Fully Automatic Deep Learning Approaches for Cervical Spine Fracture Characterisation in CT Images

Bibliographic citation

E. Goyanes, C. Lozano, J. De Moura, J. Novo and M. Ortega, "Fully Automatic Deep Learning Approaches for Cervical Spine Fracture Characterisation in CT Images," 2026 International Joint Conference on Neural Networks (IJCNN), Maastricht, Netherlands, 2026, pp. 1-6, doi: 10.1109/IJCNN69854.2026.11728787

Type of academic work

Academic degree

Abstract

[Abstract] Cervical spine fractures are a critical clinical condition that requires an accurate diagnosis to prevent severe neurological complications. Although Computed Tomography (CT) is the imaging modality of choice for cervical spine evaluation, manual interpretation takes time and is subject to inter-observer variability, motivating the development of robust automated solutions. In this work, we present a fully automatic deep learning pipeline for cervical spine fracture analysis on CT scans, the first in the state-of-the-art comprising three complementary tasks: fracture screening, vertebrae segmentation, and fracture localisation. Fracture screening is formulated as a classification problem to identify CT slices containing fractures. Vertebra segmentation is achieved through a transfer-learning strategy based on brain MRI pretraining, enabling effective transfer of knowledge between modalities to cervical CT. For fracture location, we explore two alternative approaches: coordinate-based regression and prediction of the bounding-box, providing precise spatial characterisation of fracture regions. The proposed pipeline is evaluated on large dataset comprised by 711,601 images from 2019 patients. The experimental results demonstrate strong and consistent performance across all three tasks, supporting accurate and efficient fracture characterisation. Overall, our methodology has the potential to assist radiologists by improving diagnostic accuracy while reducing analysis time and workload.

Description

Presented at: 2026 International Joint Conference on Neural Networks (IJCNN), 22-26 June 2026, Maastricht, Netherlands © 2026 IEEE. This version of the article has been accepted for publication, after peer review. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. The Version of Record is available online at: https://doi.org/10.1109/IJCNN69854.2026.1172878

Rights

Copyright © 2026, IEEE