Year: 2025
Type: article
Abstract: Cervical spine fractures are critical medical emergencies with the possible to cause permanent paralysis or death if not promptly identified and treated. This study addresses the gap in leveraging deep learning models for their detection by proposing a two-stage pipeline using a curated dataset of computed tomography (CT) images comprising both fractured and non-fractured cases. In the first stage, cervical vertebrae are identified within CT image slices using a multi-input network based on the Global Context Vision Transformer (GC ViT) architecture, benchmarked against leading Deep Learning models. During the second phase, multiple architectures such as Convolutional Neural Networks (CNNs) were employed to perform the designated classification tasks (CNN), ResNet, EnlightenmentNet, and a hybrid CNN+ResNet architecture, are employed to detect fractures, achieving accuracy of 64%, 70%, 54%, and 94%, respectively, with further comparisons against YOLOv5 for localization performance. The hybrid CNN+ResNet model demonstrated superior accuracy, significantly reducing radiologists' workload and enhancing diagnostic precision and efficiency. This approach has substantial clinical implications, offering a promising solution to improve patient outcomes through timely and reliable detection of cervical spine fractures
Authors: Santhosh T, K. T. Veeramanju
Institution: Srinivas University
Language: English
Topic: Medical Imaging and Analysis
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7121546232