Spine fractures pose a serious health risk, and timely detection is critical for preventing long-term complications. This paper introduces a new convolutional neural network (CNN) architecture to detect cervical spine fractures from computed tomography (CT) images. The proposed model incorporates depthwise separable convolutions, residual connections, and a multi-scale feature extraction block, allowing for high accuracy while maintaining low computational complexity. With only 287,723 parameters and a computational cost of 0.57 GFLOPs, the model achieves an accuracy of 99.98%, outperforming several existing state-of-the-art methods. Additionally, an ablation study underscores the impact of each architectural component, confirming their contributions to the model’s overall performance. These results demonstrate the model’s potential for real-time clinical use, offering a fast and accurate tool to assist radiologists in spine fracture detection.