Lesion detection at seven Tesla in multiple sclerosis using magnetisation prepared 3D-FLAIR and 3D-DIR
- First Online:
- Cite this article as:
- de Graaf, W.L., Zwanenburg, J.J.M., Visser, F. et al. Eur Radiol (2012) 22: 221. doi:10.1007/s00330-011-2242-z
- 908 Downloads
To examine the feasibility and value of 7 T 3D T2-weighted Fluid Attenuated Inversion Recovery (FLAIR) and Double Inversion Recovery (DIR) MR sequences for lesion detection in multiple sclerosis (MS).
High-resolution 3D-FLAIR and 3D-DIR MR sequences at 7 T were obtained using magnetisation preparation (MP), and compared with 2D-T2-weighted and 3D-T1-weighted sequences in 10 MS patients and five healthy controls. We determined contrast ratios and counted lesions according to anatomical location.
MR imaging at 7 T was safe and allowed multi-contrast imaging within clinically acceptable imaging times. Lesion to white matter (WM) and grey matter (GM) contrast ratios were higher in 3D-MP-FLAIR and 3D-MP-DIR compared with 2D-T2 and 3D-T1. Cortical (mixed+intra-cortical) and total lesion counts were 97/592 on 3D-MP-FLAIR and 100/558 on 3D-MP-DIR compared with 84/384 on 2D-T2 and 42/442 on 3D-T1. More juxta-cortical lesions were seen with 3D-MP-FLAIR (205) and 3D-MP-DIR (133) than with 2D-T2 (125) and 3D-T1 (70). Finally, higher numbers of lesions were found for deep WM lesions: 176 for 3D-MP-FLAIR and 196 for 3D-MP-DIR vs. 155 for 2D-T2 and 131 for 3D-T1.
Near isotropic 3D-MP-FLAIR and 3D-MP-DIR allows high quality T2-weighted MR imaging in MS at 7 T, improving (cortical) lesion detection.
• Magnetization prepared 3D-FLAIR and 3D-DIR 7 T MRI provide high quality isotropic images.
• MS lesions are well demonstrated by 3D-MP-FLAIR and 3D-MP-DIR at 7 T MRI.
• 3D-MP-FLAIR and 3D-MP-DIR at 7 T MRI show many more Virchov-Robin spaces.
KeywordsMagnetic Resonance Imaging/methods Imaging Three-Dimensional Multiple Sclerosis/diagnosis Multiple Sclerosis/pathology Cerebral Cortex/pathology
Multiple Sclerosis (MS) is a chronic immune-mediated inflammatory disease of the central nervous system (CNS) that causes demyelination of axonal myelin sheaths and neurodegeneration. MR imaging is a valuable diagnostic tool that is able to non-invasively show demyelinating lesions in the brain and spinal cord. Pulse sequences like fluid attenuated inversion recovery (FLAIR) and double inversion recovery (DIR) suppress the signal of cerebrospinal fluid (CSF), while DIR also attenuates the signal of white matter (WM), and have proved to be more sensitive for brain lesions in WM and grey matter (GM) than proton density (PD) or T2-weighted (T2w) and T1-weighted (T1w) MR sequences [1, 2, 3, 4, 5] at 1.5 T and 3.0 T.
While the emphasis in MS has long been directed mainly at WM, there is an increasing interest for GM abnormalities: In vivo imaging of cortical lesions has become feasible and relations between GM damage and clinical manifestation have been shown [6, 7, 8]. Detection of WM and especially GM lesions requires a high spatial resolution and a good signal-to-noise ratio (SNR) as most of the GM lesions are not of the inflammatory type. This all should be done within a clinically acceptable imaging time [4, 9, 10]. Unfortunately, the additional 180-pulses in FLAIR and DIR sequences also lower the signal intensity of the non-suppressed tissues. Therefore, MR imaging at high field strength (7 T) might be favourable as it provides an intrinsically higher SNR that can be used for higher spatial resolution and/or for faster imaging by using higher parallel imaging rates . MR examinations at 3 T detected more lesions than at 1.5 T, although the added value for the diagnosis MS remains uncertain . Also treatment monitoring may benefit from possible increased sensitivity as new lesions may be seen earlier. Finally, detection of (small) GM lesions may help to increase correlation with the clinical status of the patient. To date, most clinical 7 T studies with MS patients have been confined to the T1w, PDT2w, and T2*w protocols [13, 14, 15, 16]. So far, 3D-FLAIR and 3D-DIR, both showing increased cortical lesion detection at standard MR field strength (1.5 T and 3 T), could not easily be implemented at 7 T. This is largely because acquisition of 3D-FLAIR and 3D-DIR is challenging at high field strength, owing to a number of issues like their sensitivity to heterogeneity in the amplitude of the transmit RF field (B1), the high specific absorption rate (SAR), long T1 relaxation, and lengthy repetition times. Consequently, this leads to long acquisition times in the standard implementation of these sequences . An initial experience of DIR imaging in healthy volunteers at 7 T and 3 T demonstrated some advantage of 2D-DIR at high field strength, but showed unsatisfactory results for 3D-DIR . 2D-FLAIR was also tested but used a slice thickness of 4 mm . It needed 16 min of scan time to cover the brain as for each slice a separate adiabatic inversion pulse needs to be transmitted while using long repetition times to compensate for long T1 relaxation times and to cover SAR limitations. Recently, a 3D magnetisation prepared FLAIR (3D-MP-FLAIR) sequence was described and optimized for use at 7 T . In a clinically acceptable acquisition time this 3D-MP-FLAIR sequence gives the desired contrast in healthy subjects.
