Cervical Spine Fracture Detection in CT scans Using InceptionV3, MobileNetV2 and CNN Model

Abstract

Cervical spine fractures are one of the most urgent medical problems to address and the most difficult to diagnose. If not diagnosed quickly and properly, they can result in death and irreversible paralysis. This study uses a two-stage deep learning technique to automate the detection of fractures in the cervical spine using CT images. In the first stage, a Global Context Vision Transformer (GC ViT) model is used to locate the cervical vertebrae in the various CT slices. In the second stage, various deep learning classification models such as InceptionV3, Attention-based CNN, and MobileNetV2 are tested for diagnostic accuracy. MobileNetV2 also surpassed all other models, achieving the greatest test accuracy at 85%, as compared to InceptionV3's 74%, Attention-based CNN's 84%, and MobileNetV2's 85%. This denotes the extensive clinical significance of the developed pipeline in, accuracy of diagnosis, compression of interpretation, and assisting the rapid decision-making for patients with cervical spine fractures.