164
Posters: Perfusion-diffitsion (braht )
2.0
1.5
E
"-"0,5 i,|.~*
2~rne4~kr~ge rt~bt~ OO
Coregistration of MR and EEG coordinate systems: accuracy of matching s p l i n e - i n t e r p o l a t e d with M R I - d e r i v e d head surfaces C. LammL Ch. Windischberger2, U. Leodotter t, H. Doleisch~, E. Moser a, H Bauer 1. 1Brain Research Lab, Department of Psychology, UniversiO' of
Vienna: "Institute for Medical Physics, NMR group, University of Vienna "*lnstitute of Computer Graphics, Technical University of Vienna
Purpose: Coregistration of EEG and MRI/fMRI is increasingly used in both laboratory and clinical studies. Registering of EEG and MRI-based coordinate systems is commonly accomplished by matching a small set of discrete fidt!cial points digitized in both measurement systems. Accuracy of this approach critically depends on accurate and repeatable determination of the fiducial points used for matching. As a more precise alternative, matching of MRIderived scalp surfaces to head surfaces acquired via 3D scanning of the head using an ultrasound localizing device has been suggested [1}. Since this procedure is rather time consuming, we hypothesized that coregistration accuracy is similarly improved when the head surface is not actually digitized, but only reconstructed by means of a splineinterpolation algorithm. Subjects and Methods: 3D coordinates of 42 electrodes evenly distributed over the head, and of nasion, inion, and left and right pre-auricular points were acquired in 21 male subjects by means of a 3D photogrammetric head digitizer (3D-PHD; http://www.iinform.oeaw.ac.at/~ igor/PHD). 3D-PHD uses digital cameras taking pictures of marker lights a attached to a subjects electrodes and/or anatomical landmarks to calculate spherical and cartesian coordinates of these markers. Due to the fully automatic measurement procedure (that only takes a few milliseconds), measurement error due to inaccuracies in manual determmat on of electrode coordinates (as experienced [2] when using a sensor pen device) is reduced to a minimum. Using these 3D-coordinates, individual headshape was reconstructed by means of a spherical spline interpolation algorithm (see Fig. 1). Anatomical MR images were acquired using FLASH (64 slices, t h k = 3ram) in a 3T Medspec (Bruker Medical, Ettlingen, Germany) wholebody magnet, with nasion and left and right pre-auricular points marked
Fig. 1. Sample of a head surface which was reconstructed by means of spline-interpolation using 46 3D electrode coordinates. White objects (electrode sockets) mark actually measured coordinates.
by vitamin E pills. MR and EEG data were coregistered by means of an iterative point matching algorithm [3]. Accuracy of the surface matching technique was estimated by calculating the residual measured coordinates.error of the matched surfaces. Effects of head size. head shape and signal quality of anatomical images were evaluated in detail. Results: Preliminary results suggest good accuracy of the surface matching technique, the average root mean square being comparable to the one reported in Ref. I (3.4 ram). Conclusion: Findings suggest that surface matching based on spline-interpolated head surfaces is a precise method for matching EEG and MR coordinate systems that can be easily used in a standard laboratory and clinical setting. Acknowledgments: This study was supported by the OeNB-Jubiliiumsfonds, grant no. 7174. Development of the 3D-PHD was supported by the Austrian Scientific Research Fund (FWF), grant no. 12289.
References [1] Huppertz H J, Otte M et al. Electroencephalogr Clin Neurophysiol 1998; 106:409 - 15. [2] Towle V L Balanos J e t al. Electroencephalogr Clin Neurophysiol 1993:86: 1-6. [3] Zhang Z. Int J Comp Vis 1994;13:119-152.
Posters: Perfusion-diffusion (brain) Spatio-temporal evolution of cerebrovascular reactivity after focal brain ischemia N.G. Harris 1"2, M.F. Lythgoe 1, D.L.Thomas l, S.R. Williams x. tUnit of Biophysics, Institute of Child Health, University College London Medical School London, UK: 2Academic Neurosurgery, University of Cambridge, Clinical School, Addenbrooks Hospital, Cambridge, UK Introduction: Vasomotor reactivity monitoring in conjunction with either transcranial ultrasound Doppler or PET imaging is frequently used to determine cerebral vasodilatory reserve in stroke or head-injured patients. In normal brain the increased blood flow response reflects the coupling between metabolic activity and vascular tone while in presumed iscbemic regions a positive response will also reflect satisfactory collateral flow. Reactivity to carbon dioxide is heterogeneous in the ischemic-lesioned rat brain but regions where it is partially impaired may still be responsive to stimulation. By characterising positive reactive ischemic areas, both spatially and temporally against acute lesion pathology it might be possible to identify potentially salvageable brain regions. We have used magnetic resonance imaging techniques in conjunction with the focal is,chemic rat model to (i) characterise the spatio-temporal evolution of the CO 2 response and (ii) to determine how these responses correspond to areas of perfusion deficit a n d cell swelling. Methods: Male Wistar rats (250-300 g) were anestbetised with halothane/N20/ 02 and mechanically ventilation. Following control diffusion imaging middle cerebral artery occlusion was performed remotely using the intraluminal suture method (0.28 or 0.24 mm thread for conventional occlusion or partial occlusion, respectively, n = 4/group) during a period of consecutive T~-weighted.