The aim of this explorative study was to examine the feasibility and value of new 3D T2-weighted MP-FLAIR and MP-DIR sequences at 7 T MR imaging for lesion detection in MS patients. The potential of these sequences to detect lesions in WM and GM was explored and compared with conventional 2D-PDT2 and 3D-T1-weighted sequences at 7 T in the same MS patients.
Material and methods
Patients and healthy controls
Ethical approval for the study was received from the institutional ethical review board of the hospital and all subjects gave written informed consent before the examination. Ten MS patients (most with the primary progressive disease type) from our Neurology outpatient clinic and five healthy controls were consecutively included in this prospective study. Inclusion criteria were clinically definite MS according to the revised ‘McDonald criteria’ of 2005  and age between 18 and 60 years. Exclusion criteria were other neurological diseases (patients as well as healthy volunteers) and recent relapses (<3 months). Local high field MR safety regulations were applied that on top of the standard field regulations exclude all subjects with metal objects in or on the body and also exclude subjects with suspicion of presence of metal objects that were possibly applied during surgical interventions. The use of disease modifying treatment was not an exclusion criterion. No contrast agents were used in this study. Patients (3 men, 7 women) had a mean age of 48 years (range 43–53) and an Expanded Disability Status Scale (EDSS) median of 4.5 (range 3.5–7.5). Healthy control subjects (3 men, 2 women) had a mean age of 44 years (range 33–52). After imaging each subject was asked about any discomfort other than that experienced at lower field strength.
Sequence parameters used in the imaging protocol
TE (1/2) [ms]
21 / 80
TI (1/2) [ms]
3,150 / 550
Flip Angle [deg]
Echo Train Length [−]
250 × 250 × 190
250 × 250 × 190
250 × 200 × 190
230 × 200 × 120
312 × 304 × 238
252 × 250 × 238
248 × 312 × 238
328 × 227 × 60
2.5 × 2.5 (APxRL) *
2.0 × 3.0 (APxRL)*
Acq. Res [mm3]
0.8 × 0.8 × 0.8
1.0 × 1.0 × 0.8
0.8 × 0.8 × 0.8
0.7 × 1.0 × 2.0
Recon. Res. [mm3]
0.45 × 0.45 × 0.4
0.5 × 0.5 × 0.4
0.5 × 0.5 × 0.4
0.45 × 0.45 × 2.0
Acq. Time [min:sec]
Sagittal 3D images were reconstructed in the axial plane, corresponding to the 2D-PDT2 images. The appearance of normal and non-normal tissue (contrast of GM and WM, focal lesions, diffuse lesions, diffusely appearing WM and GM and suppression of CSF) was evaluated. A 6-level score was used to address motion artefacts and homogeneity of the images: Motion Artefacts: 0: none, 1: almost not noticeable, 2: incidental motion, 3: some artefacts, 4: multiple artefacts, needs careful analysis, 5: high numbers of artefacts, precluding valid analysis. Homogeneity: 0: good homogeneity, 1: almost homogeneous images, 2: minor heterogeneity close to skull base and in temporal lobes, 3: heterogeneous signal at multiple locations, 4: high number of heterogeneous areas, 5: highly heterogeneous images, no valid analysis possible.
