Leukoaraiosis

As a prerequisite for producing and analyzing tract diffusion profiles, we first assessed the reliability of the automated tract segmentation algorithm in identifying the tracts. We reasoned that the algorithm should produce consistent results if multiple scans were obtained for the same individual, akin to test-retest reliability in clinical assessment. Our DWI protocol included four independent repeats of a 30-direction DWI sequence. For each subject, we divided the data into two sets; we averaged scans 1 and 2 as set 1, and scans 3 and 4 as set 2. We then processed each data set with AFQ and extracted the mean FA value for each tract in each individual for the two independent scan sessions. We computed the scan-rescan reliability independently for each tract and found that the median correlation for FA values for each tract from set 1 and set 2 was r = 0.93 with a standard deviation of 0.07. This result demonstrates that the measurements generated by AFQ are highly reliable within an individual across scan sessions. Also note that the correlation reported here represents the reliability of the AFQ analysis for a DWI sequence with 2, 30-direction data sets averaged together rather than the full sequence that we typically use which averages 4, 30-direction data sets. The scan rescan reliability would be even higher if all 4 scans were averaged together.
As a more demanding measure of reliability, we then compared the mean FA of tracts obtained by two methods–manual segmentation, considered the gold standard for tract identification, (Wakana et al. 2007) and AFQ tract identification. For this analysis we selected six tracts: left and right inferior frontal-occipital fascicle, left and right uncinate fasciculus and left and right superior longitudinal fasciculus. To test the automated method in a clinical sample, we assessed the degree of correlation between tract mean FA measurements from the manual and automated methods in the preterm children. These patients had a range of white matter abnormalities on conventional MRI scans ranging from normal to severe injury, including 3 with severe ventricular dilitation [32] . Correlations between the manual and automated methods were very high for each tract. The median correlation between the FA values obtained from the two methods was r = 0.98 with a standard deviation of 0.04. Figure 9 shows the tract mean FA values obtained from manual segmentation plotted against the values from the AFQ automated segmentation. For nearly every subject the values lie on the identity line demonstrating near perfect correspondence between the methods. Hence The AFQ automated fiber tract segmentation is consistent with the time-consuming manual techniques that have served as the gold standard.
The subject that shows a discrepancy between the manual and automated methods for the right uncinate fasciculus has severe ventricular dilitation. For this subject the automated uncinate ROI placement was imperfect due to extremely abnormal brain shape. Most of the fiber tract segmentations were accurate for these severely abnormal brains, however it is important to manually inspect the ROIs and resulting fiber groups for patients with severe abnormalities because misalignment is possible.

Consecutive patients with SVD were recruited to the St George's Cognition and Neuroimaging in Stroke (SCANS) study from stroke services in three hospitals covering a geographically contiguous area of South London (St George's Hospital, King's College Hospital and St Thomas's Hospital). SVD was defined as a clinical lacunar stroke syndrome [16] (link) with an anatomically appropriate lacunar infarct on MRI, as well as confluent leukoaraiosis (Fazekas grade 2 or more) on MRI [17] (link). All patients were fluent in English to allow neuropsychological testing. Exclusion criteria were: any cause of stroke other than SVD including extra or intracranial arterial vessel stenosis >50%; any cardioembolic source; cortical infarcts; subcortical infarcts >1.5 cm in diameter as these (striatocapsular type infarcts) are often due to embolism; other major central neurological system disorders; major psychiatric disorders (except depression); any cause for white matter disease other than SVD. Individuals with contraindications to MRI including claustrophobia were excluded.
The study was granted ethical approved by Wandsworth REC. 180 patients were screened of whom 137 volunteered to participate and gave written informed consent. 121 of the 137 SVD patients completed the protocol. Of non-completers, 6 withdrew due to the length of the neuropsychology examination, 2 could not complete MRI, 6 became unwell between consenting and testing, and 2 were found to meet exclusion criteria after consent, 1 due to narcolepsy and 1 due to schizophrenia.
All cognitive testing and MRI was performed at least 3 months post-stroke to minimise acute effects of stroke on cognition.

We performed a quantitative DTI analysis of white matter microstructure alterations in POE. We selected white matter regions of interest (ROI) a priori that have been implicated in functional outcome and cognition (corpus callosum and external capsule). As we have performed previously (39 (link), 42 (link)–47 (link)), ROIs were traced by an observer masked to experimental conditions and analyzed using Bruker's Paravision 6.1 imaging software (Billerica, MA). In brief, fractional anisotropy (FA), axial diffusivity (λ₁), and radial diffusivity $\left(\frac{{\lambda }_2+{\lambda }_3}{2}\right)$  scalar maps were computed, and means were calculated individually for each ROI. For bilateral neuroanatomical ROIs, scalar means were acquired on each side and averaged per ROI. Two scans (both in the Saline group) were excluded from analysis—one due to poor field of view coverage and one due to severe motion-related artifact.

All patients were offered investigation with EEG (available from N = 60), extended MRI sequences of the neurocranium (available from N = 57), and ECG (available from N = 60). EEGs were examined by the respective ward clinician and retrospectively assessed for regional or focal slowing, intermittent generalized delta/theta activity (IRTA/IRDA), and epileptic activity. Independent component analysis (ICA) of EEGs with automatic calculation of IRDA/IRTA density was additionally performed as previously described [31 (link)]. MRI included the following sequences on a 3 Tesla scanner (MAGNETOM Prisma, Siemens Healthcare GmbH, Erlangen, Germany) in most patients: T1-weighted sequences with magnetization-prepared rapid gradient echo (MPRAGE) with isotropic 1-mm³ voxels for atrophy diagnostics, diffusion-weighted imaging (DWI) with axial 5-mm slices for stroke detection, fluid-attenuated inversion recovery (FLAIR) sequences with isotropic 1-mm³ voxels for the detection of signal alterations, and further innovative analyses such as diffusion tensor imaging (DTI) and pseudo-continuous arterial spin labeling. All MRIs were assessed and evaluated by experienced senior neuroradiologists. MRI abnormalities were categorized as white-/gray-matter alterations, atrophy, vascular changes, cysts, tumors, and anatomical variants, among others. An automated volume- and region-based approach (https://www.VEObrain.com) was used with the MPRAGE sequences for fully automated whole-brain volumetry for the detection of volume loss (VEOmorph, Freiburg, Germany). Selected cases (N = 7) with abnormalities in routine diagnostic work-up suggestive of an autoimmune cause were examined with cerebral [¹⁸F]fluorodeoxyglucose positron emission tomography (FDG-PET). Two patients received exome analysis due to suspected syndromal genesis.

The variables address the predictors of PSCI at three months; these include Age (in years), sex, alcohol use, smoking history, history of diabetes mellitus, dyslipidemia, atrial fibrillation, post-stroke depression, apathy, stroke type and characteristic (haemorrhagic/ischaemic, cortical/sub-cortical), stroke (infarct/hematoma) volume, presence of Leukoaraiosis or brain atrophy (See Table 3 for listing and a brief explanation of the variables for the Aim 3).