| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1
Division of Neuroradiology, The Johns Hopkins Medical Institutions, 600 N.
Wolfe St., Baltimore, MD 21287-7619.
2
Department of Biostatistics, The Johns Hopkins School of Public Health,
Baltimore, MD 21287.
3
The MR Perception Laboratory of The Kennedy Krieger Institute, Baltimore, MD
21287.
4
Department of Radiology, The Johns Hopkins Hospital, 600 N. Wolfe St.,
Baltimore, MD 21287-2182.
Received August 29, 2000;
accepted after revision October 23, 2000.
Supported in part by the Human Brain Project, National Institutes of Health
grant R01 AG13743, which is funded by the National Institute of Aging, the
National Institute of Mental Health, the National Aeronautics and Space
Administration, and the National Cancer Institute.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. We generated FLAIR, proton density-weighted, and T2-weighted brain images with 3-mm lesions using published parameters for acute multiple sclerosis plaques. Each image contained from zero to five lesions that were distributed among cortical-subcortical, periventricular, and deep white matter regions; on either side; and anterior or posterior in position. We presented images of 540 lesions, distributed among 2592 image regions, to six neuroradiologists. We constructed a contingency table for image regions with lesions and another for image regions without lesions (normal). Each table included the following: the reviewer's number (1-6); the MR sequence; the side, position, and region of the lesion; and the reviewer's response (lesion present or absent [normal]). We performed chisquare and log-linear analyses.
RESULTS. The FLAIR sequence yielded the highest true-positive rates (p < 0.001) and the highest true-negative rates (p < 0.001). Regions also differed in reviewers' true-positive rates (p < 0.001) and true-negative rates (p = 0.002). The true-positive rate model generated by log-linear analysis contained an additional sequence-location interaction. The true-negative rate model generated by log-linear analysis confirmed these associations, but no higher order interactions were added.
CONCLUSION. We developed software with which we can generate brain images of a wide range of pulse sequences and that allows us to specify the location, size, shape, and intrinsic characteristics of simulated lesions. We found that the use of FLAIR sequences increases detection accuracy for cortical-subcortical and periventricular lesions over that associated with proton density- and T2-weighted sequences.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
However, studies comparing FLAIR and proton density- and T2-weighted sequences have shown conflicting results regarding the visibility of brain parenchymal lesions, particularly of lesions with long T1 and T2 relaxation times [4,5,6,7,8,9,10,11,12,13,14,15]. These inconsistencies have been attributed to variation across MR scanners; implemented pulse sequence parameters; and size, location, and intrinsic characteristics of the lesions evaluated in the different studies.
Given the difficulty of controlling for these variables in the clinical setting, we sought to use computer-generated brain images for testing the effects of pulse sequence (FLAIR, proton density-weighted, T2-weighted) and lesion location on reviewers' abilities to detect simulated lesions. We formulated two null hypotheses regarding the effects of sequence and lesion location on reviewers' abilities to detect simulated lesions. First, the sequence (FLAIR, proton density-weighted, and T2-weighted) used to obtain images of a lesion does not affect reviewers' sensitivity and specificity for lesion detection. Second, the location of a lesion does not affect reviewers' sensitivity and specificity for lesion detection.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
A mixed multiecho spin-echo and inversion-recovery MR sequence was used to obtain images of the brain at the level of the lateral ventricles in a 40-year-old male volunteer. The mixed sequence simultaneously provided two image data sets: eight spin-echo images (TR = 1500 msec, 1 signal acquired) with different TEs (20, 40, 60, 80, 100, 120, 140, 160 msec) and eight inversion-recovery images (TR = 2000 msec; inversion time [TI] = 400 msec; 1 signal acquired) with different TEs (20, 40, 60, 80, 100, 120, 140, 160 msec). Image slice thickness was 5 mm, in-plane resolution was 0.80 x 0.86 mm (rectangular field of view, 165 x 220 mm; scan matrix, 205 x 256), and acquisition time was 9 min 30 sec.
