Application of Statistical Shape Models to Standard Lumbar MRI for Stenosis Treatment Stratification: Severe Versus Normal Stenosis.

Background: Lumbar spinal stenosis is a prevalent and debilitating diagnosis, which in severe cases requires surgical treatment to relieve nerve root pressure. Often, treatment plans are based in part on subjective, qualitative, and limited MRI assessment. Statistical shape models (SSMs) have the potential to improve treatment indications by uncovering morphological features that work synergistically but are difficult to assess independently or directly quantify. This study examined whether 2D SSM using standard clinically relevant MRIs can differentiate between severe and normal stenosis patients in a population with low back pain. Methods: A total of 62 patients were analyzed from an open-access Lumbar Spine MRI dataset, with variable parameters, and classified as severe or normal stenosis (at L45). Intervertebral disc (IVD) and posterior element (PE) edges were extracted using the SegNet algorithm, then aligned via generalized Procrustes analysis. Aligned shapes were used for principal component analysis and principal components (PCs) were evaluated for IVD and PE independently and together. To observe if SSMs improved from anatomical measurements, facet angles, spinous process length, IVD, and thecal sac diameters were manually extracted. ANOVA and ROC analysis was run to determine the measurements' ability to discriminate between groups. Results: PC1 explained 38% and 32% of the IVD and combined (IVD & PE) shape variances and had a clear difference between the severe and normal stenosis groups ( Conclusions: SSMs were able to distinguish between the severe and normal stenosis groups, specifically PC1 for both IVD and combined SSMs. Hence, this study demonstrates that SSMs could be a quantitative tool to improve stenosis diagnosis and treatment planning using clinical 2D MRIs.