The RSNA has published what it calls the Lumbar Degenerative Imaging Spine Classification (LumbarDISC) dataset, the largest publicly available adult MRI lumbar spine dataset for degenerative disease, according to the society.
The set includes multisequence, multiplanar MRIs from 2,697 patients (8,593 images) from eight institutions across six countries and five continents, wrote a team led by Tyler Richards, MD, of the University of Utah School of Medicine in Salt Lake City. It was created via a collaboration among the RSNA, the American Society of Neuroradiology (ASN), and the American Society of Spine Radiology (ASSR) as part of the RSNA 2024 Lumbar Spine Degenerative Classification AI Challenge, and was published January 14 in Radiology: Artificial Intelligence.
Low back pain is a significant health concern around the world, the team explained. In the U.S., direct and indirect costs of low back pain are estimated at $50 billion to $100 billion, respectively. The condition can be caused by a narrowing of the spinal canal, neural foramina, and subarticular recesses, all of which can compress spinal nerves — and may lead to spinal surgery.
MRI is the preferred imaging modality to evaluate the degree of narrowing a patient is experiencing, but interrater agreement regarding the severity of this condition varies widely, Richards and colleagues noted.
“Although the large number of lumbar spine MRIs represents a substantial workload for radiologists, the more critical challenge lies in ensuring consistent and accurate grading of stenosis severity, which impacts diagnostic confidence and potentially patient care,” they wrote.
For the dataset, team acquired imaging and demographic data from each contributing site, which included the following:
- Chiang Mai University, Thailand
- Clinical Center of University of Sarajevo, Bosnia and Herzegovina
- Gold Coast University Hospital, Gold Coast, Queensland, Australia
- Koc University School of Medicine, Istanbul, Turkey
- Thomas Jefferson University, Philadelphia, PA
- Diagnósticos da América S.A. (DASA), São Paulo, Brazil
- University of California San Francisco
- University of Utah, Salt Lake City
Inclusion criteria consisted of MRI lumbar spine studies with a sagittal “T2-like” (e.g., conventional spin echo T2, STIR, or Dixon), sagittal T1, and axial T2 weighted images. Volunteer annotators graded the degree of stenosis using a four-point scale of normal, mild, moderate, or severe for the following locations: spinal canal (labeled on the T2-like sequence), right and left neural foraminal (sagittal T1), and right and left subarticular recess (axial T2).
From the MRI images included in the work, the group found the following:
|
Results from assessment of multisequence, multiplanar MRIs for LumbarDISC dataset |
|||
|
Type of finding |
Normal/mild |
Moderate |
Severe |
| Spinal canal stenosis grades |
85.4% |
8.8% |
5.9% |
|
Neural foraminal grades |
|||
| Right |
78.3% |
17.3% |
4.4% |
| Left |
77.2% |
18.1% |
4.6% |
|
Subarticular grades |
|||
| Right |
69.4% |
19.6% |
10.9% |
| Left |
69.5% |
19.2% |
11.3% |
(A, B) MR images (A, sagittal T2; B, sagittal STIR) demonstrate the location of the localizers within the middle of the thecal sac at the level of the L1/L2 through L5/S1. Images and caption courtesy of the RSNA.
“This rich dataset has further potential utility for future investigators, including evaluation of intervertebral disc and vertebral endplate degenerative changes, which were beyond the scope of our competition,” the authors concluded.