In addition, all sequences were analysed with regard to contrast ratios (CR) of WM lesions, GM lesions and GM to WM. CR was defined as the absolute value of (SI1-SI2)/SI2 with SI1 and SI2 the signal intensity of the first and second mentioned term in the description (i.e. GM/WM, lesion/WM, and lesion/GM). On the 3D-T1 images regions of interest (ROIs) of at least 10 mm2 were drawn for GM, WM, and in MS patients for lesions. The outer rim of the cortex was not included to prevent overestimation of GM signal due to the hyperintense rim found on 3D-MP-FLAIR and 3D-MP-DIR which did not contribute to contrast between GM and WM. The ROI areas were copied to the other registered 3D sequences for CR calculation by use of MIPAV software (CIT, NIH, Bethesda, MD, USA). For 2D images this was done manually as automatic registration did not always succeed. As the appearance of normal anatomy and lesions was fundamentally different compared to conventional sequences at lower field, lesions were counted in consensus by two observers (4 years and 8 years of experience). The order of sequences and subjects was randomised. Lesions were counted and categorised according to their anatomical locations: periventricular (PV) lesions in contact with the ventricles or within a distance of 3 mm; deep white matter (DWM) lesions further than 3 mm from the ventricles and not in contact with the cortex; juxtacortical (JC) lesions in contact with the cortex but residing in WM; mixed GM/WM (Mixed) lesions residing in both WM and GM, and intracortical (IC) lesions inside the cortex, but not in the WM. GM lesions in general were defined as lesions that were of the mixed type or intracortical. Other lesion types like deep GM lesions and infratentorial lesions were infrequently found but were not used for this comparison, as the field of view did not extend to those areas for all subjects. On 3D-T1 images, all lesions were defined as hypo-intensities compared with surrounding WM and GM, whereas on the other sequences they appeared as hyper-intensities. Finally, the number of lesions detected by the 3D methods FLAIR, DIR and T1 were compared with 2D-PDT2 by use of a Friedman test and further analysed by a Wilcoxon signed rank test. When significant, results were shown as relative gain compared with 2D-PDT2.
None of the subjects experienced any additional inconvenience at 7 T compared with previous examinations they had at 1.5 T or 3 T.
Mean Contrast Ratios in healthy controls and MS Patients
Contrast Ratio [value (sd)]
Number of lesions observed at 7 T in 10 MS patients, per sequence and location
Image quality scores at 7 T in MS patients and healthy controls
Patients (n = 10)
Motion Artefacts [mean (sd)]
Homogeneity [mean (sd)]
Controls (n = 5)
Motion Artefacts [mean (sd)]
Homogeneity [mean (sd)]
Using magnetisation preparation, we were able to obtain whole-brain (near-) isotropic, high-resolution 3D-MP-FLAIR and 3D-MP-DIR images at 7 T, within clinically feasible MR data acquisition times. These optimised techniques were able to detect higher numbers of MS lesions in both WM and GM, compared with 2D-PDT2 and 3D-T1. Future studies should now compare these new sequences at 7 T with similar sequences at lower field strength within the same patient population. However, one should notice that pulse sequences that are optimal for lower field strength are not necessarily optimal at high field strength and vice versa. Whereas a 3 T vs. 7 T comparison of a 2D-PDT2 only requires slightly different TR and TEs, comparison of 3D-FLAIR and 3D-DIR requires larger modifications such as the prevention of increased, and unwanted, T1-weighting at 7 T by using magnetization preparation [19, 21].
The RF wavelength at 7 T is less than half of that at 3 T. Therefore, wavelength becomes a factor that compromises the application of a uniform B1-field, which leads to decreased signal intensity in the temporal lobes when compared with more central brain areas on images of most of the 7 T sequences. Also a lack of signal intensity can be observed around the fossa posterior. A possible solution for the problem of an heterogeneous B1-field would be the use of multi-transmit/receive coils . Nevertheless, when using the currently available hardware, MS lesions appear with high signal intensity compared with the surrounding tissue even in the temporal lobes. Although lesion detection in the temporal lobes may not be a problem (Fig. 4), the image quality will be enhanced if homogeneous B1 would be achieved in the brain by e.g. B1 shimming solutions with multiple RF channels. A higher B1 in the temporal lobes would lead to higher signal intensities, which would facilitate accurate localization of the lesions. Moreover, the T2 weighting of the long TSE train would be stronger (because the refocusing angles would be larger), which would likely further increase the lesion conspicuity. When looking at CR values of the FLAIR and DIR sequences, their high values for lesion contrast can be explained by both the inversion pulses and MP: The high CR values for DIR are caused by the suppression of WM. The MP used before the inversion pulses hardly affects CSF or tissue with a high T1 relaxation time. Therefore lesions (with high relaxation times) that are not suppressed by the inversion pulse will neither have lost signal intensity from the MP. Therefore, although the signal intensities of normal GM and WM are lower in the temporal lobes, lesions still have a high conspicuity. Future studies could address the impact of regional contrast effects on observer variability. Despite the use of adiabatic inversion pulses, B1 heterogeneity can cause incomplete spin inversion at certain locations about 80% inversion at one third of the nominal B1 , which may explain the heterogeneity in CSF suppression. As CSF has a very long T2 relaxation time, only a very small amount of remaining longitudinal magnetisation is needed to obtain a considerable but variable signal in the heavily T2-weighted FLAIR and DIR images. Moreover, the higher signal intensity of WM extending from the central sulcus to the thalamus compared with other WM is similar to that observed at lower field strength and is probably due to regional variation of both T1 and T2 relaxation times of WM [23, 24]. These effects were also shown for 7 T T2*-weighted images .
