Texture Analysis of T1-Weighted and Fluid-Attenuated Inversion Recovery Images Detects Abnormalities That Correlate With Cognitive Decline in Small Vessel Disease

Abstract

Background and Purpose

MRI may be useful to assess disease severity in cerebral small vessel disease (SVD), identify those individuals who are most likely to progress to dementia, monitor disease progression and act as surrogate markers to test new therapies. Texture analysis extracts information on the relationship between signal intensities of neighbouring voxels. A potential advantage over techniques such as diffusion tensor imaging, is that it can be used on clinically obtained MR sequences. We determined whether texture parameters (TP) were abnormal in SVD, correlated with cognitive impairment, predicted cognitive decline or conversion to dementia.

Methods

In the prospective SCANS study we assessed TP in 121 individuals with symptomatic SVD at baseline, 99 of whom attended annual cognitive testing for 5 years. Conversion to dementia was recorded for all subjects over the 5 years period. Texture analysis was performed on FLAIR and T1-weighted images. The TP obtained from the SVD cohort were cross-sectionally compared with 54 age matched controls scanned on the same MRI system.

Results

There were highly significant differences in a number of TP between SVD cases and controls. Within the SVD population TP were highly correlated to other MRI parameters (brain volume, white matter lesion volume, lacune count). TP correlated with executive function and global function at baseline and predicted conversion to dementia, after controlling for age, sex, pre-morbid IQ and MR parameters.

Conclusions

TP, which can be obtained from routine clinical images, are abnormal in SVD and the degree of abnormality correlates with executive dysfunction and global cognition at baseline and decline over five years. TP may be useful to assess disease severity in clinically collected data. This needs testing in data clinically acquired across multiple sites.

Introduction

Cerebral small vessel disease (SVD) causes lacunar stroke and is the major cause of Vascular Cognitive Impairment and Vascular Dementia [1]. The characteristic radiological features seen on MRI include lacunar infarcts, white matter hyperintensities (WMH), and cerebral microbleeds (CMB), which appear on different MR image types.

The “normal appearing white matter” (NAWM) in these images has been shown to be abnormal using advanced MR techniques, particularly diffusion tensor imaging (DTI) [2, 3]. The degree of diffuse NAWM damage, assessed on DTI, has been shown to correlate more strongly with cognitive impairment than T2-WMH lesion load (WMHL) [4, 5]. However, DTI is not routinely performed in clinical practice, and there is no simple measure that can be used to investigate NAWM damage on clinically acquired MRI.

One tool that could assess NAWM damage using conventional MR scans is texture analysis (TA). TA describes the relationship between the intensities of neighbouring pixels and is particularly suited to tissue such as NAWM where there is only subtle disruption due to disease. This post-processing technique has been used in T1- and T2-weighted images and magnetisation transfer ratio maps [6].

We hypothesised that TA might be sensitive to damage in the NAWM in SVD. Previous studies have shown correlations between NAWM DTI parameters and cognition [3, 7], suggesting that NAWM damage is important in disease progression. Therefore, we applied grey-level co-occurrence (GLCM) matrix based TA to conventional MR images of patients with symptomatic SVD and correlated the derived parameters with temporally equivalent neuropsychological test results. We also determined whether TP were independent predictors of change in cognitive function in the same population and whether they could predict conversion to dementia.

Materials and Methods

Due to subject confidentiality the imaging data are not available to researchers. Summary data and analytical methods are available from the corresponding author on reasonable request.

Subjects

Symptomatic SVD patients recruited to the SCANS study (St George’s Cognition and Neuroimaging in Stroke) were used in this study. This prospective longitudinal study, recruited patients from a geographically contiguous region served by three hospitals in South London, UK [3]. SVD was defined as a clinical lacunar stroke syndrome [8] with an anatomically appropriate lacunar infarct on MRI, as well as confluent WMH (Fazekas grade 2 or more) on MRI [9]. The study was granted ethical approval by the Wandsworth research ethics committee (ref: 07/Q0803/82). All participants gave written informed consent. Participants were fluent English speakers and MRI and cognitive tests were performed at least 3 months post-stroke to minimise acute effects of stroke on cognition. Subjects underwent MRI and cognitive testing at baseline, then, of interest in this study, cognitive testing yearly for five years, 121 underwent MRI and cognitive tests at baseline; 99 of these attended for at least one follow-up visit, details of the number of subjects at each time-point can be found in the online supplement at http://stroke.ahajournals.org.

