Advertisement

Deep Learning Classification of Spinal Osteoporotic Compression Fractures on Radiographs using an Adaptation of the Genant Semiquantitative Criteria

Published:March 26, 2022DOI:https://doi.org/10.1016/j.acra.2022.02.020

      Rationale and Objectives

      Osteoporosis affects 9% of individuals over 50 in the United States and 200 million women globally. Spinal osteoporotic compression fractures (OCFs), an osteoporosis biomarker, are often incidental and under-reported. Accurate automated opportunistic OCF screening can increase the diagnosis rate and ensure adequate treatment. We aimed to develop a deep learning classifier for OCFs, a critical component of our future automated opportunistic screening tool.

      Materials and Methods

      The dataset from the Osteoporotic Fractures in Men Study comprised 4461 subjects and 15,524 spine radiographs. This dataset was split by subject: 76.5% training, 8.5% validation, and 15% testing. From the radiographs, 100,409 vertebral bodies were extracted, each assigned one of two labels adapted from the Genant semiquantitative system: moderate to severe fracture vs. normal/trace/mild fracture. GoogLeNet, a deep learning model, was trained to classify the vertebral bodies. The classification threshold on the predicted probability of OCF outputted by GoogLeNet was set to prioritize the positive predictive value (PPV) while balancing it with the sensitivity. Vertebral bodies with the top 0.75% predicted probabilities were classified as moderate to severe fracture.

      Results

      Our model yielded a sensitivity of 59.8%, a PPV of 91.2%, and an F1 score of 0.72. The areas under the receiver operating characteristic curve (AUC-ROC) and the precision-recall curve were 0.99 and 0.82, respectively.

      Conclusion

      Our model classified vertebral bodies with an AUC-ROC of 0.99, providing a critical component for our future automated opportunistic screening tool. This could lead to earlier detection and treatment of OCFs.

      Keywords

      Abbreviations:

      AUC-PR (area under the precision-recall curve), ACU-ROC (area under the receiver operating characteristic curve), CI (confidence interval), FDR (false discovery rate), GPU (graphics processing unit), ILSVRC2012 (ImageNet Large Scale Visual Recognition Challenge 2012), MrOS (Osteoporotic Fractures in Men), NPV (negative predictive value), OCF (osteoporotic compression fracture), PPV (positive predictive value), PR (precision-recall), SQ (semiquantitative)
      To read this article in full you will need to make a payment

      Purchase one-time access:

      Academic & Personal: 24 hour online accessCorporate R&D Professionals: 24 hour online access
      One-time access price info
      • For academic or personal research use, select 'Academic and Personal'
      • For corporate R&D use, select 'Corporate R&D Professionals'

      Subscribe:

      Subscribe to Academic Radiology
      Already a print subscriber? Claim online access
      Already an online subscriber? Sign in
      Institutional Access: Sign in to ScienceDirect

