Introduction

As an important MRI marker of brain small vessel disease white matter hyperintensities (WMH) have been associated with vascular risk factors and cognitive decline (1, 2). WMH have traditionally been assessed with visual rating scales (3,4) or volumetrically quantified with a variety of automatic segmentation methods (for review see 5). A multimodal approach based on combined intensity information from FLAIR, T2w, and T1w may provide a better marker of tissue degeneration and a quantitative measure of WMH severity leading to a greater understanding of the neurobiological impact of WMH on brain function and cognition.

Methods

Subjects: 257 subjects were included in the study, as a subset randomly selected from a larger community-based cohort (age 62-70 years) of the LISA study (6). Data: Conventional structural 3D T1w (MPRAGE), T2w, and FLAIR whole-brain images were acquired on a Philips 3T MRI scanner. Whole-brain diffusion-weighted images (DWI) were additionally acquired. Analyses: WMH were delineated by simultaneous visual assessment of FLAIR and T2w images using JIM 6.0 (Xinapse Systems) resulting in a total of 8845 lesions. Grey and white matter were segmented on lesion-filled T1w images (7), to identify normal appearing white matter (NAWM). For each modality (T1w, T2w, and FLAIR), mean intensity within lesions was normalized to mean NAWM intensity within each subject. On a voxel-wise basis, a singular value decomposition was used to project all voxels within all lesions across all subjects into the 3D parameter space of normalized T1w, T2w, and FLAIR intensities. To describe how each voxel intensity within lesions deviated from NAWM, two distance measures were defined: 1) Parallel distance along the primary eigenvector, and 2) Perpendicular distance from the primary eigenvector. The DWIs were processed within each subject using FSL tools (8), and the diffusion tensor model was fitted to create fractional anisotropy (FA) and mean diffusivity (MD) maps, which were used to validate the distance measures, assuming a more severe lesion is roughly characterized by increased MD and decreased FA (9). On subject level, all lesion maps for all subjects were normalized through the T1w images with SPM DARTEL tool (10) into 1 mm MNI space, and a lesion frequency map was created.

Results

The lesion frequency map of all lesions across all subjects shows a typical spatial distribution of WMH as can be expected in an elderly subject group (Fig. 1). Zooming in on a particular lesion of one subject on the T1w, T2w, and FLAIR images visualize the heterogeneous intensities within a WMH (Fig 2B). This intensity heterogeneity is further stated in the 3D visualization of NAWM-normalized T1w, T2w, and FLAIR lesion values. This suggests a nonlinear trajectory through this parameter space reflecting the diverse neuropathology from NAWM to the state of a particular lesion (Fig. 2C). By combining the two distance measures and examining the points with a high distance value of all lesions, deviation from a linear evolution within lesions can be quantified (Fig. 3), hence the outermost points of the parallel and perpendicular distance measures are capturing the more damaged tissue part within a lesion. A scatterplot further confirms a positive correlation between MD and the two distance measures and a negative correlation between FA and the distance measures (Fig. 4), confirming agreement between lesion severity based on the T1w, T2w, and FLAIR intensities and microstructural properties of the tissue.

Discussion

The appearance of a WMH is known to first be visible on T2w contrasts and subsequently evolve as “black holes” in T1w images reflecting a chronic tissue loss. We here present a new method to characterize and quantify lesion severity based on simple measures of combined conventional T1w and T2w MRI to supplement the traditional measure of WMH volume and number. Quantification of lesion severity may increase our understanding of the impact of WMH on brain structure as well as its functional consequences in e.g. ageing and multiple sclerosis. The prospect of achieving microstructural information from conventional structural images has a great advantage over quantitative maps of e.g. diffusivity or relaxation in terms of higher image resolution, higher SNR, and shorter acquisition times.

Acknowledgements

This work was supported by Nordea Fonden, through the Center for Healthy Aging, University of Copenhagen.

References

1) Garde E, et al Lancet (2000). 2) Salat DH, et al., Neuroscience (2014). 3) Fazekas F, et al., Neurology (1993). 4) Schelten P, et al., Journal of the Neurological Sciences 114, 7–12 (1993). 5) Eugenia M, et al., Neuroinform (2015). 6) Eriksen CS, Garde E, Reislev NL, et al. BMJ Open 2016. 7) Battaglini M, et al Hum Brain Mapp (2012). 8) Andersson JLR et al., NI 125, (2016). 9) Maniega S, et al., Neruobiol. aging (2015). 10) Ashburner J et al., NI 38(1) (2007).