A control population, imaged on the same MR scanner using the same sequences (for the last time-point of the study only, which is used here [3]), comprised randomly selected community dwelling subjects recruited to the GENIE study (St George’s Neuropsychology and Imaging in Elderly) [10], these subjects had no history of psychiatric and CNS diseases (including stroke). There were data from 57 subjects available with (mean (± SD) age=70±9 years, 35 male), however, due to excessive movement or failure to tolerate the MRI protocol, data from 54 subjects was used.

Clinical and cognitive performance assessments (SVD group only)

All subjects had cerebrovascular risk factors recorded including hypertension (systolic BP of >140 mmHg, diastolic BP >90 mmHg or on treatment), diabetes mellitus (on drug or insulin treatment), hypercholesterolaemia (random total cholesterol of >5.2 mmol/L or on treatment), body mass index and smoking history (current, ex-smoker and never).

Cognitive assessment was performed by a neuropsychologist using a battery of widely used tasks chosen to characterise the cognitive impairment seen in SVD [3, 11], including: digit-span, logical-memory, visual-reproduction, BIRT memory and information processing battery, speed of information processing, digit-symbol, grooved pegboard, trail-making test, SL-verbal fluency, modified Wisconsin card-sorting test. The following cognitive indices were constructed: Executive Function (EF), Processing Speed (PS), Working Memory and Long Term Memory. A Global Cognition (GC) index was produced which summarised performance on all tasks. Premorbid IQ was estimated using the restandardised National Adult Reading Test (NART) [12]. Measures for each task were transformed into z-scores through psychometric standardisation using age-scaled normative data and averaged to calculate cognitive index scores.

Missing data from tasks which the subject was unable to complete due to the effect of cognitive decline, as judged by the attendant neuropsychologist, was addressed by substituting the minimum scaled score for that task (0 or 1; corresponding to z-scores of -3.33 and -3). Other missing data was excluded from the calculation of cognitive index scores.

The slope of change in the cognitive data over five years was determined using linear mixed effects models to determine annualised change [11], the cognitive parameters were modelled as a function of time allowing variation by fixed and random effects to determine linear trajectories and intercepts. The Wald test was used to determine whether there was a detectable change in the parameters over time for the average fixed effect slope for the group. Slopes were then estimated for each subject for the parameters showing significant change over time.

Conversion to dementia

Information on conversion to dementia over the 5 year follow-up was available for all 99 patients studied longitudinally. Dementia was diagnosed using the “Diagnostic and Statistical Manual of Mental Disorders 5” (DSM-5) [13] definition of “major neurocognitive disorder”, and was present if individuals met one of:

1. A clinical diagnosis of dementia.

2. Review of medical records and cognitive assessments by a neurologist and clinical neuropsychologist, blinded to MRI and risk factor information, who agreed that the clinical picture met DSM-5 criteria.

3. An MMSE score consistently <24, indicative of cognitive impairment and reduced capabilities in daily living as measured by a score ≤7 on the instrumental Activities of Daily Living (iADL) [14]

Date of dementia onset was defined as the date of diagnosis or the mid-point date between the visit at which the diagnosis was established and the previous visit.

MRI protocol

MRI was performed on a 1.5-T General Electric Signa HDxt MRI system (General Electric, Milwaukee, WI, USA). The full MRI protocol has been described in detail [3].

Sequences of interest to this study were:

(1) Axial Fluid Attenuated Inversion Recovery (FLAIR): TR/TE/TI=9000/130/2200ms, field of view (FOV)=240x240mm², matrix=256x192, 28 contiguous 5 mm slices.

(2) Coronal spoiled gradient recalled echo T1-weighted (SPGR): TR/TE=11.5/5ms, FOV=240x240mm², matrix=256x192, Flip Angle=18º, 176 contiguous 1.1mm slices.

Pre-processing and TA

Lesions were marked on the FLAIR images by an experienced rater, using the semi-automated lesion marking tools in DispImage [15]. The T1-weighted images were segmented using SPM8 (Wellcome Department of Cognitive Neurology, UCL Institute of Neurology, London, UK), the white matter (WM) tissue probability map was lower thresholded at 90% to produce a WM segment. The T1-weighted images were registered to the FLAIR images using FLIRT [16] and the transformation applied to the WM segment. The lesion mask and WM segment were then applied to the FLAIR images to leave NAWM. The FLAIR images were registered to the T1-weighted images and the transform applied to the lesion masks to produce a T1-weighted NAWM map. For both registrations 6 degrees-of-freedom were used to provide rigid-body transformation between the T1-weighted and FLAIR images.

