Osteoporotic spine fractures significantly impact adults over the age of 50, underscoring the need for advanced diagnostic tools to enhance their quality of life. While Convolutional Neural Networks (CNNs) have established their effectiveness in diagnosing such conditions, this study explores a novel approach by integrating quantum computing elements. We present a hybrid quantum convolutional neural network based on the ResNet18 architecture, tailored for multiclass classification of spine lesions from the VinDr-SpineXR dataset. This model incorporates a quantum circuit that mimics a classical neural network layer, enhancing feature extraction capabilities. In the training phase, the hybrid quantum model records accuracy, recall, precision, and F1 scores of 0.8274, 0.8309, 0.8262, and 0.8262, respectively, which represent decreases of 9.65%, 9.35%, 9.75%, and 9.75% compared to the classical model’s performance of 0.9158, 0.9166, 0.9155, and 0.9155. However, on the test set, the hybrid model demonstrates a significant improvement, with increases in accuracy and recall by 3.13% each, and in F1 score by 2.20%. These results underscore the model’s enhanced ability to generalize from small datasets and its potential to refine feature extraction processes for medical imaging, showcasing its advantages for practical applications where generalization is critical.