Cervical spine fracture detection utilizing YOLOv8 and deep attention-based vertebrae classification ensuring XAI

Fractures, especially in the cervical spine, pose significant challenges for diagnosis and treatment. As the incidence of these injuries rises and traditional diagnostic methods have limitations, there is an urgent need for more efficient and accurate detection techniques. This study addresses these challenges by proposing a comprehensive approach to automate fracture detection and classification of cervical spine injuries from CT scans. We have combined object detection models and a novel attention mechanism-based Convolutional Neural Network (CNN) titled (VertNet-10) to detect cervical spinal fractures and classify vertebrae from axial plane CT images. Our methodology involves refining the YOLOv8 object detection model and proposing a deep CNN model with attention blocks. Our modified YOLOv8 achieved a Mean Average Precision (mAP) of 93 % in fracture detection, outperforming existing models. The CNN model achieved an accuracy of 99.55 % in classifying cervical vertebrae, with 100 % accuracy for some vertebrae. Furthermore, we successfully generated activation maps to explain the model's classification process, enhancing model explainability. By combining these two modules, our proposed approach offers a promising solution for automating fracture detection and cervical vertebrae classification. Integrating advanced imaging algorithms and attention mechanisms significantly improves diagnostic precision and efficiency. This study emphasizes the potential of AI-driven systems in augmenting radiological diagnosis and ultimately improving patient care in cervical spine injuries.