The FLAIR and T1-weighted NAWM images were analysed using the method proposed by Haralick [17], in which 14 TP are defined based on the GLCM: Angular Second Moment (ASM), Contrast, Correlation, Variance, Inverse difference moment, Sum average, Sum variance, Sum entropy, Entropy, Difference Variance, Difference entropy, Information measure of correlation 1, Information measure of correlation 2, Maximal correlation coefficient (MCC) (for definitions please see online supplement at http://stroke.ahajournals.org).

Due to the anisotropic voxel dimensions of the FLAIR images and to maintain methodological consistency between the image types 2D TA was performed. The NAWM maps were normalised to a specified number of grey levels using histogram equalisation. Based on the number of voxels in the NAWM segments and the number of grey levels in the native images, and to avoid issues with sparsity in the GLCM the FLAIR images were normalised to 64 grey levels and the T1-weighted images to 128 grey levels. The GLCM and TP were calculated for the NAWM.

Data Analysis

For analysis the nLAC and nCMB were log 10 transformed and the T2-WMHL was expressed as a percentage of the total brain volume and log transformed.

Initially TP from the NAWM were compared between the baseline SVD and control images using the Student’s unpaired t-test.

Secondly, in the SVD cohort, the TP were correlated, using the Pearson Correlation test, with normalised brain volume (NBV), WMHL, nLAC, nCMB and peak height from mean diffusivity NAWM (MDPH) histograms as determined previously to be the best MR derived predictors of cognition [3].

A regression analysis was then performed to determine whether TP were independently predictive of the cognitive measures. As in [3] univariate regressions were run for the TP, including age, sex and premorbid IQ as confounds. For each image type the variable showing the greatest standardised beta (with associated p-value ≤0.05) was chosen to progress to the multivariate model. These were included in a multivariate regression with the parameters listed above, initially only parameters derived from conventional MR images were added, then a further model including MDPH was run. The regression tests were then repeated using the rates of change of the cognitive measures over the five year follow-up period.

Finally the ability of TA to distinguish those of the SVD subjects who developed dementia was determined. Initially univariate Cox regression was used to determine those TP which best predicted conversion to dementia (corrected for age, sex and pre-morbid IQ), these were then included in multivariate analyses as above, with addition of binary parameters indicating whether new microbleeds and/or lacunes are seen in follow-up scans, which were acquired for the first 3 years of the follow-up, to assess whether they are independent markers of conversion. To counter the possibility of overfitting the model bootstrapping with 1000 samples of the population was applied.

Results

Subjects excluded and tissue segments

Tissue segmentations were unreliable in 3 SVD subjects and these were removed. Therefore, for the texture analyses 118 subjects were included in the analysis. In a further 3 subjects artefacts prevented reliable analysis of the DTI data, so for the regression including the diffusion parameters 115 subjects were included. Demographic details and MRI characteristics are shown in table 1.

Table 1.

Group demographics and MRI characteristics for the SVD and control groups and the results of the group comparisons

	SVD	Controls	p-value
Age (mean±SD) years	70.01±9.75	70.36±9.18	0.82
Sex male (number, %)	78(65)	35(62)	0.25
Hypertension (number, %)	112(93)	28(49)	<0.0001
Systolic Blood Pressure (mean±SD) mmHg	146.8±21.5	138.5±18.0	0.013
Diastolic Blood Pressure (mean±SD) mmHg	81.0±10.8	79.3±12.3	0.36
Cholesterol (mean±SD) mmol/l	4.33±0.90	5.67±1.13	<0.0001
Diabetes Mellitus (number, %)	24(20)	0(0)	<0.0001
Smoker (number, %) current or ex	55(46)	32(56)	0.185
BMI (mean±SD) kgm-2	27.1±4.9	25.2±3.9	0.016
Lesion volume (ml)	31.87±26.97	8.7±11.73	0.0001
Normalised brain volume (ml)	1295.1±91.1	1337.3±87.3	0.005
Lacunes	4.26±5.48	0.65±1.51	<0.00001

Cognitive change over the study period

Details of the cognitive changes over the five years have been previously described [11], but in summary, PS, EF and GC scores declined significantly over the five year follow-up period. The annualised changes were -0.048±0.015 (p=0.001) for EF, -0.052±0.014 (p<0.001) for PS and -0.029±0.009 (p<0.001) for GC. It was, therefore, decided to proceed with the analysis using only these cognitive parameters.