      References

        • Looker AC
        • Borrud LG
        • Dawson-Hughes B
        • et al.
        Osteoporosis or low bone mass at the femur neck or lumbar spine in older adults, United States, 2005-2008.
        NCHS Data Brief. 2012; 93: 1-8
        • Kanis JA
        on behalf of the World Health Organization Scientific Group. Assessment of osteoporosis at the primary health-care level. Technical Report.
        WHO Collaborating Centre for Metabolic Bone Diseases, University of Sheffield, UK2007
        • Hodsman AB
        • Leslie WD
        • Tsang JF
        • et al.
        10-year probability of recurrent fractures following wrist and other osteoporotic fractures in a large clinical cohort: an analysis from the Manitoba Bone Density Program.
        JAMA Int Med. 2008; 168: 2261-2267
        • Roux S
        • Cabana F
        • Carrier N
        • et al.
        The world health organization Fracture Risk Assessment Tool (FRAX) underestimates incident and recurrent fractures in consecutive patients with fragility fractures.
        J Clin Endocrinol Metab. 2014; 99: 2400-2408
        • Robinson CM
        • Royds M
        • Abraham A
        • et al.
        Refractures in patients at least forty-five years old: a prospective analysis of twenty-two thousand and sixty patients.
        J Bone Joint Surg Am. 2002; 84: 1528-1533
        • Center JR
        • Nguyen TV
        • Schneider D
        • et al.
        Mortality after all major types of osteoporotic fracture in men and women: an observational study.
        Lancet North Am Ed. 1999; 353: 878-882
        • Meadows ES
        • Whangbo A
        • McQuarrie N
        • et al.
        Compliance with mammography and bone mineral density screening in women at least 50 years old.
        Menopause. 2011; 18: 794-801
        • Jain S
        • Bilori B
        • Gupta A
        • et al.
        Are men at high risk for osteoporosis underscreened? A quality improvement project.
        Perm J. 2016; 20: 60-64
        • King AB
        • Fiorentino DM.
        Medicare payment cuts for osteoporosis testing reduced use despite tests’ benefit in reducing fractures.
        Health Aff. 2011; 30: 2362-2370
        • 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
        • Anderson PA
        • Polly DW
        • Binkley NC
        • et al.
        Clinical use of opportunistic computed tomography screening for osteoporosis.
        JBJS. 2018; 100: 2073-2081
        • Alacreu E
        • Moratal D
        • Arana E.
        Opportunistic screening for osteoporosis by routine CT in Southern Europe.
        Osteopor Int. 2017; 28: 983-990
        • Li YL
        • Wong KH
        • Law MW
        • et al.
        Opportunistic screening for osteoporosis in abdominal computed tomography for Chinese population.
        Arch Osteopor. 2018; 13: 1-7
        • Cheng X
        • Zhao K
        • Zha X
        • et al.
        Opportunistic screening using low-dose CT and the prevalence of osteoporosis in China: a nationwide, multicenter study.
        J Bone Miner Res. 2021; 36: 427-435
        • Fang Y
        • Li W
        • Chen X
        • et al.
        Opportunistic osteoporosis screening in multi-detector CT images using deep convolutional neural networks.
        Eur Radiol. 2021; 31: 1831-1842
        • Nam KH
        • Seo I
        • Kim DH
        • et al.
        Machine learning model to predict osteoporotic spine with hounsfield units on lumbar computed tomography.
        J Korean Neurosurg Soc. 2019; 62: 442-449
        • Löffler MT
        • Jacob A
        • Scharr A
        • et al.
        Automatic opportunistic osteoporosis screening in routine CT: improved prediction of patients with prevalent vertebral fractures compared to DXA.
        Eur Radiol. 2021; 31: 6069-6077
        • Yasaka K
        • Akai H
        • Kunimatsu A
        • et al.
        Prediction of bone mineral density from computed tomography: application of deep learning with a convolutional neural network.
        Eur Radiol. 2020; 30: 3549-3557
        • Bar A
        • Wolf L
        • Amitai OB
        • et al.
        Compression fractures detection on CT.
        in: Proceedings of SPIE Medical Imaging: Computer-Aided Diagnosis. International Society for Optics and Photonics, Orlando, FL20171013440
        • Yilmaz EB
        • Buerger C
        • Fricke T
        • et al.
        Automated deep learning-based detection of osteoporotic fractures in CT images.
        in: Proceedings of Machine Learning in Medical Imaging. Springer, Strasbourg, France. Cham, Switzerland2021: 376-385
        • Husseini M
        • Sekuboyina A
        • Bayat A
        • et al.
        Conditioned variational auto-encoder for detecting osteoporotic vertebral fractures.
        in: Proceedings of the International Workshop and Challenge on Computational Methods and Clinical Applications for Spine Imaging. Springer, Granada, Spain. Cham, Switzerland2019: 29-38
        • Tomita N
        • Cheung YY
        • Hassanpour S.
        Deep neural networks for automatic detection of osteoporotic vertebral fractures on CT scans.
        Comput Biol Med. 2018; 98: 8-15
        • Lee S
        • Choe EK
        • Kang HY
        • et al.
        The exploration of feature extraction and machine learning for predicting bone density from simple spine X-ray images in a Korean population.
        Skeletal Radiol. 2020; 49: 613-618
        • Zhang B
        • Yu K
        • Ning Z
        • et al.
        Deep learning of lumbar spine X-ray for osteopenia and osteoporosis screening: A multicenter retrospective cohort study.
        Bone. 2020; 140115561
        • Murata K
        • Endo K
        • Aihara T
        • et al.
        Artificial intelligence for the detection of vertebral fractures on plain spinal radiography.
        Sci Rep. 2020; 10: 1-8
        • Chou PH
        • Jou TH
        • Wu HT
        • et al.
        Ground truth generalizability affects performance of the artificial intelligence model in automated vertebral fracture detection on plain lateral radiographs of the spine.
        Spine J. 2022; 22: 511-523https://doi.org/10.1016/j.spinee.2021.10.020
      1. IMV reports general X-ray procedures growing at 5.5% per year, as number of installed X-ray units declines.
        CISION PRWeb. 2020; (Accessed October 22)
        • Bolotin HH.
        DXA in vivo BMD methodology: an erroneous and misleading research and clinical gauge of bone mineral status, bone fragility, and bone remodelling.
        Bone. 2007; 41: 138-154
        • Kim TY
        • Schafer AL.
        Variability in DXA reporting and other challenges in osteoporosis evaluation.
        JAMA Intern. Med. 2016; 176: 393-395
        • Carberry GA
        • Pooler BD
        • Binkley N
        • et al.
        Unreported vertebral body compression fractures at abdominal multidetector CT.
        Radiology. 2013; 268: 120-126
        • Khan S
        • Rahmani H
        • Shah SA
        • et al.
        A Guide to Convolutional Neural Networks for Computer Vision.
        Morgan & Claypool, 2018
        • Szegedy C
        • Liu W
        • Jia Y
        • et al.
        Going deeper with convolutions.
        in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE Computer Society, Boston, MA. Washington, D.C.2015: 1-9
        • Bianco S
        • Cadene R
        • Celona L
        • et al.
        Benchmark analysis of representative deep neural network architectures.
        IEEE Access. 2018; 6: 64270-64277
        • Orwoll E
        • Blank JB
        • Barrett-Connor E
        • et al.
        Design and baseline characteristics of the osteoporotic fractures in men (MrOS) study-a large observational study of the determinants of fracture in older men.
        Contemp Clin Trials. 2005; 26: 569-585
        • Cawthon PM
        • Haslam J
        • Fullman R
        • et al.
        Osteoporotic Fractures in Men (MrOS) Research Group. Methods and reliability of radiographic vertebral fracture detection in older men: the osteoporotic fractures in men study.
        Bone. 2014; 67: 152-155
        • Genant HK
        • Wu CY
        • van Kuijk C
        • et al.
        Vertebral fracture assessment using a semiquantitative technique.
        J Bone Miner Res. 1993; 8: 1137-1148
        • Gonzalez RC
        • Woods RE.
        Digital Image Processing.
        4th ed. Pearson, New York, NY2018
      2. imgaug: Read the Docs. https://imgaug.readthedocs.io/en/latest/source/api_imgaug.html. Updated 2020. Accessed September 7, 2020.

