echo in the detection of multiple sclerosis lesions
INTRODUCTION
This study set out to establish whether fast spin echo (FSE) could replace spin echo (SB) as the sequence of choice for routine imaging of the MS brain. For the purposes of routine investigation of MS the main advantage of replacing SE with FSE would be saving of time. For research purposes, where optimal detection of the maximum number of lesions might be important, the time saved could be used to perform higher resolution, or higher SNR imaging. Although theoretical considerations suggest that the contrast obtained with FSE and SE should be extremely similar (Melki et al, 1991,1992; Constable et a l, 1992a), the theory has not been tested in the detection of MS plaques. Furthermore, it is often stated that heavily 7^-weighted SE images are the most sensitive MR technique for imaging the brain (for instance. De Coene et a l, 1992), without experimental backing. Given the time penalty associated with acquiring two echoes using FSE (see chapter 3, page 49), it is important to compare the contribution of long TR, long TE (T2-weighted) and long TR, short TE ("proton density"-weighted) images to the detection of plaques. The study therefore set out to answer two questions: is FSE as sensitive as SE and if so, are two echoes necessary or will one suffice?
METHODS
All scans (in this study and the others described in subsequent chapters) were carried out on a 1.5 T General Electric Signa system. Six patients with clinically definite MS underwent axial imaging of the brain first with dual echo SE (SE2ooo/32,8o taking 6 minutes
48 seconds in two patients, SE27oo/32,8o taking 9 minutes 11 seconds in four) followed immediately by dual echo FSE (FSE^gQQ/iggg, echo train of 8, taking 2 minutes 55 seconds). A matrix of 192x256, field of view 24 cm, 5 mm slices with 2.5 mm interslice gap and 1 excitation were used in all cases. Exactly the same slice position and orientation were used for the SE and FSE sequences.
The images were reviewed in three stages.
Stage 1
Initially each of the four sets of images from each patient was reviewed individually by two observers, a neuroradiologist (Dr S Halpin) and myself, working in concert. Areas of high signal considered to represent lesions were marked and numbered. Each lesion was scored according to size (diameter <5 mm, 5-10 mm or >10 mm) and the site recorded: those within the brainstem or cerebellum were referred to as posterior fossa (PE), those abutting the lateral ventricles as periventricular (PV), those in or immediately adjacent to cerebral cortex as cortical (C) and those supratentorial lesions away from the ventricles or cortex as discrete (D). In addition a certainty value was ascribed to each lesion: definite, probable or possible.
Stage 2
Next, the two echoes of the SE and of the FSE were reviewed together and each lesion was reassessed in terms of certainty and size. An overall total number of lesions seen with each pulse sequence was derived.
Stage 3
Finally the SE and FSE images were compared and each lesion once again reassessed in terms of certainty and size.
Those lesions identified at one stage which were felt on subsequent review not to be genuine were designated false positives.
RESULTS
Spin echo
Initially (stage 1) 459 lesions were recorded on either the short or the long echo or both. However, when the two echoes were then reviewed together (stage 2), 55 of these were felt not to be lesions (false positives) leaving a total of 404 lesions (321 definite, 42 probable and 41 possible): 60 PF, 139 D, 147 PV and 58 C (table 4.1).
Site SE 32 ms only SE 32 ms total SE 80 ms only SE 80 ms total SE Total PF 11 48 12 49 60 D 12 126 13 127 139 PV 32 138 9 115 147 C 15 49 9 43 58 Total 70 361 43 334 404
Comparison of short and long TE images i. TE 32 ms
There were 70 lesions identified on the short TE images which were initially (stage 1) not seen on the long TE (table 4.1). Twelve of these had been graded as possible, 15 probable and 43 definite. Fifty-five were small, six medium and nine large. Forty-seven could still not be seen on the long TE images even with further review (stage 2). Of these 27 were adjacent to the ventricles, three were of grey matter intensity and seven were considered to be indistinguishable from cerebrospinal fluid in a sulcus; the remaining ten were indistinguishable from surrounding white matter. On review of the 23 other lesions initially not detected on the long TE images, 18 were considered to be present but less visible than on the short echo, while five were seen equally well and had been missed at the initial assessment.
