Image-Guided Surgery

Meet Am Soc Stereotact Funct Neurosurg, New York, N.Y., 2003

Stereotact Funct Neurosurg 2003;80:140145
DOI: 10.1159/000075175

Stereotactic Accuracy of a 3-Tesla

Magnetic Resonance Unit
Hooman Azmi Michael Schulder
Department of Neurological Surgery, Neurological Institute of New Jersey,
New Jersey Medical School, Newark, N.J., USA

Key Words
Stereotactic accuracy W 3-Tesla magnetic resonance unit W MRI

Abstract
Recently, magnetic resonance imagers (MRIs) with 3-tesla magnets
were approved for clinical use. The spatial accuracy of these high-resolution
scanners has yet to be proven. In the present study, a computed tomography (CT)- and MRI-compatible phantom was scanned on a CT, a 1.5-tesla
MRI and a 3-tesla MRI scanner. The model was registered to the images
using an infrared-based surgical navigation system. The distance between
the predicted position of the navigation probe tip and the actual target on
the image was measured on the x, y and z axes for 13 points on each image.
Error was compared across imaging modalities, peripheral versus central
targets and along each axis. We found that 3-tesla MRI scans are accurate as
stereotactic data sets.
Copyright 2003 S. Karger AG, Basel

Introduction

Magnetic resonance images (MRIs) have been used for stereotactic localization in many patients around the world. Stereotactic images based on MRI are
used every day for tumor resection, radiosurgery and functional procedures.
While several authors have presented data on the accuracy of MRI for stereotactic

2003 S. Karger AG, Basel

10116125/03/08040140$19.50/0

Fax + 41 61 306 12 34
E-Mail karger@karger.ch Accessible online at:
www.karger.com/sfn
www.karger.com

Michael Schulder, Department of Neurological Surgery

Neurological Institute of New Jersey
90 Bergen Street, Suite 8100, Newark, NJ 07103 (USA)
Tel. +1 973 972 2907, Fax +1 973 972 2333
E-Mail schulder@umdnj.edu

Downloaded by:
NYU Medical Center Library
128.122.253.212 - 5/27/2015 4:36:18 AM

ABC

Fig. 1. MRI- and CT-compatible phantom with various internal targets.

localization, the question of image distortion remains, and the possibility of targeting errors that could affect patient outcomes is not far-fetched. The question of
accuracy is once again brought to the forefront as 3-tesla MRI scanners have
received approval for clinical use by the FDA. The higher signal to noise ratio of
these machines promises images with increased resolution and better target definition. The ability to improve stereotactic targeting with such MRI applications
as functional imaging and spectroscopy makes the use of 3-tesla scanners more
attractive. As their use becomes more widespread, it is important to establish the
accuracy of this imaging modality.
The purpose of this study was to compare the spatial accuracy of images
acquired on a 3-tesla MRI scanner to those obtained with computed tomography
(CT) and 1.5-tesla MRI scanners.

Materials and Methods

We obtained a CT- and MRI-compatible phantom from Radionics, Inc. (Burlington,
Mass., USA). The phantom had internal objects for targeting (fig. 1), and we added fiducials
to the surface of the skull to serve as peripheral targets. 13 targets in all were selected: 7
internal targets, and 6 peripheral targets. The phantom was then placed in a Radionics Cos-

Stereotact Funct Neurosurg

2003;80:140145

141

Downloaded by:
NYU Medical Center Library
128.122.253.212 - 5/27/2015 4:36:18 AM

Stereotactic Accuracy of a 3-Tesla Magnetic

Resonance Unit

Fig. 2. Error was measured on three axes as the difference between the position of the probe
as demonstrated by the StealthStation surgical navigation system and the targeted point.