Brain-Image Simulation
The generation of brain images that showed normal findings involved two
steps: First, pixel-by-pixel T1 relaxation, T2 relaxation, and proton
density-weighted brain maps (256 x 256 matrix) obtained at the level of
the lateral ventricles were generated online (software release 6.2; Philips
Medical Systems) using the image data sets from the mixed MR sequence. Second,
images simulating FLAIR, T2-weighted, and proton density-weighted sequences
were generated off-line (SUN Enterprise 5500; Sun Microsystems, Mountain View,
CA) using T1 relaxation, T2 relaxation, and proton-density pixel values from
the corresponding maps.
We developed image-simulation software using Interactive Data Language
(IDL; Research Systems, Boulder, CO). Each pixel value S(x, y) of the
generated image was calculated using the following equation:
 |
(x, y) from
the corresponding pixel values of the T1 relaxation, T2 relaxation, and proton
density maps, respectively. We selected the following parameters based on values typically used in clinical brain imaging: a TR/TE of 11,000/140 and a TI of 2600 msec; 4500/100 and 0 msec; and 2000/30 and 0 msec for the FLAIR, T2-weighted, and proton density-weighted sequences, respectively. These simulated images were expanded to a 512 x 512 matrix using linear interpolation.
Lesion Simulation
Our choice of lesion size (3 mm in diameter) was based on previous reports
that show little difference in the conspicuity of lesions larger than 5 mm in
diameter [9,
13]. The simulated lesions
were octagonal in shape, with a fixed diameter of seven pixels (3 mm in
diameter) and a total area of 37 pixels.
We chose relaxation times and proton density values of the synthesized lesions to approximate those of active multiple sclerosis plaques [6, 9, 13, 16,17,18,19,20]. First, we obtained average values for normal cortex (T1 relaxation time, 1318 msec; T2 relaxation time, 90 msec, proton density value, 355) by randomly sampling four cortical regions. We varied T1 and T2 relaxation times and proton density values across three distinct layers within each lesion (from core to periphery) to simulate lesion heterogeneity regularly encountered in the clinical setting. In particular, from core to periphery, we increased the T1 and T2 relaxation values of the three layers by 15%, 10%, and 5% of their average values, and the proton density value by 5%, 2%, and 1% of its average value (Fig. 1). Before the start of the study, a senior neuroradiologist not involved in this project judged the trilayered simulated lesions to be more realistic than homogeneous simulated lesions (Fig. 2A,2B).
|
|
|
Spatial Distribution of Lesions
The simulated brain image was divided into quadrants (right anterior, right
posterior, left anterior, left posterior) by a line through the
interhemispheric fissure and a horizontal line through the mid portions of the
lateral ventricles. Each quadrant was further divided into
cortical-subcortical, periventricular, and deep white matter regions, thus
yielding 12 possible regions for each lesion. Lesions in the
cortical-subcortical region were placed partially in cortical gray matter and
partially in subcortical white matter. Lesions in the periventricular region
were located within 5 mm of the edge of the lateral ventricle. Each simulated
image contained from zero to five lesions; an example is shown in Figure
3A,3B,3C.
We generated 12 images for each number of lesions (zero to five), resulting in
72 simulated images; these images were copied for each sequence for a total of
216 images.
|
|
|
Image Presentation
Six neuroradiologists, five of whom are board-certified, of an equivalent
training level evaluated each of the 216 simulated images in the same order.
None of the reviewers was involved in any other aspect of the study. The
reviewers received instructions that included the aim of the study,
presentation of sample images with simulated lesions for each sequence, and a
5-min practice session immediately preceding the review session. The reviewers
were also informed that a total of 216 images would be shown to them over four
sessions (54 images per session), that they could take a 5-min rest between
sessions, that some of the images would have no lesions, that no image would
have more than five lesions, that images would be presented in random order,
and that the interpretation time for one image should be minimized and must
not exceed 20 sec.
