Multiple sclerosis lesion detection in the brain: A comparison of fast fluid-attenuated inversion recovery and conventional T2-weighted dual spin echo

The extent of disease in multiple sclerosis (MS) can be monitored by recording the number and volume of lesions detected in the brain by MRI.^1-3 Many groups have used the proton density (PD) or mildly T2-weighted (long TR, short TE) conventional spin echo(CSE) sequence for this purpose; Paty et al.^4,5 used CSE in the recent study of interferon beta-1b in relapsing-remitting MS. This type of sequence has two disadvantages-it is slow to acquire (8 to 15 minutes on some scanners), and lesion conspicuity at the gray/white matter boundaries is relatively poor. The latter is difficult to improve because although lesion/brain contrast can be raised with increased T2 weighting(increasing TE), it is at the expense of increasingly bright CSF, making periventricular and subcortical lesions more difficult to identify.

The fast fluid-attenuated inversion recovery (fFLAIR) sequence has the potential to overcome these disadvantages because (1) the signal from CSF is suppressed by an initial inversion (180°) pulse before the main sequence allowing longer TEs to improve lesion/brain contrast, and (2) it achieves a relatively short acquisition time (6 minutes) by using the rapid acquisition with refocused echoes (RARE) technique.⁶

This study aimed to compare the sensitivities of CSE and fFLAIR in detection of MS lesions in the brains of a large cohort of patients with known MS, representative of the broad spectrum of clinical courses seen in practice.

Methods. Acquisition. Thirty-two patients with clinically definite MS⁷ were studied. There were eight patients in each of four clinical subgroups: relapsing-remitting (RR), secondary progressive (SP), primary progressive (PP), and benign (B) disease⁸ (for definitions seetable 1). Details of patient age, disease duration, and disability are shown in table 2. The study was approved by the local Medical Ethics Committee, and all subjects gave informed consent.

Scans were performed using a 1.5-T Signa (General Electric, Milwaukee, WI). The sequences used were dual echo CSE (TR = 2,000, TE = 34,90; 28 contiguous 5-mm slices acquired in 8 minutes) and an fFLAIR sequence derived from that described by Rydberg et al.⁹ (TI = 2,600, TR = 11,002, effective TE = 164, ETL = 8, 42 contiguous 5-mm slices acquired in 6 minutes). The TR/TI ratio was chosen to be optimal for MS lesion/white matter contrast.¹⁰ In each case, FOV was 25 cm, matrix 256², and NEX 1. The same axial slice positions were used for each sequence, and the patients were not moved between the two.

Analysis of hard copies. Each set of fFLAIR and CSE PD (SE 2,000/34)-weighted images was separately reviewed by two observers (M.G.-C., J.OR.) working in consensus and the lesions marked. Lesion identification on the fFLAIR images was guided by a previous study of the appearances of the fFLAIR in a series of 40 normal volunteers.¹¹ The fFLAIR and CSE images were then compared, and each lesion marked was recorded in one of the following groups: A, seen initially on both fFLAIR and CSE; B, seen initially on fFLAIR only; C, seen initially on CSE only. Two additional sub-groups were recorded: D, lesions initially scored on fFLAIR only but identified retrospectively on CSE (this is a subgroup of B); E, lesions initially scored on CSE only but identified retrospectively on fFLAIR (subgroup of C). Groups were subdivided according to lesion site and estimated size.

Site. Four sites were defined: (1) posterior fossa; (2) discrete-cerebral hemisphere white matter or basal ganglia; (3) subcortical/cortical-including lesions in the cerebral cortex and those touching it; (4) periventricular-adjacent to the lateral ventricles.

Size. Approximate lesion size was recorded as small (less than 5 mm), medium (5 to 10 mm), or large (greater than 10 mm). Measurements were of the long axis of lesions or perpendicular to the ventricular margin in periventricular lesions.