Cross-Sectional Analysis

i. Comparison of TP in NAWM between SVD cases and controls

There were highly significant differences between SVD cases and controls for many TP derived from both T1 and FLAIR sequences (table 2). The pattern of differences differed for the image types; only the Sum Entropy had p-values of less than 0.0001 for both image types. The p-values are not corrected for multiple comparisons due to shared variance between the measures. For reference, an uncorrected p-value of 0.004 would give a Bonferroni corrected p-value of 0.05.

Table 2.

The texture parameters for the SVD and control groups for T1-weighted and FLAIR images and the p-value from the independent samples t-test comparison. The p-values shown aren’t corrected for multiple comparisons, for reference the corrected p-value is 0.004.

	T1			FLAIR
Texture parameter (mean±SD)	SVD	Control	p-value	SVD	Control	p-value
Angular Second Moment	0.000115±0.000026	0.000121±0.000025	0.18	0.00148±0.00014	0.00162±0.00018	6x10-8
Contrast	1485±424	1164±269	9x10-7	42.5±7.3	40.1±6.1	0.044
Correlation	0.35±0.14	0.475±0.108	5x10-8	0.937±0.011	0.941±0.009	0.049
Variance	1131±91	1105±71	0.066	338.5±1.5	338.1±1.8	0.121
Inverse difference moment	0.0439±0.0062	0.0492±0.0075	3x10-6	0.265±0.017	0.273±0.016	0.002
Sum average	147.5±9.0	147.6±8.0	0.913	65.21±0.29	65.36±0.20	0.0006
Sum variance	3040±302	3257±288	2x10-5	1312±10	1312±9	0.662
Sum entropy	5.366±0.043	5.394±0.030	3x10-5	4.838±0.005	4.833±0.010	9x10-5
Entropy	9.286±0.133	9.259±0.126	0.178	6.95±0.12	6.84±0.14	8x10-7
Difference Variance	565±124	461±90	1x10-7	22.10±3.67	21.11±3.20	0.091
Difference entropy	4.353±0.122	4.24±0.122	8x10-8	2.59±0.083	2.56±0.074	0.033
Information measure of correlation 1	-0.0509±0.0197	-0.0557±0.0202	0.151	-0.283±0.020	-0.295±0.018	0.0004
Information measure of correlation 2	0.608±0.077	0.630±0.072	0.075	0.948±0.009	0.951±0.008	0.007
Maximal Correlation Coefficient	0.718±0.125	0.711±0.102	0.748	0.947±0.010	0.952±0.010	0.002

ii. Correlation between TP and other MR measures

The TP were highly correlated to a number of MR measures (table 3), suggesting abnormalities in a number of different parameters underlie the changes in TP.

Table 3.

The correlations of the texture parameters and MRI measures. The top row in each box is the Pearson correlation coefficient, with the p-value shown below in italics. Only significant correlations are shown.

	White Matter Lesion Load (% log)	Normalised Brain Volume	Mean Diffusivity Peak Height	Lacune Number (log)	Microbleed Number (log)
T1
Contrast		-0.2 0.032
Inverse difference moment		0.2 0.037
Sum Average					-0.25 0.009
Maximal Correlation Coefficient	-0.24 0.012	0.31 0.001	-0.35 0.0002	-0.27 0.004	-0.37 0.00005
FLAIR
Angular Second Moment	-0.34 0.0003	0.29 0.002	0.26 0.005		-0.23 0.016
Sum average	-0.6 1.5x10-12		0.34 0.0003	-0.19 0.043
Sum entropy	0.38 0.00003	-0.36 0.00009	-0.27 0.003	0.2 0.036
Entropy	0.36 0.0001	-0.35 0.0001	-0.3 0.001	0.19 0.04	0.23 0.013
Difference variance	-0.21 0.028

iii. Regression between TP and cognition in the SVD cases

As with previous work [3] only cognitive fields shown to change over the course of the study, and GC, were included in the analysis.

