MULTIMODAL IMAGING OF CEREBRAL MICROHEMORRHAGES AND WHITE MATTER DEGRADATION IN GERIATRIC PATIENTS WITH MILD TRAUMATIC BRAIN INJURY

Traditionally, neurobiologists have utilized microscale techniques of scientific investigation to uncover the fundamental organization and function of brain cells and neuronal ensembles. In recent decades, however, macroscale brain imaging methods like magnetic resonance imaging (MRI) and computed tomography (CT) have facilitated a wider scope of understanding neural structure and function across the lifespan. Thanks to such methods, a broader picture of the relationship between microscale processes— studied by neurobiologists—and macroscale observations—made by clinicians—has emerged. More recently, the vascular component of neurodegeneration has come under renewed scrutiny partly due to increased appreciation of the relationship between neurovascular injury, cardiovascular disease and senescence. Cerebral microbleeds (CMBs) are among the smallest lesions of the cerebrum which can be visualized using MRI to indicate blood-brain barrier (BBB) impairment; as such, this class of hemorrhages are important for the evaluation and macroscale detection of geriatric patients’ microscale pathologies associated with neurovascular disease and/or neurodegeneration. This chapter details a streamlined protocol for MRI/CT multimodal imaging data acquisition, archiving and digital processing, including methods tailored for the analysis of susceptibility-weighted imaging (SWI) and diffusion-weighted imaging (DWI) scans to reveal CMB-related alterations of the human connectome. Efficient and effective MRI/CT methods like ours, when tailored for CMB and connectome analysis, are essential for future progress in this important field of scientific inquiry.


Introduction
Microscale alterations effected upon the brain by neurological and metabolic disease are typically studied by cell biologists at the level of cells and neuronal assemblies using microscale imaging techniques.
Nevertheless, macroscale imaging methods like magnetic resonance imaging (MRI) and computed tomography (CT) may complement such efforts by enabling scientists to relate microscale pathology to macroscale findings, thereby offering a broader picture of aging and neurodegeneration.Throughout the past two decades, multisite brain imaging studies have become increasingly common, such that it is vital for imaging data to be acquired, processed and analyzed using streamlined workflows which minimize site-related differences in acquisition and analysis procedures.The adequate management of imaging data acquired from older adult populations has special significance, given geriatric patients' higher rates of overall morbidity, neurologic disease and neuropathology incidence.For example, persons aged 65 and older are more vulnerable to traumatic brain injury (TBI)-particularly mild TBI (mTBI)-and mild cognitive impairment (MCI) than any other group of adults; this can lead to substantially lower quality of life compared to younger individuals, often in tandem with cardiovascular, neurovascular, neurological and metabolic diseases, including dementia [1].
Improving the validity and reliability of procedures for the analysis of brain imaging data acquired from vulnerable populations-like the elderly-is of high priority in geriatrics and gerontology research.To achieve such improvement, research procedures and protocols must be described in detail, recorded thoroughly and published.Openly available protocol descriptions can improve the reproducibility of neuroimaging studies, define reference workflows for the implementation of multisite studies, and clarify methodological imprecisions to both scientists and clinicians.
In recent years, the vascular component of neurodegeneration has received increased attention due to mounting evidence on the relationship between neurovascular injury, cardiovascular disease and aging.
Cerebral microbleeds (CMBs) are parenchymal lesions of the cerebrum which can indicate impairment of the blood-brain barrier (BBB).They may be small as ~1 mm 3 on certain types of MRI [1], and are diagnostically important in geriatric patients because they represent macroscale manifestations of microscale pathological processes associated with neurovascular disease and neurodegeneration.This chapter describes a set of standardized, streamlined methods for the acquisition, processing and comprehensive analysis of CMB-positive MRI/CT scans of the aging brain, including the specification of suitable inclusion and exclusion criteria for older participants, MRI and CT scanner specifications, MRI imaging sequence parameters, data archiving strategies and processing methods for both CT and MRI, including T 1 -and T 2 -weighted anatomic MRI, diffusion-weighted imaging (DWI) and susceptibilityweighted imaging (SWI, which can assist in identifying CMBs) [1].

Methods
The first part of this section outlines procedures to recruit volunteers for imaging studies like those previously undertaken by our laboratory.This is followed by a description of our data acquisition protocol, and then by a detailed narrative of data archiving and processing steps to prepare data for subsequent CMB analysis and connectomics.

Inclusion and Exclusion Criteria
The following are inclusion and exclusion criteria for older healthy control (HC) participants, for individuals with MCI and for TBI survivors.