T2 calculations. The T2 values of 16 pairs of lesions from the posterior fossa and cerebral white matter were determined in nine patients(five SP, four RR) as follows: Calculated T2 images were prepared from the dual echo data using the formula T2 = (TE2-TE1)/[ln(S1/S2)] where S1 and S2 are the signal at echo times TE1 and TE2, respectively. Regions of interest(ROIs) over lesions were outlined using the image display program DispImage¹² and the mean T2 within each ROI recorded. The pairs of lesions were matched for size.

Statistical analysis. Wilcoxon's rank sum test was used to compare the number of lesions counted by CSE and fFlair (the numbers for each patient, anatomic site, and lesion size were considered separately). T2 values were compared using the t test for paired samples. All statistical calculations were performed using SPSS for Windows 6.1.

Results. Lesion identification (tables 3 and 4). In 32 patients 3,668 lesions were identified on the initial review (1,580 small, 1,165 medium, and 923 large)-2,892 by CSE and 2,943 by fFLAIR. Following comparison of the images, an additional 217 lesions could be identified on fFLAIR and 229 on CSE. The breakdown of lesions detected according to site and the number of lesions not demonstrated by each sequence, after review of the other images, are shown intable 3.

When films from the two sequences were initially viewed separately, fFLAIR detected more subcortical lesions (p < 0.0001), whereas CSE detected, overall, more posterior fossa (p = 0.001) and discrete cerebral lesions (p = 0.04). Regardless of lesion size, fFLAIR was superior in showing subcortical lesions and inferior in showing lesions in the posterior fossa (figures 1 and 2). However, in the case of discrete cerebral white matter lesions, fFLAIR showed fewer small lesions, a similar number of medium-sized lesions, but significantly more large ones. Sensitivity was similar in the periventricular region.

Table 4 shows the lesions counted when disease subtypes were considered separately. In all subtypes, fFLAIR detected more subcortical lesions and CSE more posterior fossa lesions, but the posterior fossa findings were significantly different only for the PP group. Overall, the PP and SP subgroups had a higher proportion of posterior fossa and subcortical lesions although this trend was not statistically significant.

In some cases, two to six lesions seen using one sequence appeared as one confluent lesion using the other. In such a situation, the lesion was recorded as a single large lesion identified on both scans. There was no obvious pattern of one sequence being more likely to show discrete or confluent lesions, although the pattern tended to be consistent with one patient.

T2 values. The mean T2 of cerebral white matter lesions (102.7, SD = 14.8) was greater than for posterior fossa lesions (80.2, SD = 6.4). The mean difference between pairs of lesions was 22.2 (SD = 16.2), and this was highly significant (t value = 5.52, two-tailed significance<0.001).

Discussion. The value of an MRI sequence in monitoring the progress of intracranial MS is proportional to its sensitivity of lesion detection. In this study of 32 patients with MS, fFLAIR performed better than CSE in the cortical/subcortical regions, worse in the posterior fossa, and comparably for periventricular lesions. For discrete cerebral lesions, the result was more complex-CSE showed more small lesions and fFLAIR more large ones. The results were similar when disease subtypes were considered separately, but not statistically significant, probably because of smaller numbers. The apparent trend for a higher proportion of posterior fossa and subcortical lesions in progressive disease types is of interest, but we did not examine this in detail in this study as we believe that the distribution of lesion volumes rather than numbers is likely to be more informative. This will be the subject of a separate report.

The method of scoring the films has implications for interpretation of the results. After first examining and marking the sequences separately, we counted the lesions looking at both sets of images together. If a lesion appeared small on one sequence but medium on the other, we recorded it as medium seen on both, the emphasis of the study being more on the ability of the sequences to detect lesions than to measure their size (lesion volume measurements are the subject of a separate study). Thus, the excess of small discrete lesions on CSE and larger discrete lesions on fFLAIR was not the result of the same lesions being differently attributed on the two sequences. In some cases, a single lesion on one sequence appeared as two or more lesions on the other, and this we recorded as a larger lesion detected on both. Neither sequence was more likely than the other to show confluent or separate lesions. Direct comparison of the two sequences also allowed us to assess whether the patient had moved between sequences and to match lesions appearing on different slices on the two sequences. In fact, detectable patient movement between sequences occurred in only four cases (three PP and one RR). Repositioning errors are discussed in more detail below.

