A cervical spine fracture is a break or crack in one or more of the seven vertebrae located in the neck region. Cervical spine fracture is a significant issue in medical science and is the foremost reason for death in trauma patients. Hence, accurate treatment and early detection of spine cracks can recover the patient’s results. Advanced artificial intelligence technology, such as convolutional neural networks (CNNs), can detect and classify cervical spine fractures at an early stage. The optimization in building a CNN model plays a substantial role in the training process. In this study, selected optimizers such as RMSprop, Adam, SGD, and Adadelta in the proposed CNN model require lower hardware requirements and shorter training times. This paper proposes a binary classification of cervical spine fracture using the scratch CNN model, which classifies the standard and fracture images from the computed tomography (CT) image dataset. Also, the performance measures and classification reports are compared using different optimizations. RMSprop optimization gives the best result compared to other optimizations. This work finds that the proposed CNN model is effective and achieves optimal accuracy compared to other authors’ work. This paper concludes that different optimizers play a significant role in classifying and detecting cervical spine fractures.