Images within each sequence were arranged in random order, but the order in
which the sequences were presented was fixed (proton density-weighted, FLAIR,
and T2-weighted); thus, although every third image was proton
density-weighted, the subsequent FLAIR and T2-weighted images did not have the
same lesion distributions. Reviewers examined one image at a time on a 21-inch
(53 cm) monitor (Sun Microsystems; viewable image size, 16 x 12 inches
[41 x 30 cm; 1600 dots [horizontal] x 1200 lines [vertical]). To
match values used in clinical brain imaging, we adjusted the window width and
level across sequences, using an Interactive Data Language routine based on
the following equation:
 |
Reviewers identified lesions using a crosshairs mouse pointer. The locations that the reviewer pointed out were recorded and judged for correctness by a scorekeeper, who sat next to the reviewer. The time needed for complete assessment of each image by the individual reviewer was also recorded.
Lesion Contrast Calculations
Average calculated contrast—which was calculated as
(SIlesion - SIbackground), where SIlesion
equals the signal intensity of the lesion and SIbackground equals
the signal intensity of the background—of the synthetic lesions for the
FLAIR, and T2-weighted brain maps at the different regions were compared with
that of a real multiple sclerosis lesion of similar size located in the white
matter of the cerebral hemisphere of a 35-year-old woman.
Data Analysis
Because we obtained binary responses (lesion present or absent) from our
reviewers, we used contingency table analysis
[21]. Each table consisted of
true-positive and true-negative rates for the following: reviewer, sequence
(FLAIR, T2-weighted, proton density-weighted), side, position, and region. We
performed the chi-square test and Fisher's exact test to assess differences in
true-positive and true-negative rates for reviewers, sequences, sides,
positions, and regions. Stepwise log-linear analyses were used to confirm
differences and to assess higher order associations, such as those between
sequence and region.
In addition, we performed repeated-measures analysis of variance to determine whether there were differences in the average times required to complete assessment of each image for the three pulse sequences. We used the Dunn test for paired analysis. Bonferroni-corrected p values less than 0.05 were considered significant for all analyses.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
True-Positive Rates
Across sequences, true-positive rates among reviewers ranged from 72% to
85% (p < 0.001) (Table
1). For each sequence, we found variability in true-positive rates
among reviewers, which reached significance for FLAIR images (p =
0.008) and for T2-weighted images (p = 0.004), but not for proton
density-weighted images (p = 0.054).
|
Across reviewers, the true-positive rates were highest for FLAIR images (89%), followed by proton density-weighted images (80%), and were lowest for T2-weighted images (72%) (p < 0.001) (Table 2). We did not detect a difference by side (right versus left) (p = 0.14) (Table 3) or by position (anterior versus posterior) (p = 0.29) (Table 4). With regard to region, the true-positive rates were highest for deep lesions (99%), followed by periventricular lesions (81%), and were lowest for cortical-subcortical lesions (62%) (p < 0.001) (Table 5). The model generated by log-linear analysis confirmed these associations and also contained a sequence-location interaction: Although reviewers had high true-positive rates for deep white matter lesions regardless of sequence, for periventricular and cortical-subcortical lesions the true-positive rates were highest for FLAIR images when compared with T2-weighted and proton density-weighted images (p < 0.001).
|
|
|
|
True-Negative Rates
Across sequences, true-negative rates among reviewers ranged from 99.6% to
100% (Table 1); although these
differences were statistically significant (p = 0.007), low expected
cell counts oppose this finding. For each sequence, we found no significant
variability in true-negative rates among reviewers for FLAIR images
(p = 0.42), T2-weighted images (p = 0.31), and proton
density-weighted images (p = 0.06). As was the case across sequences,
statistics for each sequence were computed from low expected cell counts.
Across reviewers, true-negative rates were highest for FLAIR images (100%), followed by T2-weighted images (99.9%), and were lowest for proton density-weighted images (99.6%) (p < 0.001) (Table 2). We did not detect a difference by side (p = 0.23) (Table 3) or by position (p = 0.11) (Table 4). With regard to region, the true-negative rates were highest for deep lesions (100%), followed by cortical-subcortical and periventricular lesions (99.8% each) (Table 5). The model generated by loglinear analysis confirmed these results, but no higher order interactions were added.
