detection of multiple sclerosis lesions
INTRODUCTION
Conventional wisdom dictates that, in order to detect the lesions of MS in the brain and spinal cord by MR^T^-weighted images are best. One problem with the SE (and FSB) pulse sequences usually employed is that the cerebrospinal fluid appears bright, which can result in lesions being obscured by partial volume effects or flow artefacts. As discussed in chapter 3, one proposed solution is the FLuid Attenuated Inversion Recovery (FLAIR) pulse sequence. This exploits differences in relaxation times between CSF and lesions, using an initial inversion pulse to null or reduce the signal from the former whilst maintaining (or even increasing) the T2 weighting of the latter (Hajnal et a l, 1992a;
Thomas et al, 1993). The disadvantages of FLAIR are its long acquisition times and low inherent signal-to-noise. Fast FLAIR (fFLAIR) has been developed by incorporating a similar inversion pulse into a FSE sequence (Hajnal et a l, 1992b; den Boer et a l, 1993; Rydberg et a l, 1994). However, a much simpler way of obtaining images with good lesion contrast and dark CSF is to use a conventional SE or FSE pulse sequence with a short TE for “proton density” (PD) weighting in conjunction with a moderate TR for mild weighting. In the previous chapter it was shown that more lesions can be seen on short TE than long TE images mainly due to difficulties in identifying periventricular and subcortical lesions in the face of high signal from CSF. In this chapter fFLAIR was first optimised for CSF suppression and lesion/background contrast and then compared both subjectively and objectively to a FSE sequence with short TE^j and mild weighting.
METHODS
A series of experiments were performed both in patients with clinically definite MS and in healthy controls. In all cases a 256x256 matrix and 6 mm slice thickness were used. Dual echo FSE images were obtained (FSE2ooo/i5,io5» echo train 8). For fFLAIR images, a TR of 6000 ms and a slice selective inversion pulse with a FSE readout of 16 echoes were used. In all cases a single signal average was obtained. Scans were reviewed by a neuroradiologist (Dr IF Moseley) and all lesions recorded. The two echoes of the dual echo FSE were reviewed together, the fFLAIR images separately. The FSE and fFLAIR images were then compared and lesions seen on one but missed on the other noted. Quantitative measures of lesion/white matter, lesion/grey matter and lesion/CSF contrast-to-noise ratio (CNR) were made using the image analysis software Dispimage (Plummer, 1992) on a Sun Workstation. Measurements were made from the fFLAIR as well as from the short TE FSE images, where CNR was defined (see chapter 2) as:
SIi^^ = Mean lesion signal intensity
SIf,g = Mean background (CSF, white or grey matter) signal intensity
5 4,y = Mean air signal intensity
Optimisation of T E and TI
Preliminary studies in which TI was varied whilst keeping TE^f constant showed that CSF signal was minimum at a TI of approximately 1600-1800 ms (figure 5.1). A series of
400 CSF White m atter 300
^
200
1002000
25 0 0 3 0 0 01000
1500 Tl/msFigure 5.1. Relative signal intensity of CSF and white matter with varying inversion time (TR 6000 ms, TE,f 120 ms).
experiments was then performed in two patients with MS, varying TE^f whilst keeping TI constant. CNR measurements were made from nine lesions. Independent of TI, lesion/white matter contrast varied little with TE^fS of between 100 ms and 240 ms and was maximal at around 120 ms (figure 5.2). This TE^f was therefore chosen for subsequent experiments.
fFLAIR vs FSE
As would be expected, CSF signal was minimal at the same TI independent of TE^f. However at the "optimal" TE^f of 120 ms, CSF did not become isointense with white matter until TI was approximately 2500 ms (figure 5.1). Four patients and four controls
were imaged at a TI of 1650 ms, to minimise CSF signal, and subsequently five patients and three controls at a TI of 2300 ms. At this longer TI, CSF was still hypointense relative to white matter and hence did not obscure lesions. All subjects also underwent dual echo FSE. FSB took approximately 4 minutes, fFLAIR nine minutes to generate 18 slices.
200
600 CSF White matter Lesion 500 - 150 Contrast100
Q.S)
200
50 100 50 100 150 2 0 0 2 5 0 3000
TE^f/ms
Co 01
I
5 A C i O % 5Figure 5.2. Lesion/white matter contrast as a function of TE^f.
RESULTS 77=7650 ms
On the fFLAIR images with the shorter TI, the signal from many lesions, especially around the ventricles was suppressed (figure 5.3). There was diffuse high signal around the ventricles (especially the occipital and frontal horns) and to a lesser degree in the centrum
semiovale in both patients and controls, which obscured lesions that were clearly visible on the short TE^f FSB images. Flow effects were prominent on the fFLAIR images especially around the basal cisterns, third ventricle and aqueduct. fFLAIR demonstrated only four cortical lesion that were poorly seen or invisible on FSF.
CNRs measured from 24 lesions showed that lesion/white matter contrast was significantly higher for FSF (8 .6 ± 2.5 for FSF, 4.9 ± 2.1 for fFLAIR, p<0.0001. Student's t test).
Figure 5.3. (a) FSF (TR=2000 ms, TF^plQ ms), (b) fFLAIR (TI=1650 ms, TFgj=120 ms). There is diffuse high signal from the white matter and low signal within lesions on fFLAIR.
TI=2300 ms
At the longer TI periventricular lesions no longer showed the loss of signal noted at 1650 ms. The anatomical definition of the fFLAIR images was inferior, the diffuse changes already noted were still present, if less marked and still obscured lesions clearly visible on FSF (figure 5.4).
