Modelling The Spread Of Atrophy In Huntington's Disease Using Network Diffusion Model (NDM)

Huntington's disease (HD) is a progressive neurodegeneration which has symptoms such as movement dysfunction, cognitive abnormalities, and psychiatric disturbances. Neurodegeneration in HD is characterised by pathology that spreads throughout the cortico-striatal network. There is growing recognition that the transfer of mutant huntingtin (mHTT) protein is cellular and shaped by structural organisation of the brain, provides a general framework underlying progressive spread of degeneration in HD. However, relatively little is known regarding how such progression occurs over time. This knowledge is critical to inform future drug discovery efforts where neuroimaging methodology can be used to develop the potential therapeutic compounds in targeting vulnerable neural circuits in HD.
This research aimed to develop and apply a novel graph theoretical network diffusion model to predict how and where in the brain neurodegenerative process is seeded in HD. We used longitudinal MRI scans (N=106 Premanifest HD (pre-HD) and N=89 Healthy Controls), collected 12 and 24 months from the Track-On HD study. We implemented Network diffusion model (NDM), which was previously applied to symptomatic individuals with HD, in the Premanifest HD population.
To evaluate the spatial patterns in change in brain volume of HD brain regions compared to controls, we segmented the T1-weighted structural MRI scans into 82 brain regions using FreeSurfer based tools. A linear mixed-effects model is statistically used to evaluate the brain volume differences (across 82 brain regions) in different time visits. The outcome of this statistical approach revealed that degeneration over time in pre-HD compared to controls shows extensively higher (p<0.05). The putamen and caudate showed the highest atrophy sub-cortically. Cortically, the highest atrophy is observed in the visual cortex (lateral occipital) and temporal cortex (superior and middle temporal).
Next, we implemented NDM to simulate the longitudinal pattern of degeneration across the 82 brain regions. The NDM assumes that the linear diffusion of pathological proteins facilitates pathology spread in the HD brain via the brain's structural connectome. Therefore, we used a canonical connectome from the healthy brain (82 x 82 IIT connectome) to simulate the brain's spread process. We found that initiating the diffusion process from pallidum, thalamus, and putamen predicts a pattern of degenerations (predicted atrophy) which is significantly correlated (p<0.05, corrected) with longitudinal degeneration measured using MRI scans. However, the predictive ability of NDM was moderate at best (maximum Pearson correlation between predicted and measured atrophy = 0.48, p <0.0001). These findings suggest that NDM can only moderately predict the pattern of degeneration in the HD brain.
Overall, this thesis explores how NDM can better understand the progression of degeneration in HD. Our comparative study between healthy and patient’s groups statistically showed a progressive pattern of atrophy in the cerebrum. Furthermore, the results from NDM suggests that NDM could be helpful to model the progression of atrophy in subcortical and cortical regions in HD. Thus, NDM may build bio-physically informed machine learning algorithms for predicting future degeneration from baseline MRI scans.