Cauda equina syndrome occurs when the nerves at the base of the spine are suddenly compressed which may result in bladder dysfunction, bowel dysfunction, and/or sensory changes in the saddle area. The condition can require emergency surgery: delayed treatment can result in long-term irreversible life-changing nerve damage.
Patients can present to their GP with non-specific symptoms but have significant and unexpected critical findings on their MRI identified at the point of reporting. Unfortunately, there is an ongoing significant backlog of radiology reporting, meaning that many patients referred from the community for MRI will not have their scan reviewed by a radiologist immediately.
A computer vision model that can correctly categorise MRI images would enable prioritised reporting of MRI scans with potential nerve compression. Provisional assessment of nerve compression with a computer vision model may provide improved consistency and productivity for reporting. We are undertaking this project to investigate the feasibility of this approach.
In partnership with The University of Edinburgh and British Neurosurgical Trainee Research Collective, our group have obtained ethical approval and agreed access to a shared curated research dataset of abnormal MRI scans.
We have created a labelling platform based on the Visual Geometry Group's VIA tool. This has enabled medically-qualified collaborators to label thousands of MRI images of the spine.
As well as developing machine learning models locally on our internal dataset, we have also joined the Kaggle RSNA 2024 Lumbar Spine competition achieving position 29th out of 1874 teams.
We are currently applying for competitively awarded funding to validate high performance models still in the research and development phase on clinical data.
If the model performs well at initial validation, the feasibility of obtaining certification for clinical use will be assessed.
Opportunities to contribute to this and other similar projects will be announced on the SUSTAINSW news page.