Classification of Lumbar Spine Degeneration Using Vision Transformer with the RSNA Dataset

Abstract This study presents a deep-learning approach for classifying lumbar spine degeneration using the RSNA dataset by employing a Vision Transformer model. The classification task was divided into three categories normal/mild, moderate, and severe. The dataset comprises approximately 2000 MRI studies annotated by spine radiology specialists to indicate the presence, vertebral level, and location of any lumbar spinal stenosis. The methodology involves preprocessing the MRI and training a Vision Transformer to classify degeneration severity. The Vision Transformer model was chosen for its ability to capture global image dependencies and superior performance in medical image analysis tasks. The quantitative results demonstrate the effectiveness of our approach, with an accuracy of 93.6%, a precision of 92.3%, and a recall of 89.8%. The proposed method achieves high classification performance and aids in the early detection and treatment planning for lumbar spine degeneration. The significance of this research lies in its potential to automate and enhance the diagnostic process, providing reliable and rapid assessments that can support clinical decision-making and improve patient outcomes.