man-Roberts-Wells (CRW) localizer and a CT scan was obtained (field of view [FOV]
32 cm, gantry 0, slice thickness 1 and 3 mm). We then placed the phantom in the MR
localizer, which consisted of a watertight tank and localizer bars. We filled the tank with
sterile water as directed and the localizer bars were filled with a 1:20 diluted solution of
gadolinium. The phantom was scanned on a GE 1.5-tesla MR unit (FOV 26 cm, matrix
512 ! 512, TR 566 ms, TE 11 ms) and a Siemens 3-tesla MR unit (FOV 27 cm, matrix
512 ! 512, TR 644 ms, TE 14 ms).
The data were then loaded into a StealthStation infrared LED-based surgical navigation unit (Medtronic/SNT, Louisville, Colo., USA). We then registered the phantom into the
computer using the fiducial markers. Once the phantom was registered, we used the Stealth
probe to identify the targets. We measured the error as the distance between the position of
the probe in virtual space and the target on the image. The distance was measured on three
axes (fig. 2). Figure 3 summarizes these data.
To see if there was any statistical significance in comparing the error measurements,
pairs of modalities were compared using Students t test. In a similar manner, we also compared the difference in errors in peripheral and central targets. Comparison of the errors on
different axes was done using a weighted one-way analysis of variance. The statistical analysis was done online using Vassar Universitys statistical website.

Stereotact Funct Neurosurg

2003;80:140145

Azmi/Schulder

Downloaded by:
NYU Medical Center Library
128.122.253.212 - 5/27/2015 4:36:18 AM

142

Fig. 3. Demonstration of the mean error in various modalities. Each bar represents a specific imaging modality, and the y axis represents the mean error measured in millimeters.
Error bars represent standard deviations.

Results

We did not find any statistical significance in comparing the errors of CT

versus all MRIs (p = 0.12), CT versus 1.5-tesla MRI (p = 0.45), or 1.5-tesla MRI
versus 3-tesla MRI (p = 0.28). In comparing CT versus 3-tesla MRI, we found that
MRI was more accurate (p = 0.03). This may have been related to observer bias.
The study was not blinded and it is possible that we were more careful in measuring the 3-tesla data and that this resulted in more accurate measurements using
the images acquired on the 3-tesla MRI. There was no statistically significant
difference in the errors obtained while comparing central versus peripheral targets on CT (p = 0.39), 1.5-tesla MRI (p = 0.41), or 3-tesla MRI (p = 0.51).
We also compared errors on different axes and did not find any statistically
significant error except in the comparison of the z axis. There was statistical significance in comparison of the 1.5-tesla 1-mm scans to the 3-tesla 1-mm (p !
0.01), 3-tesla 3-mm (p ! 0.05) and also the 1.5-tesla 3-mm scans (p ! 0.01). The
scan data for this particular image may have been prone to error in the z dimension. The scan was incomplete in that the highest part of the phantom was not
imaged; this may have resulted in falsely high error on the z axis. In addition, the z
axis is not imaged directly, and it is possible that the error in this dimension was
due to the computer reformatting of these images.

Stereotact Funct Neurosurg

2003;80:140145

143

Downloaded by:
NYU Medical Center Library
128.122.253.212 - 5/27/2015 4:36:18 AM

Stereotactic Accuracy of a 3-Tesla Magnetic

Resonance Unit

Discussion

CT has been a benchmark imaging technique for stereotactic localization.

