A hybrid 3D CNN–Transformer architecture for high-precision automated diagnosis of spinal disorders from magnetic resonance imaging volumes
The automated and precise interpretation of volumetric magnetic resonance imaging (MRI) for spinal cord anomalies is crucial for timely intervention and patient care planning. Deep learning in MRI volumetry is hampered by the need to disentangle mixed-resolution volumetric and MRI-slice data. To mitigate these challenges, this study introduces an original approach for automated multi-label classification of spinal MRI abnormalities. The proposed dual-scale spatiotemporal network (DS-STNet) processes volumetric 3D MRI data directly to achieve precise, automatic classification of spinal abnormalities. It combines dual-scale 3D convolutional feature extraction with a global attention mechanism implemented via a Transformer network, capturing both local anatomical features and inter-slice relationships. The model addresses MRI volumetric data in their raw states without computerized segmentation and feature extraction. DS-STNet was evaluated on MRI volumetric data from the Kaggle datasets (RSNA LumbarDISC and SpineWeb), where spinal abnormalities were classified into five categories. Experimental results showed an overall classification accuracy of 98.61%, with high precision, recall, F1 score, and area under the curve. State-of-the-art deep learning competitors, such as 3D convolutional neural networks (CNNs), 3D ResNets, Vision Transformers, and CNN–long short-term memory hybrids, were compared with DS-STNet. The proposed DS-STNet significantly outperformed these existing deep learning models. The findings confirm the effectiveness of incorporating dual-scale spatiotemporal feature learning with Transformer global attention for dependable identification of spinal abnormalities. The proposed DS-STNet offers a scalable, clinically feasible, and high-accuracy solution for automating MRI spinal diagnostics.
Amelot, A., Cretolle, C., de Saint Denis, T., Sarnacki, S., Catala, M., & Zerah, M. (2020). Spinal dysraphism as a new entity in V.A.C.T.E.R.L syndrome, resulting in a novel acronym V.A.C.T.E.R.L. European Journal of Pediatrics, 179(7), 1121–1129. https://doi.org/10.1007/s00431-020-03029-4
Chalif, J. I., Chavarro, V. S., Mensah, E., Johnston, B., Fields, D. P., Chalif, E. J., Chiang, M., Sutton, O., Yong, R., Trumbower, R., & Lu, Y. (2024). Epidural spinal cord stimulation for spinal cord injury in humans: A systematic review. Journal of Clinical Medicine, 13(4), 1090. https://doi.org/10.3390/jcm13041090
Chen, J., Qian, L., Ma, L., Urakov, T., Gu, W., & Liang, L. (2024). SymTC: A symbiotic Transformer-CNN net for instance segmentation of lumbar spine MRI. Computers in Biology and Medicine, 179, 108795. https://doi.org/10.1016/j.compbiomed.2024.108795
Cui, H., Ruan, Z., Xu, Z., Luo, X., Dai, J., & Geng, D. (2024). ResMT: A hybrid CNN–Transformer framework for glioma grading with 3D MRI. Computers & Electrical Engineering, 120, 109745. https://doi.org/10.1016/j.compeleceng.2024.109745
Debska, J., Kotulska-Jozwiak, K., Pawliszak, P., Bilska, M., Chmielewski, D., Puzio, I., & Jurkiewicz, E. (2017). Spinal cord lesions in children and adolescents with multiple sclerosis—Magnetic resonance imaging. Neurologia i Neurochirurgia Polska, 51(1), 77–81. https://doi.org/10.1016/j.pjnns.2016.11.006
