ISSLS Prize in Bioengineering Science 2026: Hidden in Plain Sight: Machine Learning-Assisted MRI Reveals Novel
Jeffrey C Y Ng1, Nathan Armani2, Xijun Sun1
1Department of Rehabilitation Sciences, The Hong Kong Polytechnic University, Hong Kong, China.
Purpose:
To identify signal and spatial biomarkers in the lumbar vertebrae of individuals with and without non-specific chronic low back pain (NSCLBP) using magnetic resonance imaging (MRI).
Methods:
This case-control study recruited 68 individuals with NSCLBP and 62 age- and sex-matched asymptomatic controls. Mid-sagittal T2 MRI scans of the lumbar spine were collected, and lumbar vertebral bodies from L1 to L5 were segmented using a self-developed machine learning-assisted algorithm, generating five regions of interest (ROI). MRI signal and spatial biomarkers derived from each ROI were compared between groups using linear mixed-effects models. These biomarkers included mean signal intensity, coefficient of variance (COV), skewness, kurtosis, signal intensity weighted centroid (SIWC), and geometric centroid (GC).
Results:
The mean signal intensity was significantly lower in individuals with NSCLBP compared to controls (p < 0.001), whereas the COV was significantly higher in individuals with NSCLBP (p < 0.0001). Both the SIWC and GC were displaced anteriorly in individuals with NSCLBP relative to controls (SIWC p = 0.002; GC p < 0.001).
Conclusion:
To our knowledge, this is the first study to identify unique vertebral signal and spatial MRI biomarkers in individuals with NSCLBP. Specifically, those with NSCLBP exhibited markedly lower mean signal intensity, greater signal-intensity heterogeneity, and anterior displacement of both structural mass and signal intensity within the lumbar vertebrae. These novel quantitative "vertebral body biomarker profile" (VBBP) phenotypes may provide additional insight into clinically relevant pain phenotypes and support improved patient selection for targeted therapeutics.