Ultra-high-field imaging distinguishes MS lesions from asymptomatic white matter lesions

Subjects.

Twenty-eight patients with clinically definite MS according to internationally recognized criteria were recruited from neurology outpatient clinics at Nottingham's University Hospital between August 2007 and August 2009.¹ Disease classification was as follows: 12 relapsing-remitting, 9 primary progressive, 7 secondary progressive. Two patients with demyelinating clinically isolated syndromes (CIS) were also recruited. In addition, 17 subjects who were found incidentally on clinical MRI scanning to have WM brain lesions were included in the study (referred to as the non-MS group). Five of these subjects were healthy volunteers who were incidentally found to have WM brain lesions. The remaining 12 were patients with multiple vascular risk factors and a clinical MRI scan which revealed incidental WM lesions. All lesions in the non-MS group had been reviewed by a neuroradiologist and none of these subjects were thought to have imaging or clinical features of CNS demyelination. Patient characteristics are shown in the table. Subject numbers were determined by power calculations from preliminary data on 10 patients with MS and 5 patients without MS.

Standard protocol approvals, registrations, and patient consents.

All subjects gave informed consent and the study received ethical approval from the local research ethics committee.

Image acquisition.

Subjects were imaged using a 7 T Achieva scanner (Philips Medical Systems, Best, the Netherlands), equipped with whole body gradients, 16-channel SENSE RF receive coil, and head only volume transmit coil. A 3-dimensional gradient echo sequence was acquired using 200 transverse slices acquired in 4 stacks, each stack overlapping by 10 slices (192 × 164 × 85 mm field of view, 0.5 mm isotropic voxels, echo time = 20 msec, repetition time = 150 msec, and flip angle 14°). A parallel imaging SENSE factor of 2 (right–left direction) and an EPI factor of 3 were employed; acquisition time was 8.8 minutes.

In addition, standard clinical MRI, performed on a 1.5 T or 3 T scanner within 2 years of recruitment to this study, were available for 22 of the patients (8 MS and 14 non-MS).

Image analysis.

All 7 T images were converted to Analyze format and image stacks were merged using in-house software.⁸ All analysis was performed with the primary observer (E.C.T.) blinded to disease status.

White matter lesions in patients with MS and controls were outlined on 7 T T2* magnitude images using a manually defined edge threshold technique. Each lesion was outlined on each axial slice on which it appeared so that lesion volumes could be calculated. Lesions were classified as periventricular (if within one pixel from the ventricle), subcortical (if within one pixel of the cortex), or deep WM.

The 7 T T2*-weighted magnitude images were viewed in orthogonal planes using MRIcro.⁹ For each lesion, the presence or absence of a central vein was noted. Veins were counted if they 1) could be visualized in at least 2 perpendicular planes, 2) appeared linear in at least one plane, and 3) were completely surrounded by hyperintense signal in at least one plane (to avoid the inclusion of adjacent rather than central veins). Lesions were classed as perivenous is they contained one or more central veins.

Where available, clinical T2-weighted MRI were analyzed for whether they met Fazekas criteria (more than 3 lesions plus 2 of the following: one lesion >6 mm diameter, one infratentorial lesion, one lesion abutting the body of the lateral ventricle).¹⁰

Statistics and reproducibility.

χ² was used to compare the frequencies of perivenous lesions between patients with MS and patients without MS. Regression analysis (logit with standard error adjusted for clusters in patients) was used to determine how strongly the perivenous appearance of lesions predicted disease status (MS vs non-MS) when controlling for potential confounders. The dependent variable was disease status (MS or non-MS); independent variables were lesion volume, lesion location, the presence of a central vein, age, and gender. Lesion location variable was assigned 3 categories (subcortical, periventricular, and deep). As a robustness check, in a second regression, the subcortical and periventricular lesions were combined and compared to deep white matter locations in order to mimic more closely what is done clinically. Further analysis using the same method was performed in a subgroup of patients aged ≥50 years (9 MS and 14 non-MS).

It was recognized that there was potential for unblinding of the primary observer to disease status (MS or control) on the basis of lesion distribution. To ensure that lesion classification had not been influenced by unblinding, 20 lesions, selected at random, were cropped such that images of the lesion excluded the remainder of the brain. Using cropped images, lesions were analyzed by a second observer for the presence or absence of a central vein (N.E.; blinded to the original results). Results revealed almost perfect agreement (Cohen kappa coefficient = 0.91, p < 0.001).

For patients with high lesion loads, our method of analyzing every lesion was recognized to be time-consuming. In order to estimate the smallest number of lesions that could be analyzed while maintaining an acceptable accuracy for predicting MS/non-MS, we applied hypergeometric distribution theory. Hypergeometric distribution describes the distribution of probabilities of various outcomes when n items (in our case lesions) are sampled out of a total number of N items, without replacement (each lesion can only be analyzed once). Using this model, we investigated how often sampling 10 lesions (in any patient who had >10 lesions) would provide the correct outcome where ≤40% perivenous lesions was taken to imply non-MS and >40% perivenous lesions was taken to indicate MS.