HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) All Versions of this Article: 2391050221v1 239/1/246    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Benner, T. Articles by Sorensen, A. G. Search for Related Content PubMed PubMed Citation Articles by Benner, T. Articles by Sorensen, A. G.

Comparison of Manual and Automatic Section Positioning of Brain MR Images¹

¹ From the Department of Radiology (T.B., J.J.W., A.J.W.v.d.K., B.F., A.G.S.) and General Clinical Research Center (M.G.V.), Massachusetts General Hospital, Athinoula A. Martinos Center, Harvard Medical School, 149 13th St, Rm 2301, Charlestown, MA 02129; Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, Mass (B.F.); and Department of Neurology and Brain Tumor Center, Massachusetts General Hospital, Boston, Mass (F.H.H.). Received February 8, 2005; revision requested April 6; revision received May 4; final version accepted June 13. Supported in part by the National Center for Research Resources (P41 RR14075, R01 RR16594-01A1 and BIRN Morphometric Project BIRN002, U24 RR021382, R21 EB02530), the National Institute for Biomedical Imaging and Bioengineering (R01 EB001550), the National Cancer Institute (NCI 5T32CA09502), and the Mental Illness and Neuroscience Discovery Institute. Address correspondence to T.B. (e-mail: thomas.benner{at}nmr.mgh.harvard.edu).

	ABSTRACT

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ADVANCES IN KNOWLEDGE References

 
The study protocol was approved by the institutional review board and was in full compliance with HIPAA guidelines. Informed consent was obtained from all patients. The purpose of this study was to prospectively compare intra- and intersubject variability of manual versus automatic magnetic resonance (MR) imaging section prescription. In two examinations, T2-weighted series were acquired with both methods. All intrasubject and three of six intersubject section prescription variances were significantly higher for manual prescription (P < .01). Root mean square errors confirmed better coregistration of the automated approach (P < .001). Automatic section prescription leads to improved reproducibility of imaging orientations for intra- and intersubject series in clinical practice.

© RSNA, 2006

	INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ADVANCES IN KNOWLEDGE References

 
Imaging is playing an increasing role in drug development (1). The efficacy of many therapies, from Alzheimer disease to oncology applications, is assessed by using serial imaging studies to determine changes in lesion size or atrophy progression. This follows clinical practice, where the treatment of many patients is influenced by how lesions change on imaging studies over time: Doses of drugs may be lowered or medication changes may be made on the basis of only small size changes seen at serial magnetic resonance (MR) imaging (2–5).

However, most clinical brain MR imaging is still performed with a series of sections obtained through the brain, where the in-plane resolution is higher than the through-plane resolution. As a result, the positioning of these sections—the so-called section prescription—can have a substantial effect when the change in lesion size is smaller than or similar to the section thickness, which is a common occurrence. Nonstandard section prescriptions have been shown to pose a particular problem when imaging is used as a formal end point in longitudinal clinical trials where the detection of small changes is important, such as in patients with multiple sclerosis (2–5). For multiple sclerosis, the expected yearly change in lesion load is 10% (5), and lesion load changes observed on yearly T2-weighted MR series are used as secondary end points to monitor the effect of treatment in large-scale clinical trials (4,6,7).

Reliable section prescription is important for quantitative analysis of MR data in general (8), as well as in other diseases besides multiple sclerosis, such as brain atrophy (9) and tumors (10). For example, misregistration of series makes comparison of Response Evaluation Criteria in Solid Tumors difficult over multiple time points during therapy (11). It has also been shown that the error from serial studies (interseries error), on average, exceeds the intraobserver error (12). Lower intraobserver variability for repeated measurement compared with variability due to repositioning has also been reported (7,13).