Freigang, V., Butz, K., Seebauer, C. T., Karnosky, J., Lang, S., Alt, V., & Baumann, F. (2022). Management and mid-term outcome after "real SCIWORA" in children and adolescents. Global Spine Journal, 12(6), 1208–1213. https://doi.org/10.1177/2192568220979131
Geetha, R., Kumar, E., Vasantharaj, A., & Mohamedyaseen, A. (2025). Hybrid deep learning model for spinal tumor diagnosis on MRI scans. Technology and Health Care. Advance online publication. https://doi.org/10.1177/09287329251385793
Hernandez-Navarro, A., Ros-Alsina, A., Yurtseven, M., Wright, M., & Kumru, H. (2025). Non-invasive cerebral and spinal cord stimulation for motor and gait recovery in incomplete spinal cord injury: Systematic review and meta-analysis. Journal of NeuroEngineering and Rehabilitation, 22(1), 53. https://doi.org/10.1186/s12984-025-01557-4
Horsfield, M. A., Sala, S., Neema, M., Absinta, M., Bakshi, A., Sormani, M. P., Rocca, M. A., Bakshi, R., & Filippi, M. (2010). Rapid semi-automatic segmentation of the spinal cord from magnetic resonance images: Application in multiple sclerosis. NeuroImage, 50(2), 446–455. https://doi.org/10.1016/j.neuroimage.2009.12.080
Kirshblum, S., Snider, B., Rupp, R., & Read, M. S. (2020). Updates of the international standards for neurologic classification of spinal cord injury: 2015 and 2019. Physical Medicine and Rehabilitation Clinics of North America, 31(3), 319–330. https://doi.org/10.1016/j.pmr.2020.03.005
Liang, J., Wang, L., Hao, X., Wang, G., & Wu, X. (2022). Risk factors and prognosis of spinal cord injury without radiological abnormality in children in China. BMC Musculoskeletal Disorders, 23(1), 428. https://doi.org/10.1186/s12891-022-05393-8
Ma, R., Dai, X., Yang, Z., Wei, Z., & Zhang, B. (2025). CAFR-Net: A transformer-contrastive framework for robust spinal MRI segmentation via global-local synergy. PLOS ONE, 20(7), e0327642. https://doi.org/10.1371/journal.pone.0327642
Mann, R. S., Constantinescu, C. S., & Tench, C. R. (2007). Upper cervical spinal cord cross-sectional area in relapsing remitting multiple sclerosis: Application of a new technique for measuring cross-sectional area on magnetic resonance images. Journal of Magnetic Resonance Imaging, 26(1), 61–65. https://doi.org/10.1002/jmri.20959
Margoni, M., Preziosa, P., Rocca, M. A., & Filippi, M. (2022). Pediatric multiple sclerosis: Developments in timely diagnosis and prognostication. Expert Review of Neurotherapeutics, 22(5), 393–403. https://doi.org/10.1080/14737175.2022.2064743
Margoni, M., Storelli, L., Pagani, E., Preziosa, P., Mistri, D., Gueye, M., Rubin, M., Moiola, L., Filippi, M., & Rocca, M. A. (2025). Subventricular zone microstructure in pediatric-onset multiple sclerosis. Annals of Neurology, 97(5), 979–992. https://doi.org/10.1002/ana.27180
Mezzapesa, D. M., Rocca, M. A., Falini, A., Rodegher, M. E., Ghezzi, A., Comi, G., & Filippi, M. (2004). A preliminary diffusion tensor and magnetization transfer magnetic resonance imaging study of early-onset multiple sclerosis. Archives of Neurology, 61(3), 366–368. https://doi.org/10.1001/archneur.61.3.366
Rocca, M. A., Valsasina, P., Meani, A., Gobbi, C., Zecca, C., Barkhof, F., Schoonheim, M. M., Strijbis, E. M., Vrenken, H., Gallo, A., Bisecco, A., Ciccarelli, O., Yiannakas, M., Rovira, A., Sastre-Garriga, J., Palace, J., Matthews, L., Gass, A., Eisele, P., … Filippi, M. (2023). Spinal cord lesions and brain grey matter atrophy independently predict clinical worsening in definite multiple sclerosis: A 5-year, multicentre study. Journal of Neurology, Neurosurgery & Psychiatry, 94(1), 10–18. https://doi.org/10.1136/jnnp-2022-329854
