StatisticalAtlas in HD Large 540p

[00:00:04]This project explores how advanced neuroiming and computational tools can be used to study neurodeeneration at a high spatial resolution. By combining MRI with statistical brain atlases, we aim to quantify structural and connectivity changes in the brain over time. In particular, this project focuses on Huntington's disease using the mouse model.

[00:00:26]This chapter introduces the foundational concepts and tools that make this type of analysis possible. Starting with a focus on the Duke mouse brain atlas.

[00:00:35]A central challenge in neuroiming is achieving spatial consistency across separate experiments.

[00:00:41]Let's imagine what if the mouse brain had Google maps. Conceptually, the DMBA can be thought of as a map for the general mouse brain. It is a detailed three-dimensional stereotactic atlas, meaning that it defines brain structures within a standardized three-dimensional coordinate system anchored to anatomical reference points, enabling very precise spatial localization, reproducible alignment and analysis across subjects and data sets. We have chosen to use this mouse model due to their genetic similarity to humans and their experimental tractability. While allowing us to reduce the ethical constraints that come with using human subjects for this work, they allow for high resolution imaging at a scale that is not feasible in human studies. Over 500,000 times higher resolution. Their shorter lifespan facilitates the study and data extraction of disease progression and their genetics can also be tightly controlled and easily manipulated.

[00:01:36]A small size of the mouse brain, approximately 3,000 times smaller than the human brain, enables significantly higher imaging resolution. Voxels or volutric pixels can achieve resolutions orders of magnitude around 512,000 times more greater than those in clinical imaging. In fact, diffusion MRI data sets used in the DMA reach resolutions that are up to 2.4 4 million times higher than typical clinical scans at approximately 15 microns cubed per voxil.

[00:02:06]Different imaging modalities differ in their underlying physics and applications. Starting with something most might be a little more familiar with, computed tomography or CT relies on X-rays and is well suited for imaging dense structures such as bone. Magnetic resonance imaging or MRI uses radio frequency signals to probe the water molecules in tissue providing superior contrast for soft tissues and MRI can also capture differences in cytoarchitecture or how cellular structure varies across regions of the brain. Structural MRI shows us what the brain looks like. So this is the size and the shape of regions. While diffusion MRI reveals how tissue is organized at the microructural level by tracking how water moves through fibers.

[00:02:51]Several different diffusion metrics are commonly used to interpret tissue properties. Fractional anisotropy or FA is a scalar value from 0 to one that reflects the directional organization of tissue particularly in white matter tracks. It indicates how highly organized or bundled axon fibers are with higher values indicating ordered healthy tracks and lower values near zero suggesting damage or gray matter.

[00:03:16]Axial diffusivity or AD corresponds to diffusion along axonal fibers with lower values typically signifying axonal injury while radio diffusivity RD reflects diffusion perpendicular to those fibers. A higher RD might indicate reduced structural integrity or demination. And lastly, medial diffusivity or MD is the average of diffusion in all directions and together these provide information about tissue integrity and structural organization.

[00:03:44]Older atlases suffer distortions from tissue processing outside of the skull.

[00:03:48]The Allen atlas is not stereotactic and although paxinos is, the sampling is not contiguous or uniform across all three axes. And so the DMBA allows one to correct the distortion after registration and preserves true three-dimensional structure across all axes, thus helping solve the Humpty Dumpty problem. Once brains are taken apart, it's hard to put them back together consistently without distortion. A common coordinate system ensures the same location corresponds across all specimens independent of orientation or visualization.

[00:04:20]One advantage of this common space is that it lets us integrate different types of imaging data. Here I'm overlaying a TH1 yellow fluorescent protein light sheet image onto an FA map. The TH1 YFP signal labels a subset of neurons specifically highlighting their cell bodies and long projections.

[00:04:37]So it's useful for visualizing neuronomorphology and connectivity. You can see the individual fiber pathways and how neurons extend through different regions of the brain. Everything maps into the same 3D reference space and shared coordinates enables reliable group level statistics comparisons and crosslab studies.

[00:04:57]And finally, diffusion MRI allows us to trace these white matter pathways forming the structural conneto. A map of how different brain regions are connected. In this TDI image, colors represent the direction of diffusion.

[00:05:09]Red indicating left to right, green front to back, and blue in and out of the plane.

[00:05:17]Huntington's disease is a progressive neurodeenerative disorder that affects movement, cognition, and behavior.

