Оценка эффективности алгоритмов глубокого обучения для выявления признаков spina bifida occulta на рентгенограммах позвоночника: ОРИГИНАЛЬНЫЕ СТАТЬИ

Вероника Евгеньевна  Казаринова; Андрей Сергеевич  Гацук; Кирилл Михайлович  Арзамасов; Юрий Александрович  Васильев

doi:10.56871/

Evaluating the performance of deep learning algorithms for detecting spina bifida occulta on spinal radiographs

ORIGINAL PAPERS

Veronika E. Kazarinova Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department https://orcid.org/0009-0001-3568-8138
Andrey S. Gatsuk Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department https://orcid.org/0009-0000-0549-4204
Kirill M. Arzamasov Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department; MIREA — Russian Technological University; Samara State Medical University https://orcid.org/0000-0001-7786-0349
Yuriy A. Vasilev Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department https://orcid.org/0000-0002-5283-5961

Keywords:

spina bifida occulta spine radiography artificial intelligence spina bifida occulta рентгенография позвоночника искусственный интеллект

Abstract

Introduction. Spina bifida occulta is a latent congenital spinal defect that often remains undiagnosed until neurological symptoms appear. The development of a system for the automated detection of this pathology on radiographs using neural networks requires a sound choice of architecture that ensures optimal diagnostic performance. Aim — to evaluate the performance of deep learning algorithms (Res Net50, Efficient Net-B7, RT-DETR-L, and YOLO versions 8–12) for detecting spina bifida occulta on cervical, thoracic, and lumbosacral spine radiographs. Materials and methods. A retrospective, labeled dataset was compiled from 1,732 spinal radiographs (360 cervical, 141 thoracic, and 1,231 lumbosacral) acquired between 2022 and 2024. The dataset was balanced 50:50 between pathological and normal cases. Automatic cropping of the spinal column region was performed using the Qwen3-VL vision language model. The training subset was expanded threefold using data augmentation techniques. The models were trained and tested using sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC-ROC). Results. For the cervical spine, the best performance was achieved by the YOLOv9t model (AUC=0.926 [95% confidence interval (CI): 0.816–1.000]); for the thoracic spine, by YOLOv10n and YOLOv11n (AUC=0.982 [95% CI: 0.933–1.000]); and for the lumbosacral spine, by YOLOv8m (AUC=0.956 [95% CI: 0.923–0.989]). Res Net50 and Efficient Net-B7 demonstrated significantly lower performance, yielding AUC values not exceeding 0.839. YOLO models enabled both image classification and defect localization on radiographs using bounding boxes. Conclusion. YOLO architectures outperform alternative deep learning models for detecting SBO on radiographs. These metrics support the integration of YOLOv9t, YOLOv10n, YOLOv11n, and YOLOv8m into software for automated detection of Spina Bifida Occulta on cervical, thoracic, and lumbosacral radiographs.

Author Biographies

Veronika E. Kazarinova, Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department

Junior Researcher

Andrey S. Gatsuk, Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department

Junior Researcher

Kirill M. Arzamasov, Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department; MIREA — Russian Technological University; Samara State Medical University

Dr. Sci. (Med.), Head of the Scientific Department, Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department; Associate Professor of the Department of Artificial Intelligence Technologies, MIREA — Russian Technological University; Associate Professor of the Department of Radiation Diagnostics and Radiation Therapy, Samara State Medical University

Yuriy A. Vasilev, Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of Moscow Healthcare Department

Dr. Sci. (Med.), Associate Professor, Medical Director

References

1. Schindelmann K.H., Paschereit F., Steege A., Stoltenburg Didinger G., Kaindl A.M. Systematic с lassification of Spina bifida. J Neuropathol Exp Neurol. 2021.80(4):294–305. https://doi.org/10.1093/jnen/nlab007. EDN: DEOKYD.

2. Sipos T.C., Denes L., Brinzaniuc K., Sipos R.S., Raduly G., Pap Z. Study of Spina bifida occulta based on age, sex and localization. Ars Medica Tomitana. 2019.25(3):95–99. https://doi.org/10.2478/arsm-2019-0020.

3. Заточная В.В. Формирование позвоночного канала в эмбриогенезе человека. Морфологические ведомости. 2018.26(1):33–37. https://doi.org/10.20340/mv_mn.18(26).01.33-37. Edn: Xytinv.

4. Härle A. Incidence of Spina bifida occulta and its importance in orthopaedic diseases. In: Spina bifida — neural tube defects: Basic research, interdisciplinary diagnostics and treatment, results and prognosis. P. Gless, J. Lorber, D. Voth, eds. Berlin, Boston: De Gruyter. 1986. P. 249–256. https://doi.org/10.1515/9783110850307-032.

5. Fidas A., Mac Donald H.L., Elton R.A., Wild S.R., Chisholm G.D., Scott R. Prevalence and patterns of spina bifida occulta in 2707 normal adults. Clin Radiol. 1987.38(5):537–542. https://doi.org/10.1016/s0009-9260(87)80150-2.

6. Eubanks J.D., Cheruvu V.K. Prevalence of sacral spina bifida occulta and its relationship to age, sex, race, and the sacral table angle: an anatomic, osteologic study of three thousand one hundred specimens. Spine. 2009.34(15):1539–1543. https://doi.org/10.1097/BRS.0b013e3181a98560.

7. Bessis D. Cutaneous signs of occult cranial and spinal dysraphism. Ann Dermatol Venereol. 2020.147(8-9):504–519. https://doi.org/10.1016/j.annder.2020.02.011. Edn: Blfvko.

