Oladimeji, Oladosu Oyebisi and Ibitoye, Ayodeji ORCID: https://orcid.org/0000-0002-5631-8507 (2026) MSA-CSpineNet: a multi-scale spatial attention Deep Learning framework for cervical spine fracture diagnosis. Artificial Intelligence and Application. pp. 1-9. ISSN 2811-0854 (Online) (doi:10.47852/bonviewAIA62027209)

Preview

PDF (Open Access Article) 53377 IBITOYE_MSA-CSpineNet_A_Multi-Scale_Spatial_Attention_Deep_Learning_Framework_(OA)_2026.pdf - Published Version Available under License Creative Commons Attribution. Download (1MB) | Preview

Abstract

Cervical spine fractures can cause instability in the cervical spine and may result in spinal cord injuries. If not promptly detected and treated, these fractures may deteriorate over time. Hence, the diagnosis of cervical spine injuries must be conducted urgently to prevent further complications. Current deep learning models face limitations in accurately diagnosing these fractures due to issues such as insufficient attention to subtle fracture features and poor generalization across varying scales. This research proposes MSA-CSpineNet (Multi-Scale Spatial Attention Cervical Spine Network), a deep learning framework for accurate cervical spine fracture diagnosis from computed tomography scans. The pre-trained MobileNet was used for feature extraction, which was passed to the multiscale and attention module for relevant feature selection. The results show an accuracy of 99.75%, sensitivity of 99.99%, specificity of 99.50%, and precision of 99.50%. Compared with the existing state-of-the-art approaches that used transfer learning and conventional convolutional neural network techniques, experimental results demonstrated that the proposed MSA-CSpineNet outperforms existing methods in image classification. The results of this research have the potential to greatly improve cervical spine fracture early diagnosis and treatment, which would benefit patients' outcomes. Gradient-weighted class activation mapping visualization demonstrates that the model develops spatially selective attention patterns, providing interpretability that supports clinical trust in model predictions.

Item Type:	Article
Uncontrolled Keywords:	cervical spine fracture, attention mechanism, fracture, multi-scale, computed tomography
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science R Medicine > R Medicine (General)
Faculty / School / Research Centre / Research Group:	Faculty of Engineering & Science Faculty of Engineering & Science > School of Computing & Mathematical Sciences (CMS)
Last Modified:	11 May 2026 15:23
URI:	https://gala.gre.ac.uk/id/eprint/53377

Actions (login required)

View Item

Downloads