Convolutional Neural Networks in the Diagnosis of Cervical Myelopathy

Objective  Artificial intelligence technologies have been used increasingly in spine surgery as a diagnostic tool. The aim of the present study was to evaluate the effectiveness of the convolutional neural networks in the diagnosis of cervical myelopathy (CM) compared with conventional cervical magnetic resonance imaging (MRI). Materials and Methods  This was a cross-sectional descriptive analytical study. A total of 125 participants with clinical and radiological diagnosis of CM were included in the study. Sagittal and axial MRI images in the T2 sequence of the cervical spine were used. All image parts were obtained as 8 bytes/pixel in 2 different categories, CM and normal, both in axial and sagittal views. Results  Triple cross validation was performed to prevent overfitting during the training process. A total of 242 sample images were used for training and testing the model created for axial views. In the axial view, the calculated values are 97.44% for sensitivity and 97.56% for specificity. A total of 249 sample images were used for training and testing the model created for sagittal views. The calculated values are 97.50% for sensitivity and 97.67% for specificity. After the training, the average accuracy value was 96.7% (±1.53) for the axial view and 97.19% (±1.2) for the sagittal view. Conclusion  Deep learning (DL) has shown a great improvement especially in spine surgery. We found that DL technology works with a higher accuracy than other studies in the literature for the diagnosis of CM.