8. Mirzaei S., Khoshkholghsima M., Sabaghzadeh A., Kurdkandi H.Z. Cervicothoracic (C6, C7 & T1) spina bifida occulta — A case report. Int J Surg Case Rep. 2024.117:109477. https://doi.org/10.1016/j.ijscr.2024.109477. Edn: Djapba.

9. Sakti Y.M., Lanodiyu Z.A., Sakadewa G.P., Cein C.R., Baskara A.A.N.N., Rahayu B.F.P. Spina bifida occulta at the lumbar spine level manifested as chronic low back pain and unpredictable neurologic deficit: A case report. Int J Surg Case Rep. 2024.116:109320. https://doi.org/10.1016/j.ijscr.2024.109320. Edn: Nmepsy.

10. Goto T., Sakai T., Sato N., Katoh S., Sairyo K. An adolescent athlete with low back pain associated with Spina bifida occulta at the thoracolumbar junction: A case report. J Med Invest. 2019.66(1.2):199–200. https://doi.org/10.2152/jmi.66.199.

11. Taskaynatan M.A., Izci Y., Ozgul A., Hazneci B., Dursun H., Kalyon T.A. Clinical significance of congenital lumbosacral malformations in young male population with prolonged low back pain. Spine. 2005.30(8):E210_E213. https://doi.org/10.1097/01.brs.0000158950.84470.2a.

12. Muranaka I., Kyoda Y., Nofuji S. et al. A case of neuropathic lower urinary tract dysfunction due to Spina bifida occulta discovered at the age of 19 years and successfully treated with multidisciplinary therapy. Hinyokika Kiyo. 2024.70(2):55–59. https://doi.org/10.14989/Acta Urol Jap_70_2_55.

13. Васильев Ю.А., Владзимирский А.В., Арзамасов К.М. и др. Искусственный интеллект в лучевой диагностике: Per Aspera Ad Astra. М.: Нпкц Ди Т Дзм. 2025. 493 с. Edn: Oowutv.

14. Свидетельство о государственной регистрации базы данных № 2025622039 Российская Федерация. Mos Med Data: рентгенография шейного и грудного отдела позвоночника с наличием или отсутствием Spina Bifida: заявл. 11.04.2025: опубл. 12.05.2025. Ю.А. Васильев, А.В. Владзимирский, О.В. Омелянская и др.. заявитель Гбуз «Нпкц Ди Т Дзм».

15. Свидетельство о государственной регистрации базы данных № 2024625146 Российская Федерация. Mos Med Data: рентгенография пояснично крестцового отдела позвоночника с наличием или отсутствием Spina Bifida у взрослых: № 2024624865: заявл. 31.10.2024: опубл. 13.11.2024. Ю.А. Васильев, А.В. Владзимирский, О.В. Омелянская и др.. заявитель Гбуз «Нпкц Ди Т Дзм».

16. Васильев Ю.А., Владзимирский А.В., Омелянская О.В. и др. Автоматизированная платформа для формирования наборов данных в лучевой диагностике. Менеджер здравоохранения. 2025.(7):52–67. https://doi.org/10.21045/1811-0185-2025-7-52-67. Edn: Yavsya.

17. Kirpichev Yu., Vasilev Yu., Vladzymyrskyy A. et al. Free web tool for comprehensive Roc analysis. Biological Communications. 2025.70(3):215–222. https://doi.org/10.21638/spbu03.2025.305. Edn: Rtovsr.

18. Васильев Ю.А., Владзимирский А.В., Арзамасов К.М. и др. Внедрение медицинских изделий с технологиями искусственного интеллекта в лучевую диагностику: методические рекомендации. Ч. 1. Сценарии и оценка эффективности. М.: Нпкц Ди Т Дзм. 2025. 32 с. Edn: Mkjyjc.

19. Asha R., Subashka Ramesh S.S. An optimized modified faster region convolutional neural network for Spina bifida identification from ultrasound images. Biomed Signal Process Control. 2023.86:105253. https://doi.org/10.1016/j.bspc.2023.105253. Edn: Kzvsbl.

20. Asha R., Subashka Ramesh S.S. Segmentation and classification of fetal Spina bifida using Ded with Fm2dcn. Multimed Tools Appl. 2024.83(25):66981–66999. https://doi.org/10.1007/s11042-024-18160-9. Edn: Rkukqi.

21. Omopintemi A.H., David A.O., Adeyemi O.A. Machine learning algorithm for classification of infant congenital anomaly. In: Intelligent Sustainable Systems (World S4 2024). Lecture notes in networks and systems. Vol. 1178. A. Nagar, D.S. Jat, D. Mishra, A. Joshi, eds. Springer, Singapore. 2025. P. 265–276. https://doi.org/10.1007/978-981-97-9559-8_23.

22. Chen L., Tian Y., Deng Y. Neural network algorithm‐based three‐dimensional ultrasound evaluation in the diagnosis of fetal Spina bifida. Scientific Programming. 2021.1:3605739. https://doi.org/10.1155/2021/3605739. Edn: Lviujs.

23. Duran S., Üreten K., Maraş Y. et al. Automatic detection of Spina bifida occulta with deep learning methods from plain pelvic radiographs. Res Biomed Eng. 2023.39(3):655–661. https://doi.org/10.1007/s42600-023-00296-6. Edn: Eezhwm.

Evaluating the performance of deep learning algorithms for detecting spina bifida occulta on spinal radiographs: ORIGINAL PAPERS. (2026). Visualization in Medicine, 8(2), 6-16. https://doi.org/10.56871/

Download Citation

How to Cite

Evaluating the performance of deep learning algorithms for detecting spina bifida occulta on spinal radiographs: ORIGINAL PAPERS. (2026). Visualization in Medicine, 8(2), 6-16. https://doi.org/10.56871/

Download Citation