Clinical brain MR imaging series are generally obtained by a trained MR technologist who manually selects the section prescription on the basis of a three-plane localizer series. Anatomic features visible on the localizer images are selected as reference points. Common references to which the series volume gets aligned are the anterior commissure–posterior commissure line and the midline of the brain. Manual section prescription can be time-consuming if a precise alignment is required or if the patient is in an unusual position in the MR imager. In addition, not all MR imagers allow oblique sections in all three planes, which thereby makes accurate section prescription impossible without repositioning the patient. Finally, manual section prescriptions are subject to both intraoperator and interoperator variability. As a consequence, brain MR imaging series are obtained in a nonstandardized fashion. A number of methods to decrease variability of section prescription have been presented previously (5,7,14). Methods for automatic section prescription were described by Itti et al (15), Gedat et al (16), and Welch et al (17).

We hypothesized that automatic section prescription results in more reliable positioning for intrasubject as well as for intersubject series acquisition. Thus, the purpose of our study was to prospectively compare intra- and intersubject variability of manual versus automatic MR imaging section prescription.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ADVANCES IN KNOWLEDGE References

 
Two authors (A.J.W.v.d.K and B.F.) receive licensing fees from and one author (B.F.) also provides consulting services to CorTechs Laboratories for MR image acquisition and postprocessing technology. The nonemployee and nonconsultant authors had control of the data and information presented in this article.

Patients
The institutional review board approved the study protocol. Informed consent was obtained from all patients after the nature of the procedure had been fully explained. The study was in full compliance with the Health Insurance Portability and Accountability Act guidelines. The patient population for this study was obtained from patients who were already scheduled to undergo a clinical head examination at an outpatient MR imaging facility during the time period between April 2003 and July 2004. Patients were not selected according to any specific clinical indication; however, coincidentally, patients scheduled for clinical examinations tended to be referred by physicians specializing in aging or neuro-oncology. After inclusion in the study, each patient was scheduled to undergo a second examination that took place at least 2 weeks after the first examination.

Initially, 30 patients were planned to be included in this study. However, an MR imager upgrade resulted in early termination of the study after 17 patients were included. Of these, three patients did not return for follow-up examination, and one patient was accidentally scheduled to undergo MR imaging with a different imager. These four patients were therefore excluded from the study. The mean age of the remaining 13 patients (eight women and five men) was 50.5 years (range, 21–84 years). The reasons for the clinical examination of the head were as follows: tumor follow-up (n = 7), suspicion of tumor (n = 2), or memory disorders (n = 4). The mean number of days between the first and second visit was 23.0 days (range, 14–70 days).

MR Imaging Protocol
Imaging was performed by using a 1.5-T MR system (Magnetom Sonata, software version VA21B; Siemens Medical Solutions, Erlangen, Germany). In each examination, the following block of four series was performed twice: (a) three-plane scout series, (b) automatic alignment scout series, (c) T2-weighted fast spin-echo series with manual section prescription, and (d) T2-weighted fast spin-echo series with automatic alignment. The patient was removed from the MR imager and was asked to get off the imager table between the first and second block, whereupon the head pillow was repositioned to ensure that the patient's head would not fall back into the previous position. The repetition of a series block within an examination (ie, the second block) was performed to minimize the influence of any additional factors—such as lesion growth or imager instabilities from the first to the second examination—on section prescription and succeeding coregistration. In the first examination, additional clinical series were obtained as needed for the requested medical evaluation. In the second examination, only two series blocks were performed. The order of the T2-weighted fast spin-echo series with manual and automatic section prescription was not randomized to exclude any bias of the technician owing to the fact that the results of the automatic section prescription would be presented on the monitor before performing the manual section prescription. The MR technician performed the manual alignment in the same time frame as in routine practice and was encouraged to make section prescriptions as reproducible as possible.

The automatic alignment scout parameters were as follows: three-dimensional fast low-angle shot sequence, 2.4/1.13 (repetition time msec/echo time msec), flip angles of 2° and 6°, field of view of 320 mm, isotropic voxel size of 2.5 mm, matrix of 96 x 128, 128 sections with 75% section encoding, and imaging time of 46 seconds. The parameters for the T2-weighted fast spin-echo series were as follows: 4230/95, field of view of 230 mm, 23 sections, section thickness of 5 mm with gap of 1 mm, matrix size of 256 x 192, and imaging time of 2 minutes 38 seconds.