For the univariate analysis in the EF the T1-MCC (standardised beta=0.182, p-value=0.009) and the FLAIR-Sum Average (0.196, 0.005) were the only significant regressors, only FLAIR-Sum Average significantly regressed with PS and for GC T1-MCC (0.188, 0.006) and FLAIR-Sum Average (0.24, 0.0005) were the most significant regressors. Results of the multivariate regression are shown in table 4, FLAIR-Sum Average, WMHL, nLAC, and NBV associated significantly with EF, with associations being strongest for the texture parameter, nLAC and WMHL. There was little difference whether MDPH was included in the model.

Table 4.

The results of the multivariate regressions with the significant regressors in bold. The top line contains the standardised beta coefficient and the lower line the associated p-value in italics. Results are shown both with and without the inclusion of MDPH.

	Executive Function		Processing Speed		Global Cognition
	Standardised Beta p-value		Standardised Beta p-value		Standardised Beta p-value
	No Mean Diffusivity Peak Height	With Mean Diffusivity Peak Height	No Mean Diffusivity Peak Height	With Mean Diffusivity Peak Height	No Mean Diffusivity Peak Height	With Mean Diffusivity Peak Height
Normalised Brain Volume	0.153 0.036	0.1 0.174	0.275 0.001	0.23 0.012	0.234 0.001	0.181 0.014
White Matter Lesion Load	0.254 0.007	0.308 0.002	-0.091 0.407	-0.04 0.725	0.112 0.206	0.172 0.059
Lacune Number	-0.228 0.003	-0.204 0.007	-0.214 0.016	-0.195 0.040	-0.19 0.009	-0.016 0.027
Microbleed Number	-0.081 0.308	-0.059 0.469	-0.004 0.968	0.011 0.903	-0.061 0.42	-0.043 0.578
Mean Diffusivity Peak Height		0.178 0.051		0.147 0.167		0.191 0.028
FLAIR–Sum Average	0.26 0.001	0.245 0.002	0.032 0.732	0.018 0.852	0.226 0.003	0.206 0.006
T1–Maximal Correlation Coefficient	0.08 0.241	0.055 0.425			0.913 0.418	0.023 0.72

Only nLAC and NBV independently correlated with PS. For the univariate regressions no T1-weighted texture parameter was significant and FLAIR-Sum Average was the most significant texture parameter (p=0.05), but this did not survive inclusion in the multivariate model. Again the addition of MDPH had little effect on the model.

The GC results were similar to those of EF, with NBV and nLAC being significant regressors along with the FLAIR-Sum Average, in this case the MDPH also reached significance, but did not change the results seen from the other parameters in the model.

Longitudinal analysis

Do TP predict cognitive decline?

For the univariate regressions of the TP with the rate of change of the cognitive fields over five years T1-MCC (standardised beta=0.22, p-value=0.024) and FLAIR-Entropy (0.312, 0.002) were significant regressors of EF, and the same parameters appeared for regression with GC (T1-MCC (0.205, 0.043), FLAIR-Entropy (-0.28, 0.005)), no TP had a significant regression with PS. When the other MR parameters are included none of the TP remain significant, the closest to significance is for the change in EF without the MDPH where FLAIR-Entropy a standardised beta of -0.208 and a p-value of 0.058. In addition nLAC and NART are significant for the GC change, while nLAC is significant for EF change.

For all regression models the variance inflation factor was calculated and found to be less than 3 in all cases, suggesting no significant impact of multicollinearity.

Conversion to Dementia

Twenty of the 99 (20%) patients developed dementia during follow-up. Prediction of dementia was assessed by the Cox regression, On univariate regression only one T1 image texture parameter was a significant predictor of conversion to dementia; the T1-MCC (p=0.042, hazard-ratio (HR)=0.025). Several parameters from the FLAIR images predicted conversion to dementia, the best being FLAIR-Sum Average (p=0.002, HR=0.106). In the multivariate bootstrapped model the FLAIR-Sum Average retained significance (p=0.012, HR=0.014) along with NBV (p=0.001, HR=0.97), and when MDPH was also added the same two parameters were significant in predicting dementia (FLAIR-Sum Average, p=0.006, HR=0.008, NBV, p=0.001, HR=0.971).