We expected fFLAIR to perform well. FLAIR sequences display long T2 lesions conspicuously against the normal brain, are extremely sensitive to intracranial pathology,^13-15 and, according to some workers¹⁴ but not others,¹⁶ detect more MS lesions than CSE or fast spin echo (FSE). What factors might account for the differences in fFLAIR detection of small and large discrete lesions and its relatively poor performance in the posterior fossa?

1. FLAIR images of the brain show areas of increased signal (AIS) in normal white matter.¹⁷ Possible strategies for dealing with such AIS are to count all bright regions as lesion, resulting in inevitable false positives, or not to count AIS in regions where they are recognized to be a normal feature, which risks an unknown number of false negatives. AIS are less of a problem with the sequence used in this study than with some FLAIR sequences because at 164 msec, the effective TE is only moderately long. Nevertheless, the corticospinal tracts and periventricular regions give high signal.¹¹ We attempted to minimize the risk of false positives by not counting symmetric AIS in the regions of the corticospinal tracts, but although AIS in the posterior internal capsule are consistently seen in all normal individuals, the extent of high signal in the rest of the deep cerebral white matter is variable and often asymmetric.¹¹ The excess of large discrete lesions seen on fFLAIR compared with CSE may partly be due to inclusion of such AIS, that is, they may be false positives. No excess of fFLAIR lesions was seen in the periventricular regions, and this may be because PD-weighted CSE images show"normal" AIS in this region, often referred to as periventricular caps; if periventricular false positives occurred, they occurred consistently on both CSE and fFLAIR.

2. The speed of fast or turbo FLAIR techniques derives from use of the RARE or FSE technique.^9,13,14 The acquisition of multiple echoes within one repetition time (TR) results in variable echo times for the different phase encoding steps, and the image produced has an effective rather than a true TE. This can result in blurring of edges in the phase encoded direction with decrease of contrast.¹⁸ Blurring applied to both sides of a small lesion or point may result in its complete loss (point spread function)¹⁹; this might explain the reduced number of discrete lesions less than 5-mm shown by fFLAIR in the current study. If this is correct, we would expect fFLAIR to perform as well as or better than FSE, and indeed, in a smaller study,²⁰ we found fFLAIR detected more discrete cerebral lesions than FSE. Although the effects of blurring and decreased contrast might significantly reduce detection of small(discrete) white matter lesions, they are presumably outweighed in subcortical regions by the increased inherent lesion/gray matter contrast and the greater suppression of CSF of fFLAIR.^9,16

3. A very small lesion might be visible with one scan orientation and not another, and we may have missed small lesions on one sequence solely because of repositioning error. Because we were able to follow larger lesions between slices, repositioning errors cannot account for the significant differences in their detection, but fFLAIR's detection of significantly fewer small discrete lesions might possibly be due to repositioning. If the rate of detection of small lesions is purely a result of lesion position relative to slice orientation, and the distribution of small lesions in white matter is random, then we might expect the number of lesions missed on a repositioned scan to be balanced by a similar number of new lesions detected. If bad positioning results in one sequence "losing" more lesions than the other, this is a genuine difference between the sequences, possibly due to low lesion/white matter contrast.

4. The fFLAIR sequence suffers from vascular and CSF flow artifact, especially in the posterior fossa,⁹ and this might account for the decreased detection of lesions in the region. CSF inflow artifact is particularly a problem of FLAIR sequences, but pulsation artifacts are often attributed to the use of RARE. In the previously referred to study,²⁰ fFLAIR was compared directly with FSE and was found to detect significantly fewer posterior fossa lesions.

