Background: Back pain is the most common cause of long-term disability world-wide. The basis of most back pain diagnoses is that the pain arises from degenerative changes in a specific structure of the spine. Degenerative changes are currently mainly identified by MRI; indeed MRI is the only NICE-approved imaging technique for back pain diagnosis. However, grading of the degree of degeneration of spinal units manually by radiologists is time-consuming and subject to inter- and intra-observer variation.

About This Demo: This is a demonstration of the Oxford SpineNet software [1], a machine learning based system for the automated analysis of spinal MR images to assist in clinical and algorithmic research. The system can extract a wide range of relevant measurements from MR images automatically including Pffirrmann grades, Modic changes, spinal stenosis and disc herniation. The system has been trained and validated on approximately 2000 patients from the Genodisc [5] consortium project and data from the TwinsUK dataset. This research work has resulted in numerous publications (see [1], [2], [3]) and in 2017 won the ISSLS prize in bioengineering science [4] and is currently the most robust and validated automated spinal MRI software available. The system is flexible and can support multiple grading systems in parallel allowing, for the first time, comparative studies between cohorts of data and aggregation of datasets from multiple centres. Please contact us if you'd like us to support your preferred spinal MRI grading system or wish to collaborate with us on related topics.

 

Go to Demo

 

Examples of SpineNet Radiological Gradings
For a given scan, we are able to produce several radiological gradings. Each example is a slice of a disc volume and in the case of Pfirrmann Grading and Disc Narrowing we show examples corresponding to each grade.
1. Pfirrmann Grading (Grade 1 to 5): Examples
2. Disc Narrowing (Grade 1 to 4): Examples
3. Endplate Defects (Binary): Examples
4. Marrow Changes (Binary): Examples
5. Central Canal Stenosis (Binary): Examples
6. Spondylolisthesis (Binary): Examples
More Information
More details about this work is available at http://www.robots.ox.ac.uk/~vgg/research/spine/
For any queries or to add your own grading system, contact amirj@robots.ox.ac.uk
References
[1] A. Jamaludin, M. Lootus, T. Kadir, A. Zisserman, J. Urban, M. C. Battié, J. Fairbank, and I. McCall. Issls prize in bioengineering science 2017: Automation of reading of radiological features from magnetic resonance images (mris) of the lumbar spine without human intervention is comparable with an expert radiologist. European Spine Journal, 26(5):1374–1383, May 2017.
[2] A. Jamaludin, T. Kadir, and A. Zisserman. SpineNet: Automatically Pinpointing Classification Evidence in Spinal MRIs, pages 166–175. Springer International Publishing, Cham, 2016.
[3] M. Lootus, T. Kadir, and A. Zisserman. Radiological grading of spinal MRI. In MICCAI Workshop: Computational Methods and Clinical Applications for Spine Imaging, 2014.
[4] A. Jamaludin, T. Kadir, and A. Zisserman. SpineNet: Automated classifica- tion and evidence visualization in spinal MRIs. Medical Image Analysis, 41 (Supplement C):63 – 73, 2017.
[5] http://www.physiol.ox.ac.uk/genodisc/