The automatic alignment method was based on a multispectral atlas-driven image registration technique (18,19). The atlas, which was constructed from a set of high-spatial-resolution anatomic series, contains probabilities of each voxel belonging to one of several tissue classes (eg, gray matter, white matter, cerebrospinal fluid, or skull). Given the sequence parameters, a rigid body rotation matrix is calculated that maximizes the probability of the observed voxel intensities of the localizer images given the tissue classification in the atlas. In the current implementation, this calculation requires about 12 seconds (1.8-GHz Xeon processor; Intel, Santa Clara, Calif). The calculation of the rigid body transformation matrix from the two flip angle three-dimensional localizer data was always successful—that is, automatic alignment was possible in all cases.

Image Processing
To determine the accuracy of manual and automatic section positioning, T2-weighted MR imaging volumes were retrospectively coregistered to each other (intrasubject analysis) and to a template (intersubject analysis). In the ideal case of perfect section positioning, all T2-weighted imaging volumes would match; in other words, all coregistration parameters would be zero. For intrasubject data analysis, coregistration of T2-weighted MR series was performed for intraexamination and interexamination series. This was done separately for the manually prescribed and the automatically aligned series, which resulted in a total of 12 coregistrations (four intraexamination and eight interexamination) per patient. For intersubject analysis, all T2-weighted imaging series (eight per patient) were coregistered to the Montreal Neurological Institute T2-weighted imaging template (average of 152 healthy control subjects). An image registration tool (FLIRT, version 5.0; Oxford Centre for Functional Magnetic Resonance Imaging of the Brain, Oxford, England) (20) was used for fully automated coregistration, with rigid body registration mode (six degrees of freedom [ie, three translations and three rotations]) and mutual information as cost function. Translations and rotations were determined from the coregistration matrices. The axes for translations and rotations relative to the head position were defined as follows: x-axis, right-left; y-axis, anterior-posterior; and z-axis, superior-inferior.

To assess the quality of the coregistration and as a further performance indicator of both methods, root mean square error (RMSE) values were calculated for each pair of initial and follow-up series. Signal intensities of all corresponding voxels in the two volumes were used for RMSE calculation, as follows:

where x_i and y_i are signal intensities of voxel i of the two volumes and N is the total number of voxels. RMSE values were calculated before and after coregistration, for intrasubject analysis only. To reduce contributions of tissue further from the brain and to eliminate coregistration artifacts at the edges of the volume, RMSE calculations were limited to a sphere (radius, 80 mm) centered in the middle of the volume.

Statistical Analysis
Descriptive statistics—minimum, maximum, and standard deviation—were determined for intrasubject and intersubject coregistration parameters for manual and automatic section prescription, respectively. In addition, box plots were generated for these parameters. Analyses of variance (ANOVA) were performed for absolute values of translation and rotation, with "subject" (13 levels), "method" (two levels, manual and automatic section prescription), and "image pairing" as factors. This was done for intrasubject and intersubject data analysis. Image pairings are the coregistration combinations of examination and series resulting in six and four levels for intrasubject and intersubject analysis, respectively. Box plots were generated for RMSE values before and after coregistration, split by manual and automatic section prescription. An ANOVA was performed for the log reduction in RMSE—that is, ln(RMSE_after/RMSE_before)—with "subject," "method," and "image pairing" as factors. Results were considered statistically significant for P values less than .05. All statistical analyses were performed by using R software version 2.0.1 (The R Foundation for Statistical Computing, Vienna, Austria) (21).