When the long TE images were also considered, a more certain judgement was possible in 54 lesions, less certain in seven. It was considered that there had been 19 false positives on the original analysis of the short TE images alone, of which 13 had been graded as possible and six probable.
ii. TE 80 ms
There were 43 lesions seen on the long TE which had initially (stage 1) not been noticed on the short TE images (table 4.1). Sixteen had been graded as possible, 16 probable and 11 definite. Thirty-three were small, four medium and six large. On review of both echoes together (stage 2), 12 were still not visible on the short TE images while 29 were less visible and two were as visible as on the long TE.
With the addition of the short TE images a more certain judgement was possible in 58 lesions and a less certain in 15. It was considered that there had been 36 false positives on the original analysis of the long TE images alone, of which 17 had been graded possible, seven probable and 12 definite.
Significantly more PV lesions were detected on the short TE images than the long TE = 15.00, df=l, p<0.001) whereas the differences in the other sites were not statistically significant.
Fast spin echo
Initially (stage 1) 468 lesions were recorded on either the short or the long echo or both. However, when the two echoes were then reviewed together (stage 2), 70 of these were felt to be false positives leaving a total of 398 lesions (313 definite, 40 probable and 45 possible): 80 PF, 132 D, 133 PV and 53 C (table 4.2).
Site FSE 19 ms only FSE 19 ms total FSE 95 ms only FSE 95 ms total FSE Total PF 22 73 7 58 80 D 23 123 9 109 132 PV 31 122 11 102 133 C 22 49 4 31 53 Total 98 367 31 300 398
Comparison of short and long TE^f images
i. TEgf 19 ms
There were 98 lesions identified on the short TE^f images which were not seen on the long TEgf during stage 1 of the analysis (table 4.2). Twenty-two of these had been graded as possible, 19 probable and 57 definite. Seventy-eight were small, 12 medium and eight large. After review (stage 2), 21 were still not visible on the long TE^f images: seven were adjacent to the ventricles, seven were considered to be indistinguishable from CSF in a sulcus and seven were indistinguishable from surrounding white matter. The remaining 77 lesions were now visible on the long TE^f images. Of these, 51 were felt to be less visible than on the short echo while 26 were seen as well.
When the long TE^f images were also considered, a more certain judgement was possible in 66 lesions, less certain in 14. It was considered that there had been 47 false positives on the original analysis of the short TE^f images alone, of which 26 had been graded as possible, 15 probable and six definite.
ii. TEgf 95 ms
There were 31 lesions seen on the long TE^f which had initially (stage 1) not been noticed on the short TE^f images (table 4.2). Twelve had been graded as possible, five probable and 14 definite. Twenty-six were small, four medium and one large. On review of both echoes together (stage 2), seven were still not visible on the short TE^f images while 18 were less visible and six were as visible as on the long TE^f.
lesions and a less certain in eight. It was considered that there had been 23 false positives on the original analysis of the long TE^f images alone, of which nine had been graded possible, 11 probable and three definite.
Significantly more lesions were detected on the short TE^f images than the long TE^f in all four sites (PF, = 9.48, p<0.005; D, ^ = 6.97, p<0.01; PV, = 11.31, p<0.001; C, ^ =
16.51, p<0.001).
Spin echo vs Fast spin echo
At the end of stage 2 of the analysis, 509 lesions were seen on FSE or SE or both. However, when the two pulse sequences were then reviewed together (stage 3), 20 of these were felt to be false positives (9 SE and 11 FSE) leaving a total of 489 lesions of which 395 were seen on SE and 387 on FSE. The anatomical distribution is shown in table 4.3.
