Analytic Flexibility of dMRI Preprocessing and free-water corrected DTI (fwDTI) in Parkinson's Disease (PD)
Background
Parkinson’s Disease (PD) is a promininent form of dementia that has been challenging to prodromally identify. The typical imaging derived phenotypes (IDPs) that have worked well for other conditions, like cortical thickness, have failed to reliably predict the prognosis of the condition. Different modalities, like diffusion magnetic resonance imaging (dMRI), quantify the movement of water around axons and can be modeled to describe the qualities of the white matter tissue. Given the established changes of PD occur in these white matter connections, it is a promising methods to more reliably predict this disease.
Methods
dMRI Processing needs to be correctly preprocessed in order to reliably model the underlying anatomy captured by the scan. Some steps, such as eddy current and motion correction, are necessary to reliably interpret the data. Other steps, such as denoising and Gibbs de-ringing, can be applied with different algorithms at different stages and may have a differing impact on the eventual interpretability. Much of this work is being compared through two pipelines, (QSIPrep) and (TractoFlow). These pipelines are largely identical, but standardizing the comparison between similar tools for identifying their impact on an analysis is important to understand.
The most commonly used model for interpreting dMRI data is the diffuion tensor (DTI). This summarizes the observed movement of water within the brain, providing an estimated axis of primary movement and a measure of isotropy / anisotropy for that orientation. A free-water corrected tensor (fwDTI) adds an additional anisostropic metric measuring the “background” isotropy independent of the signal in the white matter. This information corresponds to surrounding support tissue of the white matter measured by the regular tensor components. Changes in this component have been found to correspond to the clinical progression of PD in different subcortical connections.
Goal
My goal is to describe the impact of processing decisions on the ability of diffusion tensor models to make clinically relevant predictions. I am also exploring the impact of the different parameterizations and methods of predicting fwDTI. I hope to identify the processing steps and modeling procedures that maximize the utility of this promising method. With this model we can better understand the underlying neural changes associated with PD.
People