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ADVANCES IN KNOWLEDGE References

 
The variation of section orientation was larger for the manual prescription than for the automatic alignment (Fig 1). After coregistration of the follow-up series to the initial series, the resulting images (bottom half of each panel) matched the initial series better for manual section prescription, while there was no noticeable change in orientation for the automatically aligned MR images. All MR images acquired with automatic section prescription were well aligned around all three axes (Fig 2). For the manually prescribed images, more variation in the rotation around the right-left axis, as well as in the overall inferior-superior position, could be seen, resulting in less comparable images.

View larger version (127K): [in this window] [in a new window] [Download PPT slide]   Figure 1: Top row of each panel shows section 10 of original transverse T2-weighted MR images in patient 8 for all four series with manual (M1–M4) and automatic (A1–A4) section prescription. Bottom row of each panel shows images after coregistration of follow-up series (M2--M4, A2–A4) to the initial series (M1, A1). Note higher variation in manually aligned sections versus automatically aligned sections, as well as slight blurring caused by the coregistration procedure.

View larger version (93K): [in this window] [in a new window] [Download PPT slide]   Figure 2: Middle section of transverse T2-weighted MR images from all 13 patients for manual (top panel) and automatic (bottom panel) section prescription. Automatic prescription images show similar anatomic section location, good hemispheric symmetry, and vertical midline. Manually prescribed images show deviations from vertical midline because imager software did not allow manual rotation within the section plane. Hemispheric symmetry with manual prescription is comparable to that with automatic prescription, but more variation in rotation around right-left axis and in overall inferior-superior position can be seen.

View larger version (35K): [in this window] [in a new window] [Download PPT slide]   Figure 3: Box plots of translations (dx, dy, and dz on x-, y-, and z-axis, respectively, in millimeters) and rotations (ax, ay, and az on x-, y-, and z-axis, respectively, in degrees) resulting from coregistration to intrasubject T2-weighted MR images for section prescription performed manually and automatically. The ranges of translations and rotations for manual prescription clearly exceed those for automatic prescription. Factor "method" was found to be significant (P < .001) for all translations and rotations.

View larger version (24K): [in this window] [in a new window] [Download PPT slide]   Figure 4: Box plots of RMSE values (in arbitrary units [a.u.]) before and after intrasubject coregistration by using the FLIRT tool for manual and automatic section prescription. Note the greater reduction in RMSE from before to after coregistration for prescription performed manually compared with that performed automatically. Factor "method" was found to be significant (P < .001) for the log reduction in RMSE.

View larger version (36K): [in this window] [in a new window] [Download PPT slide]   Figure 5: Box plots of translations (dx, dy, and dz on x-, y-, and z-axis, respectively, in millimeters) and rotations (ax, ay, and az on x-, y-, and z-axis, respectively, in degrees) resulting from coregistration to a T2-weighted imaging template for section prescription performed manually and automatically. All ranges of translations and rotations for manual prescription—with the exception of y-axis translations—exceed those for automatic prescription. Factor "method" was found to be significant for translations along the x-axis (P < .01) and rotations around the x-axis and z-axis (P < .001).

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ADVANCES IN KNOWLEDGE References

There are a number of interesting features to our results. We note that intrasubject variance was lower than intersubject variance. Specifically, the results of intrasubject data analysis showed a clear improvement in image alignment as measured by means of variability in translations and rotations when the section prescription was performed automatically compared with when it was performed by an MR technologist. On the other hand, while still significant for three of the coregistration parameters, intersubject data analysis did not show the factor "method" to be significant for translations along the y- and z-axes and rotations around the y-axis. We suspect that our patient population made intersubject coregistration particularly challenging because many of our patients had large brain tumors, and the template from which coregistration parameters were calculated is based on normal brains. Significant improvement in the variance of rotations around the z-axis can be attributed to the fact that the version of the MR imager graphical user interface software we used did not allow adjustment of in-plane rotations when performing the section prescription manually, while automatic section prescription does take in-plane rotations into account. In other words, for manual section positioning, the observed rotations around the z-axis represent how accurately the technologist was able to place the patient's head relative to the MR imager bore. The rotations around the y-axis for manual alignment show much smaller variation compared with the other axes. We suspect this is because it is comparatively easy for a technologist to perform a section prescription in the coronal plane given a coronal localizer image with clearly visible hemispheric fissure.