Inclusion criteria for all groups
1. Be at least 40 years of age This is achieved by removing identifying information from the DICOM file headers.Next, data are transferred to a secure online database managed by the scanning facility (see Note 1).Following this transfer, data are transferred to a depository maintained by the laboratory, where they can be accessed and processed at ease (see Note 2).
This section describes preprocessing steps for both MRI and CT scans.A set of universal MRI processing steps are applied to all MRI volumes, whereas DWI and SWI data undergo additional processing steps which are specific only to these two types of data.All steps for universal MRI processing and DTI are undertaken using the fMRI Standard Library (FSL) image processing toolbox (fsl.fmrib.ox.ac.uk/fsl).

MRI Preprocessing Steps.
The following set of preprocessing steps are applied to all MRI volumes, including those acquired using DWI and SWI sequences.6. Bias field correction is applied to each volume using a fourth-order polynomial [8].

DWI Processing
Steps.The following steps are applied to DWI data in addition to those previously described.
1. B 0 volumes are identified as those volumes acquired while no diffusion gradient is being applied 2. Skull removal is implemented and the brain mask is computed 3. Susceptibility artefact correction is implemented 4. Motion correction is applied 5. Eddy-current artefact correction is performed 6.A suitable rotation of the B matrix is applied to alleviate orientation artefacts (see Note 4).   1.At each WM voxel, the uncertainty ellipsoid is defined as a volume around a voxel where any DTI-derived WM measures require detailed analysis, since they may be influenced by confounding factors to a greater extent (see Note 5).

If quantification of WM changes over time is required, the following steps are taken:
 streamlines are assumed to be piecewise-differentiable 3D space curves with an invariant curvature and torsion  any two curves are determined to have equal curvature and torsion if they lie within the same transformation of each other and if there is a curve index correspondence between parametrizations  Within each subject, curve index correspondence is established [9] by generating arc length parametrizations for each curve, time point and perilesional streamline bundle  Point-to-point co-registration is performed by implementing a local transformation using Schönemann's solution to the orthogonal Procrustes problem [10].
 The global residue of squared inter-curve distances within the local transformations is minimized to approximate a global transformation which matches source and target curves  FA values and pointwise descriptive statistics of each measure are calculated for the arclength coordinates of each streamline [9] and [11]  The range of arc-length coordinates shared between most streamlines within each bundle of interest is identified  To compare mean FA differences over time within perilesional streamline bundles and then evaluate whether or not they may be due to artefacts, two empirical reference mean FA distributions are established (one for younger subjects and one for older subjects)  The mean and variance of the reference distributions are calculated over the streamlines  Distribution parameter values are pooled within each age group  The null hypothesis of no difference in FA across time points is tested for each perilesional streamline bundle of interest.

WM streamline prototyping is implemented as follows:
 An anatomical clustering atlas is used to annotate and categorize WM streamline clusters (see Note 6) as deep WM bundles (such as major association and projection bundles, commissural bundles, cerebellar and brainstem connection bundles), short bundles and medium-range superficial bundles, which are organized by the brain lobes that they connect  Connections identified as potential false positives are annotated for exclusion  Following WM parcellation, WM prototyping is implementing by identifying the most representative WM streamline in each cluster, i.e. the streamline whose trajectory is most similar to the average trajectory of the streamline in the bundle (see Note 7).The sensitivity and precision of the CMB identification method can be evaluated against a manual CMB labelling method conducted by a neuroanatomy and neuroradiology expert.Specifically, to calculate the sensitivity and precision of the approach, a manual detection algorithm is used, as follows.

