SpinalAI: A Deep Learning Approach to Predict Vertebrae-Column Level, Structure, and Foraminal on Cervical Spine Axial MRI Images

Spinal cord injuries are a significant cause of disability worldwide. Accurate diagnosis and treatment of these injuries are critical to prevent further damage and improve patient outcomes. This paper presents SpinalAI, an intelligent system that utilizes deep learning techniques to predict the vertebrae column level, structure, and foraminal detection on axial MRI images of the cervical spinal cord. SpinalAI incorporates a trio of models: a vertebral column level assessment (a regression-based or a classification-based), one for structural analysis of the cervical spinal cord, and another dedicated to foraminal detection. The models undergo fine-tuning utilizing a pre-trained model trained on RadImageNet a comprehensive dataset of radiological images or ImageNet, a vast dataset of natural images. The system performs well on a dataset of 1862 axial MRI images from two centers, demonstrating its potential to enhance the efficiency and objectivity of interpreting cervical spine MRI scans. The system achieved a mean absolute error (MAE) of 0.459 and an F1 score of $89 \%$ in predicting the level of the vertebrae column. It also detected the cervical spinal cord structure and the foraminal with $95 \%$ and $96 \%$ accuracy, respectively.