All sequences used showed many perivascular spaces at 7 T. For the 2D-PDT2 and 3D-T1 images this may be explained by the high spatial resolution, which reduces partial volume effects. For FLAIR and DIR it may also partly be explained by the fact that fluid in peri-vascular spaces has a different composition compared with CSF, which may shorten T1 relaxation times [26, 27]. Inversion recovery timing would then not meet the T1 relaxation. The risk of confusing peri-vascular spaces with lesions may occur and therefore it is necessary to follow them over several slices and/or observe them on orthogonal views (the near isotropic resolution of the 3D acquisition allows reformatting of the data in any desired orientation). Lesions and peri-vasuclar spaces can then be discriminated on their shape: most lesions have an oval, sphere-like appearance, while peri-vascular spaces have an elongated, linear shape and have small diameters over several slices. One of the readers is a board certified neuro-radiologist with over 10 years of experience in neuro-inflammatory diseases and had no problems discriminating lesions from peri-vascular spaces using the described method.
Visualisation of the cortex
To enable correct classification of lesions (especially juxta-cortical and mixed lesions), a good GM/WM contrast is necessary. The 3D-MP-FLAIR and 3D-MP-DIR sequences showed good contrast between GM and WM, except for some areas, e.g. pre- and post-central gyrus. The hyper-intense rim, which represents the outer layers of the cortex, may reduce the visibility of small intra-cortical lesions. Next to the hyper-intense rim also a banding pattern of the cortex is visible, mostly from the motor cortex. This effect has also been observed on DIR images at 3 T and is expected to visualise layer III and V (both pyramidal) of the motor cortex . The amount of myelin in these layers is higher than in the other layers which may cause the observed effects. The observed contrast within the cortex due to differences in cortical structure has also been shown in several T1 relaxation time and T2* weighted MRI studies e.g. [14, 23], although more subtly.
Detection of lesions
Many deep white matter and juxta-cortical lesions in particular were found in MS patients by both 3D-MP-FLAIR and 3D-MP-DIR. The number of pure intra-cortical lesions however, was relatively low. This might be caused by the afore-mentioned hyper-intense rim that reduced conspicuity of these lesions, but may also be due to the specific patient group which had a relatively low EDSS score (4.5). To our knowledge no clinical studies so far have been performed using 3D-FLAIR and 3D-DIR at 7 T, possibly because implementation of these sequences is not straightforward at high field strength and needs dedicated modifications of the sequences . A number of lesions were detected on the FLAIR and DIR images but not on the T1 and PD/T2. As for the 2D PD/T2 sequences, this may be caused by the larger slice thickness. For the 3D-T1, with a comparable spatial resolution as the FLAIR and DIR sequences, further study with a one-on-one lesion comparison between sequences may help to explain these differences. The studies published before have used 3D-T1w, 2D-PDT2 and 2D-T2*w sequences  or worked with ex vivo material . The T2*w sequence type did show high conspicuity of lesions, as well as the same WM to GM lesion ratio as observed with histopathology. As the contrast type in T2*-weighted images strongly differs from that known from more conventional sequences, it is unknown how the lesion conspicuity on T2* weighted images compares with that in FLAIR and DIR images. A practical complication of T2*-weighted images is the long acquisition time that is needed for whole brain coverage. The shorter imaging times in combination with the high contrast of the hyperintense lesions in the imaging types presented in this study may therefore be more feasible for clinical applications in MS.
Unfortunately, the single transmit coil that was used in this study is unable to induce a fully homogeneous B1 field at 7 T. Possibly the best way to optimise the homogeneity of the images, and improve the degree of CSF suppression, would be to use parallel excitation by use of multiple transmit. If not available, a normalisation filter may be used to obtain more homogeneous images at the cost of some signal intensity and contrast. In this explorative study normalisation was not used so as not to loose any signal intensity. Owing to the presence of the bright rim in the outer layer of the cortex on both 3D-MP-FLAIR and 3D-MP-DIR it cannot be excluded that subpial lesions are not being detected. The conspicuity of this cortical lesion type should be validated on post-mortem slices in combination with histopathology. Also a comparison of images of the same type acquired at lower field strength would be valuable to show the additional strength of 7 T MRI in MS.
To conclude, we have shown the feasibility of both 3D-MP-FLAIR and 3D-MP-DIR for (cortical) lesion detection at 7 T with high contrast ratios and high spatial resolution in MS patients. Future studies that also look at the relation with clinical impact may further reveal its value in imaging for diagnostics and research.
Mr. F. Visser is a part-time employee of Philips Healthcare, Best, the Netherlands.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.