Diffusion tensor imaging (DTI) has emerged as a powerful modality for examining microstructural alterations within the spinal cord, offering invaluable insights into various neurological conditions such as traumatic and non-traumatic spinal cord injury, degenerative diseases, and tumors [1–3]. DTI parameters including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), radial diffusivity (RD), apparent diffusion coefficient (ADC), and relative anisotropy (RA) have been demonstrated to be good indicators of normative white matter microstructure and potential predictors of demyelination in pathological states in human and animal studies [3, 9–13].
Numerous studies have investigated adult [2, 3] and pediatric [4–8] spinal cord DTI data.
DTI metrics have shown to be potential biomarkers for injury and disease, but the quantitative interpretation of DTI can be challenging, especially for the spinal cord. The unique architecture of the spinal cord, characterized by its small size, complex fiber organization, and susceptibility to physiological motion, poses significant challenges for accurate imaging and interpretation of DTI data. Furthermore, the lack of standardized acquisition protocols and the inherent variability across magnetic resonance (MR) scanners further complicate the comparison and synthesis of findings across studies and clinical sites. Different MR scanners vary based on magnetic field strength, gradient performance, pulse sequence designs, processing techniques and calculation methods [14]. In recent years, efforts have been made to address these challenges through the development of advanced imaging techniques. Reduced field-of-view (rFoV) diffusion-weighted imaging sequences have demonstrated promise in reducing geometric distortions and artifacts, particularly beneficial for the small dimensions of the spinal cord [5, 15–19]. Despite these advancements, significant gaps remain in our understanding of the reproducibility and reliability of DTI measurements in the spinal cord [4].
The clinical translation of DTI biomarkers for spinal cord pathologies relies heavily on the establishment of robust and standardized imaging protocols, as well as validated harmonization techniques to ensure consistency and comparability across diverse patient populations and clinical settings. Longitudinal ComBat (longComBat), an empirical Bayesian method, is one of a harmonization method that removes additive scanner effects and corrects multiplicative scanner effects by removing heteroscedasticity of model errors across scanners [20]. It is a generalization of a method was originally used in genomics, which has been adapted for brain functional MRI and DTI with promising results [20]. This technique has never been applied to spinal cord imaging prior to our studies. In our prior study, we demonstrated the efficacy of longComBat in decreasing scanner effects on the data from different scanners and field strengths [14].
Few studies have been conducted to show spinal cord DTI reproducibility within scanner [4]. In this study, we concentrate on demonstrating the efficacy of longComBat in decreasing scan-rescan variability on the longitudinal data obtained from the same scanner of the cervical and thoracic spinal cord. We examined the variability of DTI of the spinal cord between longitudinal scans with a single 3T Siemens MR scanner by scanning a sample of forty-two healthy pediatric subjects and twenty-seven pediatric subjects before and after a 2-hour interval. We continue to show that harmonization of human spinal cord DTI data is a crucial prerequisite for facilitating longitudinal and multisite clinical research as well as clinical trials.