten/-innen im Rahmen des opportunistischen Screenings zu identifizieren und präventive Maßnahmen früher einzuleiten.

Abstract

Background

Osteoporotic hip fractures are associated with high morbidity and mortality. Opportunistic screening by incidental analysis of routine clinical CT scans for fracture risk could reveal the need for prevention at an early stage. However, a freely available fully automated method for determining volumetric bone mineral density (vBMD) of the proximal femur is still lacking.

Methods

The open-source AI tool TotalSegmentator was combined with two in-house AI models to segment both the proximal femur and a calibration phantom, enabling fully automated vBMD calculation. The accuracy of AI vBMD measurements was evaluated in 1070 hip QCT scans from the AGES study by comparison with the semi-automated gold standard MIAF. For an initial assessment of suitability, 289 clinical CT scans (ARTEMIS study) were analyzed regarding prediction of incident hip fractures.

Results

AI HU vBMD values correlated closely with MIAF vBMD values (r=0.88–0.97). After calibration, correlation was r=0.96 with a bias of 1.6 mg/cm³ (integral) and 21.9 mg/cm³ (trabecular), and RMS errors of 15.1 mg/cm³ (integral) and 9.8 mg/cm³ (trabecular). Predictive performance for hip fractures (AUC 0.771–0.836) was significantly higher (p<0.031) than the baseline model of age and sex (AUC=0.641).

Conclusions

The developed AI enables fully automated, rapid, and calibrated assessment of proximal femur vBMD directly from clinical CT scans and allows prediction of hip fracture risk. The positive results of this first prognostic study, however, need to be confirmed in independent and larger datasets. This approach offers the potential to identify at-risk patients in opportunistic screening and to initiate preventive measures at an earlier stage.

Schlüsselwörter

künstliche Intelligenz - Computertomographie - Knochenmineraldichte - inzidentes osteoporotisches Frakturrisiko - Opportunistisches Screening

Keywords

computer tomography - bone mineral density - incident fracture risk - opportunistic screening - artificial intelligence