We used an off-line coregistration tool to quantify the performance of both alignment methods, and statistical analysis showed that the factor "method" was highly significant for the reduction in RMSE values—that is, the overall magnitude of change was much larger when the sections were prescribed manually. However, we note that the two methods used to align brain images—automatic alignment and the FLIRT program—are fundamentally different, and therefore the residual variance reported by using the FLIRT tool may not actually represent true error in alignment of the brain itself. This is because, while automatic alignment takes only brain tissue into account, FLIRT uses the whole data set for matching, thereby including neck and other tissue in its alignment, and these may or may not stay in a constant spatial relationship to the brain. To minimize the influence of non–brain tissue on RMSE measurements, the latter were limited to a sphere in the middle of the series volume. In any case, small improvements in RMSE values after application of the FLIRT tool are therefore not necessarily an indicator for better matching of brain tissue only, but reflect overall better matching of the head.

Our method compares favorably with alternative approaches. A variety of methods to decrease the variability of section prescription have been presented previously. Some of these methods require special hardware, such as face masks (5) or other hardware (22), and may be time consuming and uncomfortable for the patient and are therefore not practical in clinical routine. Other methods that are based on the careful acquisition of a series of single-section images (5,7,14,23) can be time consuming as well. As a procedure to increase the accuracy of lesion volume measurements, reduction of section thickness has been suggested (4,5,24,25); however, this approach has not been integrated into clinical routine.

Methods similar to the one used in our study were described by Itti et al (15), Gedat et al (16), and Welch et al (17). The method described by Itti et al (15) is based on a rapid MR imaging pilot series, brain surface segmentation, and coregistration to a reference surface. The standard deviations of rotation angles found in our study (0.9°, 1.0°, and 0.7°) compare favorably with those reported by Itti et al (2.70°, 1.35°, and 0.76°). Gedat et al (16) coregistered three-plane localizer data and applied the resulting transformation to succeeding series. This method is less accurate if the difference in positions between the reference and target localizer data is too great. The reported standard deviations of translations and rotations compare well with our results. The method by Welch et al (17) is based on linear and spherical navigator echoes. While the method presented in our study relies on one statistical atlas that can be applied to all subjects in any MR imager, the methods used by Gedat et al and Welch et al can only be applied to the same subject and require storage of a small amount of data for each subject in the MR imager or in a picture archiving and communications system. To be able to use these methods on other imagers, the data would have to be transferred, which makes it less practical to use these methods by using multiple MR imagers within one hospital or at different hospitals. The times needed for these alternative methods and for our approach are comparable.

Using a registration procedure either to a statistical atlas or to another anatomic reference (15) requires that there be sufficient overlap between the reference and the subject to be imaged for the procedure to work. The presented method is therefore expected to be less accurate in cases in which the target brain anatomy is strongly different from that used for generating the reference atlas, such as in patients with large tumors or patients who have undergone brain surgery. However, we have shown that the method still works reliably even in patients with tumors. It is also possible to create atlases that include these patient groups to achieve better matching. Given the flexibility of being able to set the target series volume in the protocol, application of this method to other MR techniques, such as spectroscopy, is straightforward.

While the presented approach corrects for examination-to-examination variability, it does not correct for variability introduced, for example, by changes in patient position from series to series. Although this could be resolved by acquiring a separate three-dimensional localizer series before each series, this is not practical because of the additional imaging time required. The same applies to the methods presented by Itti et al (15) and Gedat et al (16). More rapid acquisition types, like navigator echo imaging (17), that quickly depict in a few seconds changes from the previous series but do not compare to the full template might be better suited for this purpose. A combination of both the template automatic alignment technique with a quick series-to-series correction could result in an imaging protocol with consistent images obtained over the course of a study and standardized section prescription over the course of multiple studies.

