Low back pain is a widespread problem throughout the developed world. There is a need for a rigorous multidisciplinary approach to studying this problem and evaluating solutions. Computational methods are beginning to be used to evaluate and test new surgical treatments and medical devices. With the application of computational models in a clinical setting the need for more robust, well validated, and efficient computational techniques is increasing. Patient-specific and population-based computational methods have been applied to some regions of the body but have not been previously published for the lumbar spine. The primary goal of this research was to develop robust methods for performing probabilistic simulation based on anatomical variation in the full lumbar spine. The research has been broken up into three separate aims.
Aim 1 was to develop a method to repeatably and reliably identify and extract geometric features and landmarks of lumbar vertebrae and use that information to automatically create finite element models with subject-specific geometry. An automated method was developed and tested. Eighteen subject-specific full lumbar spine finite element (FE) models were created based on automated landmark identification of 90 lumbar vertebrae. The subject-specific FE models were produced with good accuracy, quality, and robustness. The new automated method represents an improvement over manual and semiautomated methods previously reported in the literature.
Aim 2 was to validate the automation process and resulting FE models. Mesh convergence, direct validation, and indirect validation studies were performed. The studies showed that the automated models can be used to reliably evaluate lumbar spine biomechanics, specifically within the intended context of use: in pure bending modes, under relatively low non-injurious in vivo loads, to predict torque rotation response, disc pressures, and facet forces.
Aim 3 was to create and evaluate a statistical shape model (SSM) of the lumbar spine for use in probabilistic modeling. The research successfully demonstrated the use of a SSM combined with automated methods for landmark identification and FE model generation to create a fully parameterized FE model of the lumbar spine. The SSM was evaluated using compactness, generalization ability, and specificity. The shape modes were also evaluated visually, quantitatively, and biomechanically. Functional FE models of the mean shape and the extreme shapes (±3 standard deviations) of all 17 shape modes were created demonstrating the robust nature of the methods. This research represents an advancement in FE modeling of the lumbar spine and will allow population-based modeling including anatomical variation in the future.