      3. TensorFlow. https://www.tensorflow.org. Accessed July 25, 2020.

        • Silberman N
        • Guadarrama S
        TF-Slim: a high level library to define complex models in TensorFlow.
        Google AI Blog. 2020; (Published August 30, 2016. Accessed July 25)
      4. GoogLeNet-Inception. GitHub. https://github.com/conan7882/GoogLeNet-Inception. Accessed February 5, 2022.

      5. TensorFlow Model Garden. GitHub. https://github.com/tensorflow/models. Updated July 24, 2020. Accessed July 25, 2020.

        • Russakovsky O
        • Deng J
        • Su H
        • et al.
        ImageNet large scale visual recognition challenge.
        IJCV. 2015; 115: 211-252
        • He K
        • Zhang X
        • Ren S
        • et al.
        Delving deep into rectifiers: surpassing human-level performance on ImageNet classification.
        in: Proceedings of the IEEE international conference on computer vision. IEEE Computer Society, Las Condes, Chile. Washington, D.C.2015: 1026-1034
        • Kingma DP
        • Ba J.
        Adam: a method for stochastic optimization.
        in: Proceedings of the 3rd International Conference on Learning Representations. NY: Association for Computing Machinery, San Diego, CA. New York2015
      6. tf.nn.weighted_cross_entropy_with_logits. TensorFlow. https://www.tensorflow.org/api_docs/python/tf/nn/weighted_cross_entropy_with_logits. Accessed September 7, 2020.

        • Goodfellow I
        • Bengio Y
        • Courville A.
        Deep Learning.
        MIT press, Cambridge, MA2016
        • Davis J
        • Goadrich M.
        The relationship between precision-recall and ROC curves.
        in: Proceedings of the 23rd International Conference on Machine learning. New York, NY: Association for Computing Machinery, Pittsburgh, PA2006: 233-240
        • Lentle BC
        • Berger C
        • Probyn L
        • et al.
        Comparative analysis of the radiology of osteoporotic vertebral fractures in women and men: cross-sectional and longitudinal observations from the Canadian Multicentre Osteoporosis study (CaMos).
        J Bone Miner Res. 2018; 33: 569-579
        • He K
        • Zhang X
        • Ren S
        • et al.
        Deep residual learning for image recognition.
        in: Proceedings of the IEEE conference on computer vision and pattern recognition. IEEE Computer Society, Las Vegas, NV. Washington, D.C.2016: 770-778