SWI Processing
1. Potential CMBs are identified in images using a voxel classifier based on Microbleed Anatomic Rating Scale (MARS) guidelines [12].
2. Identified CMBs are classified as either certain (represented by small, clearly delineated circular shapes) or doubtful (represented as components which are not well-delineated and which do not conform to strictly circular shapes) 3. Hypointensities on the boundary of the brain are disregarded, as CMBs are not associated with the meninges 4.An object-based classifier which assesses CMB shape is applied to eliminate false positives CMB assignments 5. CMB sphericity violations are determined by calculating CMB object curvature and identifying values which exceed the curvature of a sphere by a substantial amount [13].
6.If CMB etiology must be classified as either traumatic or non-traumatic (e.g.related to cerebral amyloid angiopathy or hypertensive vasculopathy), a CMB is assumed to be TBIrelated if (A) at the acute stage of TBI, the CMB is partially or fully surrounded by one or more focal FLAIR hyperintensities, and (B) at the chronic stage of TBI, the FLAIR hyperintensities are no longer detectable.
1.The parameters of Gaussian intensity distributions are computed for three tissue classes [WM, grey matter (GM) and cerebrospinal fluid (CSF)] 2. Using the following steps, probabilities are assigned to voxels to reflect their likelihood of belonging to a specific tissue class in the Gaussian mixture model:  An objective function is derived from several Gaussian random variable models and its value is minimized using a nonlinear optimization procedure of the user's choice  To complete the classification process, a priori tissue probability maps available in a modified MINI 152 probabilistic atlas [20] are identified for each tissue class  To accommodate the intensity profile of CT scans-where CSF is hypointense-the atlas is modified to assign low intensities to CSF voxels in the atlas  The objective function is used to weigh the probability maps using Bayesian inference principles  The probability maps are deformed to match each subject's brain CT volumes to be segmented  The volumes are segmented by assigning probabilities to each voxels, which specify the likelihood that the voxel belongs to a certain tissue class  Posterior probabilities are computed according to each subject's voxel intensity values  The interface between the resulting GM and WM volumes is smoothed using a nonlinear Wiener filter, subject to the topological constraints of the brain's intensity profile.
 Topology-constrained probabilistic classification refinement is performed [16]   2. MRI data transfer is made to a HIPAA-compliant Amazon Web Services (AWS) online cloud environment 3. Intensity normalization is implemented to account for the potentially confounding effect of intersubject intensity brain differences.
4. The additional rotation of the B matrix in DWI image processing is necessary because resulting orientational information may be affected by the increased head motion which is often associated with DWI, given this modality's relatively long acquisition times [21].
5. In the estimation of confounding factors, the use of the ellipsoid-rather than that of a sphere-is advantageous because it facilitates the evaluation of the local water-diffusion WM anisotropy 6.The anatomically-curated streamline clustering atlas used is described in reference [6].This atlas is suitable because it allows the whole-brain WM to be parcelled into 800 streamline clusters.
Streamline prototyping is important in diffusion MRI tractography, as the latter technique may generate false streamlines which are not representative of real WM connections.Establishing a streamline prototype for an entire cluster allows for a more accurate detection of underlying WM structure and of its trajectory.

2. Speak English fluently 3 .
Be competent to provide informed consent Exclusion criteria for all groups 1.A diagnosis of comorbid neurological or psychiatric disease 2. Being currently nursing or pregnant, or intending to become pregnant 3.Not speaking English fluently 4. Suffering from AIDS 5. Suffering from claustrophobia 6.Having any metal objects in the body which are affected by strong magnetic fields Additional, group-specific inclusion criteria 1.For TBI participants, the volunteer must have experienced one or more TBIs 2. For MCI participants, the volunteer must have reported memory concerns 2.2 Data Acquisition and Archiving MRI acquisition.MRIs are acquired using a whole-body, MAGNETOM Prisma FIT 3T scanner (Siemens Corporation, Erlagen, Germany) with a 60 cm bore diameter.All head scans are acquired using a 20channel head-neck matrix coil.The following are the technical specifications of the Prisma FIT scanner:  magnetic homogeneity of 0.103 parts per million (ppm) within the 40 cm diameter spherical volume of the bore  64 parallel radiofrequency (RF) receiver channels and up to 204 coil elements  dual-channel parallel transmit and slice-accelerated "multi-band" acquisition  A total imaging matrix (TIM) of 204 × 64  A 2x Tesla C2075 graphics processing unit (GPU)  A XR 80/200 gradient system with a maximum amplitude of 80 mT/m per axis and a maximum slew rate of 200 T/m/s per axis  Active shimming using 3 linear and 5 second-order channels  A field of view (FOV) of at most 500 mm and at least 5 mm  A maximum matrix size of 1024 × 1024  A maximum in-plane resolution of 7 µm  Integrated Tx/Rx solid-state DirectRF technology  Syngo MR VD13D software A total of 10 sequences are included in this protocol for acquiring MRI scans.The sequence names and their parameters are listed in Table 1.CT Acquisition.CT scans are acquired using a 16-slice General Electric (GE) BrightSpeed TM scanner.Images are acquired using the following parameters:  width: 512 pixels  height: 512 pixels  mode: helical  slice thickness: 0.6250 mm  slice spacing: 0.6250 mm  kilovoltage peak (kVp): 120 kV  data collection diameter: 250 mm  reconstruction diameter: 223 mm  gantry detector tilt: 0 degrees  rotation direction: clockwise  exposure time: 1397 ms  X-ray tube current: 140 mA  filter type: head filter  generator power: 16800 kW  focal spot: 0.7 mm  convolution kernel type: soft  revolution time: 1 s  single collimation width: 0.625 mm  total collimation width: 10 mm  table speed: 9.3750 mm/s  table feed per rotation: 9.375 mm  spiral pitch factor: 0.9375  pixel spacing: 0.435547 mm MRI/CT Archiving.After acquisition, data are stored in DICOM (digital imaging and communications in medicine) format.They are then de-identified and de-linked to preserve privacy and confidentiality.