Site SE only SE total FSE only FSE total Total
PF 13(4) 57 (37) 33 (10) 77 (40) 90 (53)
D 29 (13) 137(114) 23 (19) 131 (119) 160(137)
PV 37(32) 145 (135) 19(16) 127(119) 164 (153)
C 23 (12) 56 (34) 19(11) 52 (30) 75 (50)
Total 102 (61) 395 (320) 94 (56) 387 (308) 489(393)
Table 4.3. Lesions detected by SE vs FSE. Numbers in brackets are definite lesions.
Of the 395 lesions seen on SE, 102 had not been seen on FSE during stage 2 analysis, of which 61 were felt to be definite, 25 probable and 16 possible: 13 were PF, 29 D, 37 PV
and 23 C. After review (stage 3), 53 lesions were still not visible on FSE (five cortical lesions had grey matter intensity on both echoes; 15 PV lesions could not be distinguished from adjacent CSF; an additional 33 lesions - 3 PF, 10 D, 11 PV and 9 C - were simply invisible). Forty lesions were visible but much less clearly seen on FSE; nine were equally visible but had been missed.
Of the 387 lesions seen on FSE, 94 had not been seen on SE during stage 2 analysis, of which 56 were felt to be definite, 14 probable and 24 possible: 33 were PF, 23 D, 19 PV and 19 C. After review (stage 3), 49(18 PF, 16 D, 4 PV and 11 C) were still not visible on SE of which five (all cortical) had grey matter intensity on both echoes. Thirty-seven lesions were now seen but were less visible; eight were felt to be equally visible on SE but had been missed.
Significantly more PF lesions were detected by FSE than by SE (%^ = 11.68, p<0.001) whereas more PV lesions were detected by SE (%^ = 6.98, p<0.01). The differences in the other areas were not significant. Confining the analysis to definite lesions there was no difference in the number of PF lesions detected, but the difference between the PV lesions remained significant (%^ = 5.93, p<0.025).
Most of the lesions which were only seen on one or other sequence were small (86/102 of those seen on SE and 73/94 on FSE). However 15 large lesions were missed on FSE: 13 of these were PV. Three large lesions were missed on SE (2 PV and 1 D).
DISCUSSION
The . advantage of replacing SE with FSB as the standard long TR sequence for brain imaging is the significant saving of time, hence improving patient compliance and reducing costs. This study was the first to carry out a systematic comparison of the two sequences
in MS.
There are theoretical reasons why FSE might prove less sensitive in detecting MS lesions. First, many MS plaques are located around the lateral ventricles; edge effects with FSE cause signal enhancement at interfaces (in this case between CSF and white matter) which
might obscure periventricular lesions. Secondly, with conventional SF, periventricular lesions are often best seen on shorter TF images as the CSF appears relatively dark. This is because the TR is sufficiently short to cause a mild degree of weighting to the CSF. On equivalent FSE images the CSF can appear quite bright, mainly because a longer TR is often required to permit sufficient slices to be collected to cover the whole brain. The CSF can be made darker by using a shorter TR, or a shorter echo train (eg 4) but the former has the disadvantage of limiting the number of slices that can be acquired in one acquisition and the latter that of increasing the scan time. Thirdly, theoretical calculations show that the "blurring" effect of FSE can cause small objects to become invisible (Constable and Gore, 1992). Finally, the SF images were acquired using "flow compensation"; this reduces artefacts in areas adjacent to flowing blood or CSF; this was not available on FSE, which might make lesion detection problematic in areas, such as the posterior fossa, where there is prominent CSF pulsation.
Despite these theoretical concerns regarding FSE, the study showed that a remarkably similar total number of lesions was identified on SE and FSE and that about three quarters of the lesions reported on one sequence were also identified on the other. This suggests that FSE can replace SE as the standard long TR sequence in the routine diagnostic work up of suspected MS.