Publikationsverlauf

Eingereicht: 09. August 2025

Angenommen nach Revision: 06. Oktober 2025

Artikel online veröffentlicht:
14. November 2025

• Literatur

• 1 Hernlund E, Svedbom A, Ivergård M. et al. Osteoporosis in the European Union: medical management, epidemiology and economic burden. Arch Osteoporos 2013; 8: 136
• 2 Cummings SR, Browner W, Cummings SR. et al. Bone density at various sites for prediction of hip fractures. The Lancet 1993; 341: 72-75
• 3 Schuit SCE, van der Klift M, Weel AEAM. et al. Fracture incidence and association with bone mineral density in elderly men and women: the Rotterdam Study. Bone 2004; 34: 195-202
• 4 Yang L, Burton AC, Bradburn M. et al. Distribution of bone density in the proximal femur and its association with hip fracture risk in older men: The osteoporotic fractures in men (MrOS) study. J Bone Miner Res 2012; 27: 2314-2324
• 5 Engelke K, Chaudry O, Bartenschlager S. Opportunistic Screening Techniques for Analysis of CT Scans. Curr Osteoporos Rep 2023; 21: 65-76
• 6 Pickhardt PJ, Pooler BD, Lauder T. et al. Opportunistic Screening for Osteoporosis Using Abdominal Computed Tomography Scans Obtained for Other Indications. Ann Intern Med 2013; 158: 588-595
• 7 Wang L, Museyko O, Su Y. et al. QCT of the femur: Comparison between QCTPro CTXA and MIAF Femur. Bone 2019; 120: 262-270
• 8 Yosibash Z, Katz Y, Nir T. et al. Femurs segmentation by machine learning from CT scans combined with autonomous finite elements in orthopedic and endocrinology applications. Comput Math Appl 2023; 152: 16-27
• 9 Deng Y, Wang L, Zhao C. et al. A deep learning-based approach to automatic proximal femur segmentation in quantitative CT images. Med Biol Eng Comput 2022; 60: 1417-1429
• 10 Xiao J, Li X, Xie Y. et al. Maximum chest CT score is associated with progression to severe illness in patients with COVID-19: a retrospective study from Wuhan, China. BMC Infect Dis 2020; 20: 953
• 11 Patton DM, Henning EN, Goulet RW. et al Deep learning-based segmentation of high-resolution computed tomography image data outperforms commonly used automatic bone segmentation methods. 2021 2021.07.27.453890
• 12 Bjornsson PA, Baker A, Fleps I. et al. Fast and robust femur segmentation from computed tomography images for patient-specific hip fracture risk screening. Comput Methods Biomech Biomed Eng Imaging Vis 2023; 11: 253-265
• 13 Apivanichkul K, Phasukkit P, Dankulchai P. et al. Enhanced Deep-Learning-Based Automatic Left-Femur Segmentation Scheme with Attribute Augmentation. Sensors 2023; 23: 5720
• 14 Sultana J, Naznin M, Faisal TR. SSDL – an automated semi-supervised deep learning approach for patient-specific 3D reconstruction of proximal femur from QCT images. Med Biol Eng Comput 2024; 62: 1409-1425
• 15 Xie W, Chen P, Li Z. et al. A Two stage deep learning network for automated femoral segmentation in bilateral lower limb CT scans. Sci Rep 2025; 15: 9198
• 16 Imani M, Buratto J, Dao T. et al. Deep Learning Technique for Automatic Segmentation of Proximal Hip Musculoskeletal Tissues From CT Scan Images: A MrOS Study. J Cachexia Sarcopenia Muscle 2025; 16: e13728
• 17 Du C, He J, Cheng Q. et al. Automated opportunistic screening for osteoporosis using deep learning-based automatic segmentation and radiomics on proximal femur images from low-dose abdominal CT. BMC Musculoskelet Disord 2025; 26: 378
• 18 Uemura K, Otake Y, Takao M. et al. Development of an open-source measurement system to assess the areal bone mineral density of the proximal femur from clinical CT images. Arch Osteoporos 2022; 17: 17
• 19 Finck MJ, Koser NC, Hövener J-B. et al. FemoraLyze: A Modular Framework for Proximal Femur Analysis Med Imaging Deep Learn-Short Pap. 2025
• 20 Wasserthal J, Breit H-C, Meyer MT. et al. TotalSegmentator: robust segmentation of 104 anatomical structures in CT images. Radiol Artif Intell 2023; 5: e230024
• 21 Isensee F, Jaeger PF, Kohl SAA. et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods 2021; 18: 203-211
• 22 Harris TB, Launer LJ, Eiriksdottir G. et al. Age, Gene/Environment Susceptibility-Reykjavik Study: Multidisciplinary Applied Phenomics. Am J Epidemiol 2007; 165: 1076-1087
• 23 Nicks KM, Amin S, Melton LJ. et al. Three-dimensional structural analysis of the proximal femur in an age-stratified sample of women. Bone 2013; 55: 179-188
• 24 Bousson V, Le Le Bras A, Roqueplan F. et al. Volumetric quantitative computed tomography of the proximal femur: relationships linking geometric and densitometric variables to bone strength. Role for compact bone. Osteoporos Int 2006; 17: 855-864
• 25 Wang L, Yang M, Liu Y. et al. Differences in Hip Geometry Between Female Subjects With and Without Acute Hip Fracture: A Cross-Sectional Case-Control Study. Front Endocrinol 2022; 13
• 26 Jr DWH Lemeshow S, Sturdivant RX. Applied Logistic Regression. John Wiley & Sons; 2013
• 27 Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric Approach on JSTOR. . Im Internet: https://www.jstor.org/stable/2531595?seq=1 Stand: 07.08.2025.
• 28 Engelke K, Adams JE, Armbrecht G. et al. Clinical use of quantitative computed tomography and peripheral quantitative computed tomography in the management of osteoporosis in adults: the 2007 ISCD Official Positions. J Clin Densitom Off J Int Soc Clin Densitom 2008; 11: 123-162
• 29 Kopperdahl DL, Aspelund T, Hoffmann PF. et al. Assessment of Incident Spine and Hip Fractures in Women and Men using Finite Element Analysis of CT Scans. J Bone Miner Res Off J Am Soc Bone Miner Res 2014; 29: 570-580
• 30 Black DM, Bouxsein ML, Marshall LM. et al. Proximal Femoral Structure and the Prediction of Hip Fracture in Men: A Large Prospective Study Using QCT1*. J Bone Miner Res 2008; 23: 1326-1333
• 31 Orwoll ES, Marshall LM, Nielson CM. et al. Finite Element Analysis of the Proximal Femur and Hip Fracture Risk in Older Men*. J Bone Miner Res 2009; 24: 475-483