Seijas, V., Schrepfer, L., Posada, A. M., Spir, M. A., Machado, B., Sigrist-Nix, D., Scheel-Sailer, A., Eriks-Hoogland, I., & Sabariego, C. (2024). Evidence-based recommendations for the rehabilitation and management of the ageing population with spinal cord injury: A systematic review of clinical practice guidelines. European Journal of Physical and Rehabilitation Medicine, 60(3), 433–444. https://doi.org/10.23736/S1973-9087.24.08244-3
Sergeenko, O. M., Savin, D. M., & Diachkov, K. A. (2024). Association of spinal cord abnormalities with vertebral anomalies: An embryological perspective. Child's Nervous System, 40(5), 1415–1425. https://doi.org/10.1007/s00381-024-06336-5
Shinozaki, M., Nagoshi, N., Nakamura, M., & Okano, H. (2021). Mechanisms of stem cell therapy in spinal cord injuries. Cells, 10(10), 2676. https://doi.org/10.3390/cells10102676
Suleman, S., Harwood, M., Kang, P., Narayan, R., & Hughes, J. N. (2022). Assessment of cervical spinal cord volume in pediatric-onset multiple sclerosis. Clinical Imaging, 90, 44–49. https://doi.org/10.1016/j.clinimag.2022.06.021
Thompson, A. J., Banwell, B. L., Barkhof, F., Carroll, W. M., Coetzee, T., Comi, G., Correale, J., Fazekas, F., Filippi, M., Freedman, M. S., Fujihara, K., Galetta, S. L., Hartung, H. P., Kappos, L., Lublin, F. D., Marrie, R. A., Miller, A. E., Miller, D. H., Montalban, X., … Cohen, J. A. (2018). Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. The Lancet Neurology, 17(2), 162–173. https://doi.org/10.1016/S1474-4422(17)30470-2
Valsasina, P., Aboulwafa, M., Preziosa, P., Messina, R., Falini, A., Comi, G., Filippi, M., & Rocca, M. A. (2018). Cervical cord T1-weighted hypointense lesions at MR imaging in multiple sclerosis: Relationship to cord atrophy and disability. Radiology, 288(1), 234–244. https://doi.org/10.1148/radiol.2018172311
Valsasina, P., Horsfield, M. A., Rocca, M. A., Absinta, M., Comi, G., & Filippi, M. (2012). Spatial normalization and regional assessment of cord atrophy: Voxel-based analysis of cervical cord 3D T1-weighted images. American Journal of Neuroradiology, 33(11), 2195–2200. https://doi.org/10.3174/AJNR.A3139
Valverde, S., Cabezas, M., Roura, E., González-Villà, S., Pareto, D., Vilanova, J. C., Ramió-Torrentà, L., Rovira, A., Oliver, A., & Lladó, X. (2017). Improving automated multiple sclerosis lesion segmentation with a cascaded 3D convolutional neural network approach. NeuroImage, 155, 159–168. https://doi.org/10.1016/j.neuroimage.2017.04.034
Verhey, L. H., Branson, H. M., Makhija, M., Shroff, M., & Banwell, B. (2010). Magnetic resonance imaging features of the spinal cord in pediatric multiple sclerosis: A preliminary study. Neuroradiology, 52(12), 1153–1162. https://doi.org/10.1007/s00234-010-0755-9
Wang, Y., Zeng, L., Zhu, F., Huang, G., Wan, Y., Yao, S., Chen, K., & Guo, X. (2022). Acute hyperextension myelopathy in children: Radiographic predictors of clinical improvement. Spinal Cord, 60(6), 498–503. https://doi.org/10.1038/s41393-021-00755-7
Weeda, M. M., Middelkoop, S. M., Steenwijk, M. D., Daams, M., Amiri, H., Brouwer, I., Killestein, J., Uitdehaag, B. M. J., Dekker, I., Lukas, C., Bellenberg, B., Barkhof, F., Pouwels, P. J. W., & Vrenken, H. (2019). Validation of mean upper cervical cord area (MUCCA) measurement techniques in multiple sclerosis (MS): High reproducibility and robustness to lesions, but large software and scanner effects. NeuroImage: Clinical, 24, 101962. https://doi.org/10.1016/j.nicl.2019.101962