This study is limited by a number of factors. First, for the most accurate three-dimensional coregistration, use of isotropic high-spatial-resolution MR series would be preferable compared with the T2-weighted MR series used in this study. Nevertheless, the FLIRT program performed well on the T2-weighted MR data sets, and acquisition of high-spatial-resolution series would require a much longer imaging time in addition to the clinically mandated series. Second, the study population included mostly patients with tumors or memory disorders. The performance of automatic section positioning in patients with other diseases might therefore deviate from the performance shown here. Still, taking into account the fact that automatic section positioning worked well even in patients with substantial changes of the brain structure, a failure of the method would not be expected in other patients. Third, the automatic method was compared with the performance of only one technician, and the ability to perform an accurate section prescription is likely to vary from technician to technician. However, we have no reason to believe that this one technician performed substantially worse than the average technician, and the highly significant results make a significant difference still likely even in the case of a better-performing technician. A comparison using multiple technicians would require many more patients to undergo acquisition of additional series and is beyond the scope of this article.

In summary, we found that automatic section prescription based on a statistical atlas was robust in clinical practice and led to improved reproducibility of series orientations and, in turn, less variance in clinical MR brain imaging for both intrasubject and intersubject studies.

	ADVANCES IN KNOWLEDGE

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ADVANCES IN KNOWLEDGE References

 

• Automatic prospective section prescription based on a statistical atlas of the brain is feasible in clinical practice and can be performed successfully in patients.
  
• Intrasubject and intersubject variance of section prescription can be reduced by using automatic section prescription rather than manual section prescription.
  

	ACKNOWLEDGMENTS

 
We thank Mary T. Foley, BS, for technical assistance.

	FOOTNOTES

 

Abbreviations: ANOVA = analysis of variance • RMSE = root mean square error

See Materials and Methods for pertinent disclosures.

Author contributions: Guarantors of integrity of entire study, T.B., J.J.W., A.G.S.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; maunscript final version approval, all authors; literature research, T.B., B.F., F.H.H., A.G.S.; clinical studies, T.B., J.J.W., A.J.W.v.d.K., F.H.H., A.G.S.; statistical analysis, T.B., M.G.V.; and manuscript editing, all authors

	References

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ADVANCES IN KNOWLEDGE References

 