Nevertheless, it is notable that about a quarter of lesions identified on one sequence were not seen on the other. Part of this may relate to patient movement between scans causing small lesions to be lost (there was a 2.5 mm interslice gap in all cases). A few periventricular lesions were missed by FSE, as might be expected due to the edge enhancement effect and the brightness of the CSF on the short TE^f images. Conversely, despite the lack of flow compensation FSE was as good as, or slightly superior to, SE in detecting posterior fossa lesions. Other authors have commented on intermittent high signal artefact crossing the brainstem on FSE, presumed to be due to CSF pulsation (Tien et a i,
1992). In our experience this did not cause problems with interpretation, but it is possible that some of the "lesions" detected in the brainstem on FSE were artefactual. However, it did seem that the contrast obtained with FSE was often slightly different to that from SE such that some lesions stood out more clearly, others less so (see figure 4.1). There are several factors potentially influencing contrast in FSE images, including point spread function effects (Constable and Gore, 1992), magnetisation transfer, the decoupling of J- modulation effects, diffusion, the production of stimulated echoes and direct saturation effects (Constable et al, 1992). Of these, point spread function and magnetisation transfer (MT) effects seem to be the most important. The former, which describes the blurring which can occur in the phase encoding direction resulting from the collection of different
i
y
Figure 4.1. Axial SE2700/32 (a) and (b), and FSE3500/19 (c) and (d), of patient with clinically definite MS. For most lesions, the contrast is very similar but a few lesions are only seen on SE (eg horizontal arrow) or on FSE (eg vertical arrow). CSF within the lateral ventricles appears brighter on the FSE image, making detection of periventricular lesions more difficult.
lines of k-space at different echo times (see chapter 3), could reduce lesion contrast. By the same token it is theoretically possible (but less likely) that point spread function effects could increase lesion contrast, by causing edge enhancement. MT effects would be more likely to increase lesion contrast. Multi-slice FSE is more sensitive than SB to signal loss due to MT (Tien et al, 1992). MS lesions tend to have lower MT ratios than white matter
(Dousset et a l, 1992) and would therefore lose less signal through MT effects than the tissue around them. Hence contrast would be increased. Thus the relative conspicuity of lesions on SB and FSB probably depends on a variety of different effects some operating to increase contrast on FSB, others to reduce it. The precise size, shape, location, 7^ relaxation time and macromolecular structure of each lesion will determine which sequence demonstrates it most clearly.
The short echo images demonstrate more lesions than the long echo images with both SB and FSB. There were only 11 definite lesions seen on the long echo SB images which had not been seen on the short echo images, whereas there were 43 seen on the latter which had not been seen on the former: the equivalent numbers for the FSB images were 14 and 57. Nevertheless, the additional information provided by the long echo images is useful in confirming the presence of lesions seen on the short echo and, in many instances, analysis of both echoes together improved the certainty with which a lesion was classified. Thus there would seem to be a good case for acquiring both echoes in a diagnostic work up for suspected MS.
Use of FSE to monitor treatment
sequences monthly in a group of patients with relapsing-remitting or secondary progressive MS. They found that approximately 55% of active (new, enlarging or enhancing) lesions were seen with Gd-DTPA enhancement alone, 20% only on the long TR sequence and 25% on both sequences. This suggests that both long TR and Gd-DTPA-enhanced short TR sequences are of value in monitoring disease activity in these clinical subgroups. Maximal enhancement of MS lesions occurs at least 5 minutes after Gd-DTPA injection (Kermode
et a l, 1990b), during which time a long TR FSE sequence can be obtained without any loss of lesion visibility: in fact the uptake of Gd slightly increases lesion visibility on long TR images (Barkhof et a l, 1992b). The complete examination, injection of Gd-DTPA followed immediately by long TR FSB and then short TR SB, takes only 10-15 minutes. Such a protocol is highly efficient and has since been adopted in trials of monoclonal anti- CD4 (Van Oosten et al, 1997) and anti-CDw52 (Moreau et a l, 1994). To date, however, lesion volume measurements, increasingly a part of treatment trials (Paty et a l, 1993), have proved less reproducible on FSB images than on SB images (Rovaris et a l, 1997). This is clearly an area in need of further study.