Linear X-ray beams are not subject to distortion and the CT scan provides an
image with accurate targets [1].
MRI offers excellent neuroanatomic resolution, the ability to obtain direct
nonreformatted multiplanar imaging and reduced artifacts from bone and from
stereotactic frames [2]. These factors make MRI an attractive tool for imaging of
stereotactic targets.
Despite these advantages, the use of MRI for stereotactic localization has had
some obstacles: restricted availability, longer image acquisition times compared
with CT scanning and higher costs [2]. Most importantly, however, concerns
regarding the use of MRI for stereotactic localization have been raised by questions about the spatial accuracy of MRI [1, 3].
Errors in the accuracy of MRI can be from several different sources. Gradient field nonlinearities are produced from imperfections with the linear gradients generated by coils in the MR scanner. Most MR scanners, however, have
incorporated software and other mechanisms to correct for this kind of error [4].
The other type of error is from resonance offset. This results from changes in the
frequency of magnetic resonance signals caused by mechanisms other than linear
imaging gradients [5]. There are two sources of resonance errors, one being from
chemical shifts resulting from the different chemical makeup of various tissues.
The other, which is probably the most important source of error [5], arises from
magnetic field inhomogeneities induced by the scanner, the object, or both. Field
inhomogeneities caused by a specific object are related to that specific object and
are harder to predict [4].
Despite these concerns, many centers are using MRI as the primary imaging
technique for stereotactic localization, and there are several studies that support
the accuracy of MRI in stereotactic targeting [2, 6, 7].
More recently, with the approval of the 3-tesla MR scanner by the FDA, a
new tool has been introduced for imaging patients. The higher signal to noise ratio
allows the new 3-tesla scanner to acquire images with better resolution. In fact, a
3-tesla scanner theoretically provides twice the resolution of a 1.5-tesla scanner.
This allows for better imaging of the lesion and better definition of the lesion and
the important structures. The improved resolution of this modality makes it an
attractive option for stereotactic localization. However, no study to date has
assessed the stereotactic accuracy of a 3-tesla scanner. With so many centers using
MRI as the sole imaging method for stereotactic protocols, the question of accuracy becomes an important one.

Stereotact Funct Neurosurg

2003;80:140145

Azmi/Schulder

Downloaded by:
NYU Medical Center Library
128.122.253.212 - 5/27/2015 4:36:18 AM

144

Conclusion

In comparing CT, 1.5-tesla MRI and 3-tesla MRI, we did not find any statistical significance in the error measurements of the different modalities. We also
compared error measurements in peripheral and central targets and again did not
find any statistically significant error. In addition, in comparing the error in different axes, we found the imaged axes to have no significant error.
In conclusion, it is important to note that image distortion is a real phenomenon that occurs with MRI. However, each institution should be able to calibrate
their scanners in order to eliminate predictable sources of error.
We have shown that 3-tesla MRI provides an accurate data set for stereotactic localization. Further work needs to be done to validate these results on stereotactic frames, and subsequently on patients.

References

3
4

Peters TM, Clark J, Pike B, Drangova M, Olivier A: Stereotactic surgical planning with magnetic
resonance imaging, digital subtraction angiography and computed tomography. Appl Neurophysiol
1987;50:3338.
Kondziolka D, Dempsey PK, Lunsford LD, Kestle JR, Dolan EJ, Kanal E, Tasker RR: A comparison between magnetic resonance imaging and computed tomography for stereotactic coordinate
determination. Neurosurgery 1992;30:402407.
Hassenbusch SJ, Pillay PK, Barnett GH: Radiofrequency cingulotomy for intractable cancer pain
using stereotaxis guided by magnetic resonance imaging. Neurosurgery 1990;27:220223.
Sumanaweera TS, Adler JR Jr, Napel S, Glover GH: Characterization of spatial distortion in magnetic resonance imaging and its implications for stereotactic surgery. Neurosurgery 1994;35:696
704.
Holtzheimer PE 3rd, Roberts DW, Darcey TM: Magnetic resonance imaging versus computed
tomography for target localization in functional stereotactic neurosurgery. Neurosurgery 1999;45:
290298.
Alexander E 3rd, Kooy HM, van Herk M, Schwartz M, Barnes PD, Tarbell N, Mulkern RV, Holupka EJ, Loeffler JS: Magnetic resonance image-directed stereotactic neurosurgery: Use of image
fusion with computerized tomography to enhance spatial accuracy. J Neurosurg 1995;83:271276.
Schulder M, Fontana P, Lavenhar MA, Carmel PW: The relationship of imaging techniques to the
accuracy of frameless stereotaxy. Stereotact Funct Neurosurg 1999;72:136141.

Stereotactic Accuracy of a 3-Tesla Magnetic

Resonance Unit

Stereotact Funct Neurosurg

2003;80:140145

145

Downloaded by:
NYU Medical Center Library
128.122.253.212 - 5/27/2015 4:36:18 AM