Evaluation of Publicly Available Deep Learning Lumbar Spine MRI Analysis Tool for Prioritizing MRI Evaluation

Motivation: Lumbar spine MRIs (LSP) are obtained regularly around the world for evaluating back pain. Many of these cases needing urgent care will show high spinal canal narrowing. An automated system that can find studies that have moderate to severe spinal stenosis could prioritize study review. Goal(s): Can the SpineNetV2 deep learning model identify high spinal canal narrowing as well as trained neuroradiologists, and how do they complement each other. Approach: SpineNetV2 was run on a public LSP dataset, and was compared with human readers scores. Results: SpineNetV2 scores showed same moderate agreement with readers, as the readers did with each other. Impact: The SpineNetV2 publicly available deep learning software has potential to serve as a prioritization tool for moderate to severe vertebral canal narrowing, which can aid in screening for patients with potentially critical spinal stenosis.