ASSESSING MAGNETIC RESONANCE IMAGING DIFFUSION TENSOR VARIABILITY IN HEALTHY SUBJECT ’BIG DATA’

Abstract

Diffusion Tensor Imaging (DTI) is a powerful tool for assessing microstructural integrity in both healthy and diseased populations. However, variations due to magnetic resonance imaging (MRI) vendor differences and demographic factors must be quantified to improve clinical utility. This study integrated three analyses: (1) consistency of DTI-derived metrics across vendors using a DTI calibration phantom, (2) age, sex, region, and vendor-related differences in DTI metrics in a large healthy control population, and (3) an application of ‘Big Data’ DTI to perform outlier detection in the cerebellum of a mild traumatic brain injury (mTBI) patient population. The calibration phantom study evaluated consistency of the acquisition of a rank-2 diffusion tensor. DTI metrics, primarily fractional anisotropy (FA), a measure of white matter integrity, was evaluated across three MRI vendors (GE, Siemens, Philips). No significant differences in FA were found between vendors (p > 0.05, 1.3% differences), suggesting that DTI-derived metrics are reliable and reproducible in a phantom environment across scanner vendors. Compared to a study using a large dataset of healthy controls (n = 2,700), significant variability effects (p < 0.0001) were found to be associated with specific demographic and vendor factors. The main findings included a significant decline in FA due to aging, which we expected, sex differences with FA being overall greater in males (p < 0.0001), regionally, the tapetum section of the corpus callosum exhibited the highest variability across subjects, and all three vendors had significantly different FA values from each other. This was the first ‘Big Data’ study that evaluated subject age, sex, and MRI vendor all in one statistical model. This important advancement allowed for assessment of interactions between these variables. A final study compared microstructural abnormalities in the cerebellum of mild traumatic brain injury (mTBI) patients (n = 51), to an age, sex, and vendor matched control group (n = 150 per matched group), while performing a case-wise two-tailed Z-score outlier analysis. This revealed unique cerebellar FA reductions, particularly in males (n = 24), and older populations (56-60 yrs) appeared more likely to have numerous cerebellar regions affected (i.e. n > 8 ROIs). Linear regression, random forests, and hierarchical clustering analysis further revealed the importance of Z-score analysis compared to its Post Concussion Symptom Score (PCSS) counterpart. These findings highlight the clinical utility of ‘Big Data’ DTI but also demonstrates the need for vendor harmonization tools such as ComBAT and underscores age, sex, and vendor related DTI differences. This study also shows that investigating neuropathology through a ‘Big Data’ DTI lens can demonstrate novel findings in understanding mTBI such as cerebellar vulnerability. Future work should focus on developing standardized harmonization pipelines, while also leveraging machine learning approaches and targeting regions of interest like the corpus callosal tapetum or the cerebellum to enhance the detection of subtle neuropathological changes in mTBI.