1. Smith JJ, Sorensen AG, Thrall JH. Biomarkers in imaging: realizing radiology's future. Radiology 2003;227:633–638.[Abstract/Free Full Text]
2. Gawne-Cain ML, Webb S, Tofts P, Miller DH. Lesion volume measurement in multiple sclerosis: how important is accurate repositioning? J Magn Reson Imaging 1996;6:705–713.[Medline]
3. Goodkin DE, Ross JS, Medendorp SV, Konecsni J, Rudick RA. Magnetic resonance imaging lesion enlargement in multiple sclerosis: disease-related activity, chance occurrence, or measurement artifact? Arch Neurol 1992;49:261–263.[Abstract/Free Full Text]
4. Molyneux PD, Tofts PS, Fletcher A, et al. Precision and reliability for measurement of change in MRI lesion volume in multiple sclerosis: a comparison of two computer assisted techniques. J Neurol Neurosurg Psychiatry 1998;65:42–47.[Abstract/Free Full Text]
5. Filippi M, Marciano N, Capra R, et al. The effect of imprecise repositioning on lesion volume measurements in patients with multiple sclerosis. Neurology 1997;49:274–276.[Abstract/Free Full Text]
6. Filippi M, Horsfield MA, Tofts PS, Barkhof F, Thompson AJ, Miller DH. Quantitative assessment of MRI lesion load in monitoring the evolution of multiple sclerosis. Brain 1995;118:1601–1612.[Abstract/Free Full Text]
7. Rovaris M, Gawne-Cain ML, Sormani MP, Miller DH, Filippi M. The effect of repositioning on brain MRI lesion load assessment in multiple sclerosis: reliability of subjective quality criteria. J Neurol 1998;245:273–275.[CrossRef][Medline]
8. Plante E, Turkstra L. Sources of error in the quantitative analysis of MRI scans. Magn Reson Imaging 1991;9:589–595.[CrossRef][Medline]
9. Gunter JL, Shiung MM, Manduca A, Jack CR Jr. Methodological considerations for measuring rates of brain atrophy. J Magn Reson Imaging 2003;18:16–24.[CrossRef][Medline]
10. Clarke LP, Velthuizen RP, Clark M, et al. MRI measurement of brain tumor response: comparison of visual metric and automatic segmentation. Magn Reson Imaging 1998;16:271–279.[CrossRef][Medline]
11. Husband JE, Schwartz LH, Spencer J, et al. Evaluation of the response to treatment of solid tumours: a consensus statement of the International Cancer Imaging Society. Br J Cancer 2004;90:2256–2260.[Medline]
12. Simon JH, Scherzinger A, Raff U, Li X. Computerized method of lesion volume quantitation in multiple sclerosis: error of serial studies. AJNR Am J Neuroradiol 1997;18:580–582.[Abstract]
13. Tan IL, van Schijndel RA, van Walderveen MA, et al. Magnetic resonance image registration in multiple sclerosis: comparison with repositioning error and observer-based variability. J Magn Reson Imaging 2002;15:505–510.[CrossRef][Medline]
14. Miller DH, Barkhof F, Berry I, Kappos L, Scotti G, Thompson AJ. Magnetic resonance imaging in monitoring the treatment of multiple sclerosis: concerted action guidelines. J Neurol Neurosurg Psychiatry 1991;54:683–688.[Abstract/Free Full Text]
15. Itti L, Chang L, Ernst T. Automatic scan prescription for brain MRI. Magn Reson Med 2001;45:486–494.[CrossRef][Medline]
16. Gedat E, Braun J, Sack I, Bernarding J. Prospective registration of human head magnetic resonance images for reproducible slice positioning using localizer images. J Magn Reson Imaging 2004;20:581–587.[CrossRef][Medline]
17. Welch EB, Manduca A, Grimm RC, Jack CR Jr. Interscan registration using navigator echoes. Magn Reson Med 2004;52:1448–1452.[CrossRef][Medline]
18. van der Kouwe AJ, Benner T, Fischl B, et al. On-line automatic slice positioning for brain MR imaging. Neuroimage 2005;27:222–230.[CrossRef][Medline]
19. Fischl B, Salat DH, Busa E, et al. Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain. Neuron 2002;33:341–355.[CrossRef][Medline]
20. Jenkinson M, Smith S. A global optimisation method for robust affine registration of brain images. Med Image Anal 2001;5:143–156.[CrossRef][Medline]
21. R Development Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing, 2005.
22. Oshio K, Panych LP, Guttmann CR. A simple noninvasive stereotactic device for routine MR head examinations. J Comput Assist Tomogr 1996;20:588–591.[CrossRef][Medline]
23. Filippi M, Rovaris M, Sormani MP, et al. Intraobserver and interobserver variability in measuring changes in lesion volume on serial brain MR images in multiple sclerosis. AJNR Am J Neuroradiol 1998;19:685–687.[Abstract]
24. Molyneux PD, Tubridy N, Parker GJ, et al. The effect of section thickness on MR lesion detection and quantification in multiple sclerosis. AJNR Am J Neuroradiol 1998;19:1715–1720.[Abstract]
25. Filippi M, Horsfield MA, Campi A, Mammi S, Pereira C, Comi G. Resolution-dependent estimates of lesion volumes in magnetic resonance imaging studies of the brain in multiple sclerosis. Ann Neurol 1995;38:749–754.[CrossRef][Medline]

This Article Abstract Figures Only Full Text (PDF) All Versions of this Article: 2391050221v1 239/1/246    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Benner, T. Articles by Sorensen, A. G. Search for Related Content PubMed PubMed Citation Articles by Benner, T. Articles by Sorensen, A. G.