[00:05:23]Symptoms of Huntington's include Korea, which is involuntary spasmotic movements, impaired coordination and balance, muscle rigidity, difficulty speaking and swallowing, cognitive symptoms like dementia, and psychiatric symptoms like depression. In the HTT gene, which provides instructions for making a protein called Huntington, HD is caused by a genetic mutation where three DNA bases, cytosine, adanine, and guanine repeat over and over in this section. The more repeats, the earlier and more severe the disease. The Huntington gene is especially toxic to the striatam, which is an area of the brain important for mood, motivation, and movement. In this study, we use two mouse models, Q11 and ZQ175DN, which differ in how many CAG repeats they carry and how quickly the disease progresses. We compare hetererozygous mice, which carry the mutation to wild type controls. And because all of the brain data is aligned into the same spatial framework, we can track how these changes evolve over time at 2, 6, 10, and 15 months. In specific connections, neurons degenerate, connections weaken, axons and neurons die, and are no longer effective, especially in the primary motor cortex, which causes these severe motor deficits often seen in HD patients. These changes aren't always obvious on standard imaging, but diffusion MRI allows us to detect these subtle differences in tissue structure and connectivity. Now that I've introduced the data, this chapter walks through how we have turned these brain images into meaningful quantitative measurements. Our goal is to build a statistical atlas of the brain. To do this, we break the process into a pipeline of a couple major steps that take us from raw imaging data to a final atlas. A key part of this process is the DMBA, which gives us a consistent spatial map. The pipeline is an automated process that moves data from acquisition all the way to statistical analysis. It includes image acquisition, reconstruction, registration to a template, labeling regions, and finally statistical testing. Image derived phenotypes are quantitative measurements extracted from imaging data. So basically turning brain images into numbers we can analyze. For example, volume decrease might reflect tissue health with a decreased number of neurons, while diffusion metrics might capture how well neuropathways are preserved and organized. Each scan takes about 24 hours and the MRI process is non-destructive, meaning that the brain remains intact inside the skull to avoid warping. Samples are perusion fixed and prepared through caization before scanning. After reconstruction, we calculate diffusion scalers like FA, AD, and volume and the box level. These values then allow us to quantify brain structure and connectivity. We also generate connetos and may use AI to help account for statistical errors during registration.

[00:08:15]The DMBA allows us to put all animals into the same space through a process called registration. This is critical because it ensures that differences we observe are biological and not just due to differences in orientation or size between brains. As an example, this LSM light sheet data is shown in red and after correction into the geometric space is shown in green. The original light sheet image is greatly distorted as a product of the processing required to collect the data. Instead of registering each brain directly to the atlas, we first create a minimum deformation template or MDT. This is built by averaging all specimens together using aphine transformation and iterative refinement. The MVT improves consistency because all specimens are first aligned to a common template. This ensures that each brain has similar success in registration before mapping to the atlas.

[00:09:08]Once registered, we label regions of interest or ROIs. We then calculate mean values for each region which become the basis for statistical comparisons. The statistical workflow involves computing regionwise averages, running ANOVA tests, applying multiple comparison corrections and then calculating effect sizes like Coen's D. We use Nway ANOVA models to compare groups such as transgenic versus control animals.

[00:09:35]Because we test many regions, we apply the Benjamini Hawkberg correction to control the false discovery rate and reduce false positives. Coend measures effect size by comparing the difference in means between groups relative to their standard deviation. Unlike some metrics, it includes directionality showing whether values increase or decrease.

[00:09:57]Coen's F on the other hand comes from the ANOVA model and reflects total variability, but it doesn't show direction.

[00:10:05]In addition to CoenD, we also use metrics like statistical power or p value and percent change. So our method improves on previous work by offering much higher spatial resolution, better contrast, minimal distortion, and the ability to process large numbers of specimens efficiently. Overall, this pipeline allows us to move from raw imaging data to statistically meaningful brain atlas. By combining multiple advanced imaging modalities, registration, and statistical modeling, we can identify real biological differences across groups. And because everything is in the same space now, we can quantify how patterns evolve and compare stages, which is critical for watching the progression of Huntington's disease and identifying effective time windows for potential treatment. First, we can compare the CoenD for volume change for the brain at 2 months and at 15 months. Coend is a standardized measure of effect size. It tells us how large a change is relative to the variability in the data. A coend value of one means the difference between two groups is one standard deviation.

[00:11:07]Instead of just looking at percent change, which only shows raw differences, coend accounts for how consistent that change is across animals. This makes it a much more informative and meaningful measure of statistical robustness of the changes between diseased and non-deseased animals. In general, we see that the direction of changes are similar from two months to 15 months except for a slight difference in the olfactory bulbs with much more drastic changes in the 15 months. As expected, there's significant volume increase in the cerebellum, a region responsible for coordinating smooth, precise voluntary movements, maintaining balance and posture and motor learning. This increase is possibly due to inflammation and other factors. There is decrease mostly everywhere else and notably in the striatam which is a region that regulates the initiation and decision making regarding movement, emotion and cognitive function. These two regions also have noticeably low P values indicating that these changes are unlikely due to chance.

[00:12:05]Next we can compare the coen or FA change reflecting differences in white matter organization and cytoarchitecture.

[00:12:12]In contrast with volume change, FA change is much more heterogenous throughout different regions and the twomonth and 15month atlases are much more different compared to the comparison for volume. There is slight decrease in the cerebellum for the twomon but increase for the 15month and there is increase in the stridum for the twomon but decrease for the 15mon. This makes sense as lower FA indicates damage to white matter connectivity.

[00:12:37]We see that the p value is still pretty significant for the cerebellum but not as much for the stratum. Again, there is much decrease in the frontal and parietal temporal loes areas responsible for sensory input, executive function, and comprehension.

[00:12:52]Overall, MRytology provides quantitative whole brain three-dimensional mapping of changes in mouse models of disease. By mapping this data to the Duke mouse brain atlas, we can build statistical atlases of neurodeeneration.

[00:13:06]These atlases can summarize complex highdimensional evolution of disease progression and together these provide a powerful new platform for quantitative studies of neurodeeneration.