It has just hit the press that I have been awarded a Ruth L. Kirschstein Predoctoral Individual National Research Service Award from the National Institutes of Health (NIH) for my thesis project entitled “A novel multi-modal, multi-scale imaging pipeline for the validation of diffusion MRI of the brain”. This award will fund up to three years (totaling $120,979) of my thesis research through the National Institute of Neurological Disorders and Stroke (NINDS).
Specific aims of the fellowship:
Diffusion Tensor Imaging (DTI) is a powerful magnetic resonance imaging tool used to noninvasively report 3D microstructural properties of nervous tissue on a macroscopic scale, and it has played an important role in the understanding and diagnosis of a number of neurological disease processes. New methods of reconstructing orientation distribution functions (ODFs) from DTI data are rapidly being developed, each seeking to identify and model the orientation distribution of distinct, sub-voxel axon fiber populations. The 3D orientations of these fiber populations are passed into tractography algorithms as a potential noninvasive means of performing neural connectivity analysis across whole brains. However, efforts to validate tractography pipelines have found poor specificity when compared primarily to histological- or phantom-based ground-truth datasets. These failures have led to a push for more microstructure-driven, multi-scale validation and improvement efforts for DTI.
Studies seeking to validate ODF reconstruction methods have typically relied on serial optical histology as a ground-truth dataset. The typical pipeline is to generate ODFs from the high-resolution histology data using a computer-vision technique called “structure tensor analysis,” which uses image intensity gradients to estimate local, voxel-wise fiber orientations. These orientation estimates are then binned across regions of interest (ROI) the size of a DTI voxel in order to form ground-truth ODFs, which are compared pair-wise to ODFs reconstructed from DTI data of the same specimen. Histology-based ground-truth datasets used for this purpose rely on the labor-intensive process of physically sectioning, staining, and optically scanning hundreds of slices of the tissue of interest. The slice thickness is necessarily at least 4-20 times thicker than the achievable in-plane resolution ~5000 nm vs. 250 nm), yielding non-isotropic volumetric reconstructions; distortions introduced by sectioning further limit the ability to align the slices and extract faithful information on the 3D orientation of fiber populations.
We are pioneering the use of metal-stained synchrotron micro computed tomography (microCT) as a means of performing isotropic, 3D imaging of whole mouse brain specimens at micron resolution, with the potential ability to resolve every axon in the brain. We propose to optimize a pipeline to use microCT data together with serial electron microscopy (EM) to validate and characterize DTI ODF reconstruction algorithms with respect to the underlying neurological tissue structure. Imaging the same mouse brain using DTI, microCT and EM will provide an unprecedented DTI validation dataset with resolution scales spanning six orders of magnitude. The results of this work will address the limitations of previous histology-based validation studies, and provide a key microanatomical understanding of the basis of the DTI signal. The specific aims are:
Aim 1: Model phase contrast to optimize microCT data acquisition. Techniques to exploit phase contrast in synchrotron x-ray imaging typically make the assumption of a non- or weakly-absorbing sample. However, in preliminary studies, we have observed phase effects in strongly-absorbing mouse brain specimens stained with heavy metals. We propose to develop a novel theoretical imaging model that accounts for both phase and absorption contrast in these samples. This model will be used to optimize parameters of the microCT data acquisition to enhance the contrast of axon fiber bundles.
Aim 2: Validate DTI reconstruction methods using ground-truth microCT ODFs. We will compute ground-truth ODFs from the microCT data across a whole mouse brain using structure tensor analysis. After spatial registration, the ground-truth ODFs will be compared to ODFs calculated using a variety of reconstruction methods on DTI data from the same specimen, generating algorithm-specific spatial maps of DTI accuracy. These maps will highlight gross anatomical regions associated with DTI success and failure.
Aim 3: Characterize DTI performance using underlying tissue microstructure. A number of ROIs from aim~2 displaying either large or small ODF discrepancies will be selected for follow-up imaging with EM. Morphological features from the nanometer-scale EM data and micron-scale microCT data will be used to characterize the performance of DTI. The statistical relationship between these features and scalar ODF validation metrics will be used to identify significant anatomical characteristics that are correlated with both the success and failure of ODF reconstruction algorithms.
Upon completion, this project will generate a comprehensive validation map for a number of DTI reconstruction methods, exploiting the advantages of a novel dataset uniquely tailored to this purpose. The understanding of key microstructural features that influence DTI algorithm performance can be used in future work to create better, adaptive models and acquisition schemes to better leverage fiber orientation and connectivity information in the treatment of neurological disease.