SIENA-SC: accurate, robust, and automated longitudinal spinal cord change analysis

In the thesis I present a new method to assess longitudinally spinal cord (SC) atrophy, called SIENA-SC. The spinal cord (SC) is an important area of the central nervous system (CNS) as it plays a critical role in both motor and sensory functions. Magnetic resonance imaging (MRI) allows for in-vivo visualization of the SC, providing valuable insights into its structure and function and helping the diagnostic workup of several neurological conditions such as multiple sclerosis (MS), amyotrophic lateral sclerosis (ALS), and spinal cord injury. In addition, the quantification of SC atrophy can be used to monitor disease progression and as an outcome measure in clinical trials. However, these measures have not reached yet a degree of robustness and reliability comparable to those of brain atrophy, which are currently used to monitor disease progression and treatment efficacy in neurodegenerative diseases. To date, there are two possible strategies to assess longitudinal volumetric changes of SC over time. The first is to compare the cross-sectional area (CSA) from segmentation maps obtained independently at each timepoint. This approach provides an indirect estimation of the atrophy rates and is limited by the inaccuracy of segmentation maps due to partial volume effects. The second approach relies on these SC segmented masks, but they are registered on a common reference space, providing a direct estimation of atrophy measurements. The first attempt to provide a reliable tool to measure SC atrophy longitudinally was made through the generalized boundary shift integral (GBSI)., an optimization of an algorithm that has been already validated to assess brain atrophy. Structural Image Evaluation using Normalization of Atrophy (SIENA) is a widely recognized method that employs registration techniques to measure changes in brain volume over time. Over the past two decades, SIENA has been extensively used to evaluate brain atrophy due to its user-friendly nature and its reliability. SIENA calculates the zeros of the second derivative of the intensity profiles of the lines perpendicular to the surface in the two images to be compared. This approach helps to reduce the influence of MRI intensity inhomogeneity, as it relies on the shape of the intensity profiles rather than their specific intensity values. By using this method, the variability introduced by intensity variations is minimized, allowing for more robust and reliable measurements of atrophy. The study presented in this thesis introduces SIENA-SC, which is an adapted version of the SIENA method, designed to calculate the percentage of spinal cord change (PCVC) over time directly on the cord edges. Our main objective was to provide a fully automated approach that reduces variability in measuring SC atrophy and offers a solution similar to longitudinal brain atrophy measurements. In the first experiment, using a multicenter dataset including 13 scan-rescan and 190 Healthy Control (HC) subjects, SIENA-SC showed to have a lower measurement error than GBSI and CSA, reflected by lower standard deviation, coefficient of variation and median absolute variation. In the second experiment, the lower measurement variability of SIENA-SC than GBSI and CSA, was confirmed in a dataset of 65 Multiple Sclerosis (MS) subjects and the same 190 HC of the previous experiment, thus resulting into a better differentiation between patients with MS and HC, an improvement of statistical power, and reduction of sample size estimates. In conclusion, SIENA-SC showed to be robust and feasible when assessing SC atrophy using brain MRI scans routinely acquired in clinical practice. Longitudinal spinal cord atrophy measured through SIENA-SC has the potential to become a recognized outcome measure for clinical trials. However, it should currently be considered as a secondary outcome measure until additional advancements enhance the ease of acquisition and processing. Further developments of the methods are needed to make the process more streamlined and user-friendly.