Quantification of intervertebral displacement with a novel MRI-based modeling technique: Assessing measurement bias...

The purpose of this study was to develop a novel magnetic resonance imaging (MRI)-based modeling technique for measuring intervertebral displacements. Here, we present the measurement bias and reliability of the developmental work using a porcine spine model. Porcine lumbar vertebral segments were fitted in a custom-built apparatus placed within an externally calibrated imaging volume of an open-MRI scanner. The apparatus allowed movement of the vertebrae through pre-assigned magnitudes of sagittal and coronal translation and rotation. The induced displacements were imaged with static (T₁) and fast dynamic (2D HYCE S) pulse sequences. These images were imported into animation software, in which these images formed a background 'scene'. Three-dimensional models of vertebrae were created using static axial scans from the specimen and then transferred into the animation environment. In the animation environment, the user manually moved the models (rotoscoping) to perform model-to-'scene' matching to fit the models to their image silhouettes and assigned anatomical joint axes to the motion-segments. The animation protocol quantified the experimental translation and rotation displacements between the vertebral models. Accuracy of the technique was calculated as 'bias' using a linear mixed effects model, average percentage error and root mean square errors. Between-session reliability was examined by computing intra-class correlation coefficients (ICC) and the coefficient of variations (CV). For translation trials, a constant bias (β₀) of 0.35 (±0.11) mm was detected for the 2D HYCE S sequence (p=0.01). The model did not demonstrate significant additional bias with each mm increase in experimental translation (β₁Displacement=0.01mm; p=0.69). Using the T₁ sequence for the same assessments did not significantly change the bias (p>0.05). ICC values for the T₁ and 2D HYCE S pulse sequences were 0.98 and 0.97, respectively. For rotation trials, a constant bias (β₀) of 0.62 (±0.12)° was detected for the 2D HYCE S sequence (p<0.01). The model also demonstrated an additional bias (β₁Displacement) of 0.05° with each degree increase in the experimental rotation (p<0.01). Using T₁ sequence for the same assessments did not significantly change the bias (p>0.05). ICC values for the T₁ and 2D HYCE S pulse sequences were recorded 0.97 and 0.91, respectively. This novel quasi-static approach to quantifying intervertebral relationship demonstrates a reasonable degree of accuracy and reliability using the model-to-image matching technique with both static and dynamic sequences in a porcine model. Future work is required to explore multi-planar assessment of real-time spine motion and to examine the reliability of our approach in humans.