Discussion

In this study we demonstrated that TA, using standard MR sequences, detects abnormalities in NAWM structure in SVD compared with aged matched controls. Furthermore TP correlated with EF and GC at baseline and predicted conversion to dementia, after controlling for age, sex, pre-morbid IQ and other MR parameters. Of particular interest in this study are entropy, MCC and Sum Average. Entropy describes the randomness of the grey level pattern, the more random the pattern of grey levels in the image the greater the entropy. The sum average measures the sum of the diagonals in the GLCM, the more variation in the number of signal pairs the greater the value. Lastly, the MCC is the second principal component of the correlation matrix of rows of the GLCM, it identifies variance not accounted for by the correlation parameter.

Our results are consistent with studies using different MR sequences, particularly DTI [3, 5, 18], which have demonstrated that the NAWM in patients with SVD is not normal, and its structural integrity correlates with cognitive impairment. What ultrastructural changes each of the individual TP are measuring is not clear, and it is likely that individual TP relate more closely to different underlying aspects of the pathology. However, within the SVD cohort we found strong correlations between TP and brain volume, nLAC, WMHL and atrophy suggesting that the texture abnormalities are related to a number of different pathologies. These results support those seen in a recent paper where a relationship between TP with ‘SVD score’ was seen [19].

We found specific TP predicted EF and GC at baseline and prospectively with longitudinal decline in both fields, although this did not survive in the multivariate model. In contrast TP did not correlate with PS or predict change in this parameter. This difference in relationships between executive dysfunction and PS is consistent with previous analysis of this dataset in which diffuse white matter changes were more closely related to EF [11], while lacunar infarcts possibly acting through disruption of distributed brain networks were more closely related to impairment in PS [20].

While TP did correlate with cognition at baseline and predict decline in cognition, such correlations were only moderate and less strong than those with DTI parameters [11] where MDPH was a significant predictor of cognitive decline in the multivariate model. Furthermore TP did predict conversion to dementia, however the relationship was only of a similar order to the diffusion parameters which also showed a clear relationship with conversion to dementia (p=0.009, HR=0.002) [11]. Also, the small number of subjects who converted to dementia means that the Cox regression may still be optimistic, despite the use of bootstrapping. Thus, these results should be treated accordingly.

These results suggest that TA may be less sensitive to changes in NAWM ultrastructure than DTI, in particular those predicting cognitive decline. However TA has an advantage, particularly for large clinically acquired datasets, in that it can be performed on standard MRI sequences acquired during routine clinical scanning. TA produces a large number of highly correlated parameters. In this study we selected those which were most strongly associated on univariate analysis to take through to multivariate analysis. Further replication studies are required to confirm that the parameters which we prioritised are most strongly associated with cognitive function in SVD in other populations. The third criterion for diagnosing dementia is not as clinically definite as the former two, this may lead to a misdiagnosis of some subjects, however in the absence of a definite diagnosis it was considered that those subjects showing this significant level of impairment should be considered as having dementia.

A major strength of this study is its longitudinal design with data on progression of cognitive decline over a 5 year period. A further strength is the inclusion not only of TP but other MR parameters so that the independent contribution of TP to disease prediction could be determined. The SCANS cohort included patients with relatively homogenous SVD in that all had presented with lacunar stroke and all had confluent WMH. This is both a strength, and a weakness of the study. However further studies in cohorts of patients with different severity of SVD are required to determine whether these changes can be seen across the spectrum of SVD severity.

SVD presents a major health problem and there are few proven treatments to prevent its progression. Recurrent stroke rates are relatively low, and it has been shown that neuropsychological testing is relatively insensitive to change both due to the slow rate of change and due to practice effects reducing the sensitivity to change [21]. This has led to increasing interest in the use of MRI techniques to both monitor disease progression, and test new therapeutic approaches in smaller phase 2 studies prior to large phase 3 trials with clinical end points [22]. MRI markers, particularly DTI, have been shown to be potential surrogate markers being both sensitive to change over short time periods, and able to predict future cognitive decline and dementia [3, 11]. However, despite an increase in availability DTI is not acquired in all clinical MRI settings, and takes additional time and image processing, and can present challenges in multi-centre studies. As TA can be performed on conventional MRI data it has potential advantages in these situations. Our results suggest that TP may be useful markers of disease in SVD, but further studies are required to replicate these results and determine how TA performs in cohorts with multi-centre imaging.