Of 479 lesions identified in five patients on short TE^p FSE, 144 were not seen on fFLAIR; conversely there were 414 lesions identified on fFLAIR of which 140 were not seen on FSF. Most lesions not seen on one sequence were small. Lesions that were discrete on FSF often appeared more confluent on fFLAIR.
Figure 5.4. (a) FSF (TR=2000 ms, TF,f=19 ms), (b) fFLAIR (TI=2300 ms, TF,f=120 ms). Some small lesions on fFLAIR (arrowed) are difficult to see against diffuse background high signal.
CNRs measured from 119 lesions showed that lesion/white matter contrast was essentially identical for short TF^f FSF and fFLAIR (table 5.1) although lesion/grey matter and lesion/CSF contrast were superior on fFLAIR.
FSF fFLAIR
Lesion/white matter 9.4 ±3.3 9.3 ±3.4
Lesion/grey matter 3.9 ±2.7 7.2 ±2.8*
Lesion/CSF 9.2 ±2.8 11.6 ±2.7*
* fFLAIR > FSE, p<0.001, Student's t
DISCUSSION
Previous comparisons of FLAIR with SE have tended to concentrate on heavily 7^- weighted images. However, as shown in the previous chapter, it is short rather than long TE images that demonstrate most lesions, mainly due to the relatively low signal from CSF. By using a TR of only 2000 ms in the current study, CSF on the short TE^f FSE images remained appreciably darker than grey matter and so did not obscure periventricular or cortical lesions.
Early studies of FLAIR at 1 T (Hajnal et a l, 1992c) used very long echo times (up to 240 ms). For the detection of MS lesions, however, shorter echo times are better and in this study an optimum TE^f of 120 ms was found. This corresponds well with what would be expected from published values of Tj of lesions and white matter at 1.5 T (Larsson et al,
1988, see figure 5.5). In contrast to previous reports of FLAIR at 1 T, the null point of CSF was found to be at a TI of around 1700 ms, at which point there were marked detrimental effects on lesion contrast. At longer TIs there was no advantage of fFLAIR over FSE in terms of lesion detection and essentially identical lesion/white matter contrast.
The fFT^AIR sequence used in this study was not very efficient and hence time consuming. A much more efficient sequence, using an ingenious "sequential interleaving" technique, has been developed by other workers (Rydberg et a l, 1994) permitting the generation of 36 196x256 images in just over 5 minutes. To generate the same slices using the FSE sequence employed in this study would take over 6 minutes. Rydberg’s sequence employed a TR of 11000 ms, compared to 6000 ms in the current study. Theoretical calculations would suggest that the longer TR results in images with CNR of approximately 50% higher
than those in this work (Rydberg et al., 1995).
200
- 1 0 0 White matter Lesion Contrast - 8 0 - 6 0100
- 4 0- 2 0
100 150 2 0 0 2 5 0 3000
50 n o 3 A % TE/msFigure 5.5. Graph of predicted white matter/lesion signal intensity and contrast (arbitrary units) vs echo time using the values of average Tj of white matter and lesion (81 and 174 ms respectively) of Larsson e ta l, 1988. (The effects of proton density differences, which might skew the contrast curve slightly to the left, have been ignored).
The higher lesion/grey matter contrast of fFLAIR might be expected to improve the detection of cortical lesions but in practice little advantage was found. This may partly be explained by the small proportion (only 5%) of MS lesions that are purely cortical (Brownell and Hughes, 1962). Other studies comparing fFLAIR with FSE and/or SB have however tended to find fFLAIR superior in detecting cortical and subcortical lesions (Filippi et a l, 1996a; Bastianello et a l, 1997; Gawne-Cain et a l, 1997; Stevenson et a l.
1997). This may partly be explained by the longer TRs employed in these studies (9000- 11000 ms) and resultant improved CNR. On the other hand, the TRs of the FSE and SE sequences with which fFLAIR has been compared have generally been 2500-3000 ms which, for the reason outlined above, may be too long. On balance, the evidence seems to be mounting that, with the correct parameters, fFLAIR can have contrast superior to SE and FSE (Rydberg et a l, 1994, 1995; Bastianello et a l, 1997).
A potential pitfall of FLAIR and fFT,AIR is the diffuse high signal seen in both healthy subjects and patients, which obscures some lesions whilst causing others to fade imperceptibly into the surrounding white matter. This poses potential problems if these techniques are to be used for measurement of lesion volume, a technique gaining wider acceptance in the context of MS treatment trials (Paty et a l, 1993). Filippi et a l (1996a), comparing fFLAIR with SE and using a semiautomated thresholding technique to measure lesion volume, found that more manual editing of lesion contours was required for periventricular lesions on fFLAIR because of diffusely increased signal. Overall, however, it seems that lesion volume measurement from fFLAIR images is quicker. Reproducibility is comparable (Gawne-Caine et al, 1998) or superior (Filippi et a l, 1996a; Bastianello et a l, 1997). Nevertheless, it is lesions in the posterior fossa and spinal cord that account for much of the disability in MS and in these locations FLAIR sequences are inferior to FSE and SE (Filippi et a l, 1996a; Bastianello et al, 1997; Gawne-Cain et al, 1997; Stevenson
et al, 1997). Correlation of MRI lesion load with disability might be expected to be worse with fFLAIR, although this has not be found in a cross-sectional study (Gawne-Caine et a l, 1998). Further, especially longitudinal, studies are needed.