Assessment Times
Means and standard deviations of the times required to assess individual
images were 7.4 ± 1.2 sec, 7.7 ± 1.6 sec, and 7.8 ± 1.3
sec for FLAIR images, proton density-images, and T2-weighted images,
respectively. The assessment time was shorter for FLAIR images than for
T2-weighted images (p = 0.02); however, we did not detect significant
differences between FLAIR and proton density-weighted images (p =
0.21) or between T2- and proton density-weighted images (p = 0.33) in
assessment times.
Lesion Contrast Calculations
On the FLAIR maps, the average calculated contrast was 29.1, 33.1, and 37.8
for the synthetic lesions located in the cortical-subcortical, deep, and
periventricular regions, respectively. On the T2-weighted maps, the average
calculated contrast was 49.3, 52.4, and 57.7 for the synthetic lesions located
in the cortical-subcortical, deep, and periventricular regions, respectively.
The calculated contrast of the real multiple sclerosis lesion was 33.5 and
55.7 on the FLAIR images and T2-weighted MR images, respectively.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Our results show that detection accuracy (true-positive and true-negative rates) for simulated lesions is highest for FLAIR images, particularly for lesions in the periventricular and cortical-subcortical regions, despite the greater average calculated contrast on T2-weighted maps compared with FLAIR images. This result is in agreement with the previously reported superiority of FLAIR images in revealing subtle lesions in these regions [7,8,9,10, 13] and confirms the utility of FLAIR sequences in the evaluation of diseases that have a predilection for these regions [10, 11, 13].
We have also shown that the average assessment time for FLAIR images is shorter than that for T2-weighted images. This result underscores the usefulness of FLAIR sequences for the practicing radiologist involved in clinical brain imaging, for which efficiency and consistency are essential.
Our method for calculating T1, T2, and proton density maps from image data generated by a multiecho spin-echo interleaved with a multiecho inversion-recovery MR sequence is based on the ratios and least squares algorithm [22]. This method has been validated with phantoms and human volunteers, and the calculated T1, T2, and proton density values are in agreement with values obtained using spectrometry [22].
Our simulation of inversion-recovery and spin-echo sequences did not account for the effects of fast imaging techniques, such as rapid acquisition with relaxation enhancement (RARE) or echoplanar readouts, on image signal-to-noise ratios, spatial resolution, contrast, and artifacts [23,24,25]. Furthermore, the influence of fast readouts on the point-spread function of the simulated lesions [26] is not addressed in this study. Because FLAIR and T2-weighted MR sequences are most commonly performed with similar fast readout techniques (hybrid RARE), the effects of fast readouts on image quality and lesion visualization should be relatively fixed across sequences.
We chose lesion T1 and T2 relaxation times on the basis of values reported for active multiple sclerosis plaques [16,17,18,19,20], thus limiting our ability to generalize these results to a wide range of white matter diseases. Our simulation of lesion heterogeneity by varying T1 relaxation, T2 relaxation, and proton density values resulted in subjectively more realistic lesions—of the type regularly encountered in clinical practice [6, 9, 13, 17, 19]. However, we did not formally analyze the "naturalness" of these lesions to radiologists. Furthermore, the average calculated contrast of the synthetic lesions on FLAIR and T2-weighted maps is similar to that of a real multiple sclerosis lesion.
A limitation of this study is the lack of variability in size and signal intensity across simulated lesions, which could have inflated reviewers' sensitivities and specificities for all pulse sequences compared with the clinical setting. We chose a uniform 3-mm lesion size so that we could evaluate our techniques on relatively inconspicuous lesions. Similarly, although we distributed lesions among many regions in which multiple sclerosis plaques are commonly found, including regions adjacent to cerebrospinal fluid where lesions are more difficult to detect, we did not model the size or spatial distribution of these plaques. We could obtain these distributions from large cohorts of patients with multiple sclerosis whose brain MR examinations have been registered to a common standard. It might be the case that more realistic lesion simulations would decrease the true-positive rate differences for all factors, including sequence.
