Deep Learning-Based Osteoporosis Screening from Lumbar Spine X-ray Images: A CNN Approach Compared with DXA

Md. Nesar; Kamruzzaman Mithu; Md. Ataur; Shahanara Begum; Khondaker Abdullah Al

doi:10.25163/data.6110745

Data Modeling

Mathematical and Computational Data Modeling

Citations

1.5k

Views

Articles

Submit

Volume 6 Number 1 2025

Figures and Tables

RESEARCH ARTICLE (Open Access)

Previous Contents Vol 6 (1)

Deep Learning-Based Osteoporosis Screening from Lumbar Spine X-ray Images: A CNN Approach Compared with DXA

Md. Nesar Uddin¹*, Kamruzzaman Mithu ¹, Md. Ataur Rahman¹, Shahanara Begum ², Khondaker Abdullah Al Mamun¹

+ Author Affiliations

Data Modeling 6 (1) 1-8 https://doi.org/10.25163/data.6110745

Submitted: 31 December 2024 Revised: 28 February 2025 Published: 08 March 2025

Abstract

Osteoporosis, a condition often progressing silently until fracture occurs, continues to pose a substantial clinical and public health burden worldwide. While dual-energy X-ray absorptiometry (DXA) remains the gold standard for diagnosis, its limited accessibility and underutilization raise important questions about alternative screening strategies. In this context, the present study explores whether deep learning—specifically convolutional neural networks (CNNs)—might offer a complementary approach using routinely acquired lumbar spine X-ray images. A retrospective multicenter dataset comprising 1,255 postmenopausal women and 2,510 radiographic images was analyzed. Regions of interest were manually identified, and a dual-channel CNN model was developed using anteroposterior and lateral views. Model predictions were evaluated against DXA-derived bone mineral density classifications based on World Health Organization criteria. Performance was assessed using area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. The model demonstrated moderate diagnostic capability. In the validation dataset, osteoporosis detection achieved an AUC of 0.786, with sensitivity of 60.6% and specificity of 85.5%. In independent test datasets, AUC values ranged from 0.726 to 0.810, with variability observed across imaging channels and classification categories. Notably, sensitivity for osteopenia detection improved in certain configurations, suggesting differential feature representation across views. Although these findings do not yet support clinical replacement of DXA, they suggest that deep learning applied to routine radiographs may provide a feasible adjunct screening tool. Further refinement, validation, and integration with clinical risk factors will be essential before such approaches can be meaningfully translated into practice.

Keywords: Osteoporosis, Deep Learning, Convolutional Neural Network, Lumbar Spine X-ray, Bone Mineral Density.

References

Amarnath, A. L., Franks, P., Robbins, J. A., Xing, G., & Fenton, J. J. (2015). Underuse and overuse of osteoporosis screening in a regional health system: A retrospective cohort study. Journal of General Internal Medicine, 30(12), 1733–1740.

Brown, C. (2017). Osteoporosis: Staying strong. Nature, 550, S15–S17.

Centers for Medicare & Medicaid Services. (2019). National physician fee schedule. Retrieved April 17, 2019, from http://www.cms.hhs.gov/PFSlookup/

Cheung, A. M., Papaioannou, A., Morin, S., & Osteoporosis Canada Scientific Advisory Council. (2016). Postmenopausal osteoporosis. New England Journal of Medicine, 374, 2096.

Curry, S. J., Krist, A. H., Owens, D. K., Barry, M. J., Caughey, A. B., Davidson, K. W., Doubeni, C. A., Epling, J. W., Kemper, A. R., Kubik, M., Landefeld, C. S., Mangione, C. M., Silverstein, M., Simon, M. A., Tseng, C. W., & Wong, J. B. (2018). Screening for osteoporosis to prevent fractures: U.S. Preventive Services Task Force recommendation statement. JAMA, 319(24), 2521–2531.

Dimai, H. P. (2017). Use of dual-energy X-ray absorptiometry (DXA) for diagnosis and fracture risk assessment: WHO criteria, T- and Z-score, and reference databases. Bone, 104, 39–43.

Erhan, D., Bengio, Y., Courville, A., Manzagol, P.-A., Vincent, P., & Bengio, S. (2010). Why does unsupervised pre-training help deep learning? Journal of Machine Learning Research, 11, 625–660.

Esteva, A., Robicquet, A., Ramsundar, B., Kuleshov, V., DePristo, M., Chou, K., Cui, C., Corrado, G., Thrun, S., & Dean, J. (2019). A guide to deep learning in healthcare. Nature Medicine, 25, 24–29.

Ferizi, U., Honig, S., & Chang, G. (2019). Artificial intelligence, osteoporosis and fragility fractures. Current Opinion in Rheumatology, 31(4), 368–375.

King, A. B., & Fiorentino, D. M. (2011). Medicare payment cuts for osteoporosis testing reduced use despite tests’ benefit in reducing fractures. Health Affairs, 30(12), 2362–2370.

LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521, 436–444.

Lever, J., Krzywinski, M., & Altman, N. (2016). Model selection and overfitting. Nature Methods, 13(9), 703–704.

National Health Commission of China. (2019). Epidemiological investigation of osteoporosis in China. Retrieved May 15, 2019, from http://www.phsciencedata.cn/Share/jsp/PublishManager/foregroundView/1/9eaadbb3-bd64-4531-9d9f-753ec183f26d.html

Schuit, S. C., van der Klift, M., Weel, A. E., de Laet, C. E., Burger, H., Seeman, E., Hofman, A., Uitterlinden, A. G., & Pols, H. A. (2004). Fracture incidence and association with bone mineral density in elderly men and women: The Rotterdam Study. Bone, 34(1), 195–202.

U.S. Preventive Services Task Force. (2002). Screening for osteoporosis in postmenopausal women: Recommendations and rationale. American Family Physician, 66(8), 1430–1432.

Wainwright, S. A., Marshall, L. M., Ensrud, K. E., Cauley, J. A., Black, D. M., Hillier, T. A., Hochberg, M. C., Vogt, M. T., Orwoll, E. S., & Study of Osteoporotic Fractures Research Group. (2005). Hip fracture in women without osteoporosis. Journal of Clinical Endocrinology & Metabolism, 90(5), 2787–2793.

Data Modeling

Article Contents

Deep Learning-Based Osteoporosis Screening from Lumbar Spine X-ray Images: A CNN Approach Compared with DXA

Abstract

References

Recommended articles

Stay connected