1 .
DICOM files are converted into NIFTI format using dcm2niix (https://www.nitrc.org/plugins/mwiki/index.php/dcm2nii:MainPage) 2. Motion correction is performed 3. Voxels beyond the area of the brain are excluded (i.e.imaging volumes are skull-stripped) 4. Image intensities within the brain mask are identified 5.The intensities of voxels within the brain are normalized across subjects by dividing their intensities by their median intensity (see Note 3).

7 . 3 
B 0 images are co-registered to their corresponding T 1 -weighted volumes using a rigid, affine (six-parameter) registration 8. Diffusion tensors are fitted to DWI images 9. FA calculations are performed 10.Deterministic tractography via fixed step-length streamline propagation is used to reconstruct white matter (WM) streamlines using the following parameters:  seed spacing: 0.5 mm  linear measure start threshold: 0.stopping value: 0.17


minimum streamline length: 20 mm  maximum streamline length: 110 mm  stopping criterion: FA value  stopping criterion value: 0.17

 1 
stopping track curvature: 0.96 mm -integration step length: 0.5 mm 11.Streamlines are visualized in TrackVis (trackvis.org)and 3D Slicer (www.slicer.org)If longitudinal MRI/DWI data are available, one can estimate and reduce the extent to which diffusion tractography measures can be influenced by confounding factors, such as subject motion, scanner noise, MRI inhomogeneity differences and other computational errors and inaccuracies.Such estimation is completed through the following technique: Steps.The following steps are implemented to process SWI volumes and identify CMBs: 1. SWI volumes are skull-stripped and co-registered rigidly to their respective T 1 -weighted volumes 2. CMB identification is accomplished using the following algorithms  The image intensity gradient of the skull-stripped SWI volume is computed at each brain voxel using MATLAB  Binary erosion of the gradient magnitude volume is performed using a spherical morphological structuring element (radius = 11 pixels) to remove areas exhibiting sharp gradient changes at the edges of the brain  A gradient magnitude histogram is calculated over all brain voxels. The eroded volume is thresholded to exclude values belonging to the tails of its intensity distribution  Within the thresholded volume, the volume of each connected component consisting of contiguous non-zero voxels is computed  The largest component-associated with voxels on the borders between cerebrospinal fluid (CSF) and WM-is removed  Components associated with voxels within the sulcal banks of the neocortex and on the boundaries between brain and blood vessels are eliminated by removing voxels which were identified as isointense (i.e.within µ ± 2σ of the image intensity distribution of the SWI image)  Remaining non-zero voxel components are assigned CMB labels.
Figure1shows the output produced by the preprocessing of DWI and SWI scans acquired from a sample geriatric mTBI patient.

3 . Notes 1 .
Interaction with the secure database to which the MRI scans were uploaded is facilitated by Flywheel software (flywheel.io).

Figure 1 .
Figure 1.Example output of a DWI/SWI analysis revealing the co-localization of two CMBs (arrows) relative to the arcuate fasciculus (AF) of the two cerebral hemispheres.The AF is a WM structure with substantial involvement in speech production.Two CMBs are depicted as small, red, ovoidal objects located in the spatial vicinity of the AF.Shown are anterior (top row), inferior (middle row) and right (bottom row) views of the cerebral cortex, cerebellum, lateral ventricles and brain stem.Because of their locations, only one CMB is visible in any of the three views.Specifically, one (smaller) CMB is visible in the anterior and lateral views (top and bottom rows), and one (larger) CMB is visible in the inferior view (middle row).T 1weighted images of the cortex are overlaid to indicate the location of the CMBs and AF relative to the anatomy of the brain.WM streamline orientation is color-coded (red: left to right; green: anterior to posterior; blue: inferior to superior).The left column localizes the AF relative to the brain, whereas the right column isolates this WM structure for easier visualization. parameter

Table 1 .
MRI Acquisition Parameters.Listed are selected parameters for eight MRI sequences, as utilized with the acquisition protocol described.Importantly, (A) the TR, TE and TI are fundamental temporal parameters which determine the properties of the MRI signal, whereas (B) pixel width, height, slice thickness and interslice gap determine the spatial resolution of a scan.