To distinguish flexible and rigid lumbar curve from MRI texture analysis in adolescent idiopathic scoliosis: A feasibility study.

Abstract

Imaging in side bending, supine, traction, fulcrum, and push prone are examples of methods used to evaluate the curve reduction of scoliotic spine. However, being able to determine spine curve flexibility from MRI would eliminate the need of additional X-ray radiation related to radiograph acquisition in side-bending. To find specific texture features of lumbar postural muscles on MRI that can distinguish flexible from rigid lumbar scoliotic curves. We hypothesized that the changes occurring in postural muscles with scoliosis can be seen with MRI. Retrospective study case control. With Institutional Review Board approval and informed consent, 15 adolescents with idiopathic scoliosis and scheduled for surgery were involved. T1 -weighted MR images were performed on a 1.5T system using a spin echo sequence in the axial direction. The spinal erector, quadratus lumborum and psoas major muscles were analyzed using textural features. Principal component analysis (PCA) and agglomerative hierarchical clustering (AHC) were used to classify the lumbar postural muscles and calculate performance metrics. The lumbar flexibility index, measured from suspension tests, was used as ground truth measurement. The five discriminant features (out of 34 tested features) obtained from PCA were able to keep over 90% of the variability of the dataset. The right and left spinal erector and the left psoas major had the highest performance metrics to classify the spinal curve flexibility, with an accuracy over 0.80, a sensitivity over 0.82, a specificity over 0.68, and a Matthews correlation coefficient over 0.57. This study analyzed MRI using texture information of muscle to distinguish flexible from rigid scoliotic curves. Some postural muscle such as the spinal erector and the psoas major are more likely to reflect the curve flexibility of a scoliotic participant. 2 Technical Efficacy: Stage 1 J. Magn. Reson. Imaging 2017.