5. Lesions in the posterior fossa could have different MRI characteristics from those in the cerebral hemispheres. We found that although MS lesions look brighter on fFLAIR slices high in the brain they are less conspicuous in the posterior fossa, and we found the T₂ values of lesions in the posterior fossa to be significantly lower than those in the cerebral white matter. We did not design the study to measure T₂ values of lesions; the intention was to compare lesions from different regions rather than to determine absolute T₂ values. Because many of the lesions were small, it was rarely possible to place an ROI completely within a lesion, and unavoidable partial voluming of white matter would account for the relatively low absolute T₂ values we found. Although not directly comparable with T₂ values from other studies, our results are still valid for comparison within the current study because we matched posterior fossa and cerebral lesions for size, and the partial voluming effect is likely to have been consistent. The fFLAIR sequence used TR/TI ratios optimized for detection of MS lesions whose T₂ values (110 to 150 msec)¹⁰ were derived from the results of studies of large, predominantly cerebral lesions.^21,22 If T₂ values of lesions in the posterior fossa are genuinely lower than those in the cerebral hemispheres, they might not be optimally demonstrated by Rydberg's¹⁰ "optimized" sequence parameters: even longer TR/TI values might improve contrast, albeit only slightly. The difference in lesion T₂ values in the posterior fossa might reflect the influence of the local tissue environment (e.g., closely packed fibers in the brainstem) on the pathologic characteristics of lesions in different sites.

Although the total numbers of lesions detected by fFLAIR and CSE are similar, they are made up of more subcortical and large discrete cerebral lesions on fFLAIR and a higher proportion of posterior fossa and small discrete lesions on CSE. It is difficult to know which are the more important lesions to detect, but it might be helpful to review why the MRI studies are being performed. In the context of MS, MRI is used in two distinct ways: (1) to confirm the diagnosis and (2) to monitor progress of the disease, particularly in the context of therapeutic trials.

For the initial diagnosis of MS, the total number of lesions seen is less important than the demonstration of definite abnormality. If fFLAIR can show abnormality where CSE does not, as has been reported,¹⁵ it may be more sensitive for diagnosis. On the other hand, the reduced detection of infratentorial lesions, which are a common and characteristic finding in MS,²³ is disadvantageous. We might speculate that because fFLAIR is more able to show lesions in sites not usually involved in normal aging or ischemia, such as cortical gray/white matter boundaries,²⁴ it could improve specificity, but this assumes that fFLAIR does not give a similarly increased yield of subcortical lesions in other conditions. A systematic review of the specificity and sensitivity of the sequence in MS and other white matter disorders is needed.

For monitoring the progress of MS in clinical trials, we are interested in the accumulation of lesions over time, and thus the emphasis is on number and size of lesions detected; the MRI change should also correlate with the clinical outcome if MRI is to be a useful surrogate marker. CSE, having been used for almost a decade, and in some large cohorts of patients,^1,4,5,25-27 is a well-tried sequence, and the volume of disease it detects is known to correlate with motor disability, albeit modestly.⁵ The importance of the different range of lesions detected by fFLAIR needs to be examined in more detail; lesions in various anatomic sites might be expected to contribute differently to clinical scales of motor or neuropsychological impairment.^28-30 The apparent insensitivity of fFLAIR to posterior fossa lesions is of concern as this is a site of many disabling lesions in MS. Of equal concern is the much lower sensitivity of fFLAIR in detecting cord lesions, which we noted in a recent study.²⁰

Conclusions and future work. Although more sensitive than CSE for subcortical lesions, fFLAIR detects fewer posterior fossa lesions, and we should be cautious in our use of fFLAIR for the monitoring of MS. A future paper will explore, for each sequence, the relationships between measured lesion volume and site versus disability and other clinical features. The apparently higher proportion of posterior fossa and subcortical lesions seen on CSE in the PP and SP groups also warrants further investigation.

Acknowledgments

The authors acknowledge Drs. S.J. Riederer and J.N. Rydberg of the Mayo Clinic for providing the fFLAIR pulse sequence.

We thank D.G. MacManus, S.R. Webb, and H. Gallagher for performing the scans.