Our study design does not assess the effects of varying the window width and level, pulse sequence parameters, and lesion intensity values on lesion detection; we also did not assess infratentorial lesions. These issues are particularly important in view of reports that emphasize the occasional poor visibility of infratentorial lesions and lesions with extremely prolonged T1 relaxation times on FLAIR compared with T2-weighted images [7,8,9, 12, 14, 15].
The equation used to generate the weighted images is an approximate
solution to the Bloch equation. A more exact solution includes the following
term in the denominator [27]:
 |
In the future, we envision this software being used to answer specific questions related to visibility of simulated lesions as a function of varying pulse parameters (pulse-sequence optimization) and lesion characteristics (e.g., relaxation time, proton density, and shape and size). It may also be used to test detection skills of radiologists interested in interpreting brain MR images and, thus, could assist those who want to establish standard-of-care guidelines.
In conclusion, we have implemented a computer-based MR image and lesion simulator with which we found that FLAIR images increase detection accuracy for cortical-subcortical and periventricular lesions, and allow more efficient review, compared with T2-weighted images. Results such as these may focus subsequent clinical trials, thus reducing their expense and increasing their efficiency.
Acknowledgments
We thank Richard Swensson for providing advice regarding statistical
analysis of these data.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  E. S.C. Korf, L. R. White, P. Scheltens, and L. J. Launer Brain Aging in Very Old Men With Type 2 Diabetes: The Honolulu-Asia Aging Study Diabetes Care, October 1, 2006; 29(10): 2268 - 2274. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. H. Woo, L. P. Henry, J. Krejza, and E. R. Melhem Detection of Simulated Multiple Sclerosis Lesions on T2-weighted and FLAIR Images of the Brain: Observer Performance Radiology, October 1, 2006; 241(1): 206 - 212. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Pikus, J. H. Woo, R. L. Wolf, E. H. Herskovits, G. Moonis, A. F. Jawad, J. Krejza, and E. R. Melhem Artificial Multiple Sclerosis Lesions on Simulated FLAIR Brain MR Images: Echo Time and Observer Performance in Detection Radiology, April 1, 2006; 239(1): 238 - 245. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J H Fu, C Z Lu, Z Hong, Q Dong, Y Luo, and K S Wong Extent of white matter lesions is related to acute subcortical infarcts and predicts further stroke risk in patients with first ever ischaemic stroke J. Neurol. Neurosurg. Psychiatry, June 1, 2005; 76(6): 793 - 796. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. C. Goehde, P. Hunold, F. M. Vogt, W. Ajaj, M. Goyen, C. U. Herborn, M. Forsting, J. F. Debatin, and S. G. Ruehm Full-Body Cardiovascular and Tumor MRI for Early Detection of Disease: Feasibility and Initial Experience in 298 Subjects Am. J. Roentgenol., February 1, 2005; 184(2): 598 - 611. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A Saleh, F Wenserski, M Cohnen, G Furst, E Godehardt, and U Modder Exclusion of brain lesions: is MR contrast medium required after a negative fluid-attenuated inversion recovery sequence? Br. J. Radiol., March 1, 2004; 77(915): 183 - 188. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. J. Bastos Leite, E. C.W. van Straaten, P. Scheltens, G. Lycklama, and F. Barkhof Thalamic Lesions in Vascular Dementia: Low Sensitivity of Fluid-Attenuated Inversion Recovery (FLAIR) Imaging Stroke, February 1, 2004; 35(2): 415 - 419. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. R. Melhem, E. H. Herskovits, K. Karli-Oguz, X. Golay, D. A. Hammoud, B. J. Fortman, F. M. Munter, and R. Itoh Defining Thresholds for Changes in Size of Simulated T2-Hyperintense Brain Lesions on the Basis of Qualitative Comparisons Am. J. Roentgenol., January 1, 2003; 180(1): 65 - 69. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I L. Tan, R. A. van Schijndel, P. J. W. Pouwels, H. J. Ader, and F. Barkhof Serial Isotropic Three-Dimensional Fast FLAIR Imaging: Using Image Registration and Subtraction to Reveal Active Multiple Sclerosis Lesions Am. J. Roentgenol., September 1, 2002; 179(3): 777 - 782. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