165
Posters: Perfusion-d!ff'usion (brahl) gradientecho image acquisitions to map the region of perfusion deficit. Cerebrovascular reactivity testing was performed at 0.5, 1.5, 2.5 and 3.5 h post-occlusion by supplementing the anesthetic gases with 15% CO,_ for 3 min during the centre of a 20 min acquisition period of 75 successive gradient-echo experiments (15 s temporal resolution). Post-occlusion diffusion images were acquired interleaved with the reactivity experiments. Regional perfusion deficits following occlusion were determined from T* subtraction images. Trace maps were constructed to resolve areas of reduced apparent diffusion coefficient (ADC) indicative of cell swelling and signal intensity changes were determined in the corresponding regions in the CO2-reactive data sets. Results: Occlusion resulted in an 18 +4.0% decrease in ADC within 5 min post-occlusion and this dropped to 34 + 4.7% by 4 h. In conventionally-occluded rats the region of reduced ADC was significantly larger when compared to the partially-occluded group (38.3 + 5.1 and 5.4_+0.8 mm z, respectively, P <0.01) despite similar areas of perfusion deficit (35.2_+ 1.8 and 35.5 +4.1 mm2). Analysis of the CO 2 reactivity images revealed three distinct patterns of signal intensity (Fig. 1), Area 1: CO.~ induced a ~ 5% increase in signal intensity in the contratateral hemisphere at all four timepoints post-occlusion. Area 2: in regions of reduced ADC with perfusion deficit the response was either very reduced, absent or there was an undershoot indicating the steal phenomenon. Area 3: in the region of normal ADC but reduced perfusion (partially occluded group) the response was generally delayed at the initial time-points and either delayed and reduced or absent at the later timepoints post ischemia (Area 3). Conclusion: In conclusion, CO2-reactivity becomes attenuated or delayed in different regions of hypoperfused brain, though the delay is too long to be explained solely on the basis of a delayed arrival of arterial CO 2. The ability to monitor differential CO_, reactivity following cerebrovascular occlusion, in combination with other MR measurements, should provide a powerful tool to improve our understanding of the development of stroke lesions at both a tissue and vascular level.
Results: Fignre 1A shows the offset. M~(0) - M~(O). in % of the equilibrium magnetization. M o, as a function of the thickness of the inversion pulse (FWHM). The imaging slice had a constant thickness of 10 ram. Taking into consideration that the FAIR signal in grey matter is about 1% of M o, it can be seen that the slice thickness of the inversion pulse must be above 45 mm for the offset to be negligible. Figure I B shows the calculated perfusion as a function of the slice thickness. Compared to the )erfusion level of grey matter ( ~ 0.01 ml/g per s), the measured perfusion is negligible at all slice thicknesses. This shows that the effect of the slice thickness can be separated from perfusion. We are currently testing the perfusion measurements in vivo with different slice thicknesses. 2O 15
"-'o ~Z
~, lO
0 -5 20 40 60 Slice thickness (FWHM) [mini
Area 2
Area 1
0.01
80
(B)
0.005
== E
.~
o
Area 3
-0.005
-0,01
Fig. 1. Gradient-Echo signal intensity change (y axis) versus image no. (x axis) at four times postocclusion and in three areas.
~-~
Zero perfusion calibration of FAIR imaging with arbitrary reversion slice profiles
K. Sidaros 1'2, I.K. Andersen2, H.B.W. Larsson:, H. Gesmar~, E. Rostrup2. W.M.M.J. Bovre s. IThe Technical University of Denmark; eHvidovre University
Hospital, Copenhagen, Denmark: 3Delft University, The Netherlands
Introduction:Perfusion measurement studies using Arterial Spin Labelling (ASL) have been shown to be sensitive to the slice profiles of the RF inversion pulses [1]. The interaction between the inversion and imaging RF pulses in FAIR imaging [2], results in an offset in the FAIR image at zero perfusion unless the slice-selective inversion pulse profile is several times wider than that of the 90~ imaging pulse. However, this reduces the sensitivity to perfusion and introduces larger transit delays for the incoming blood [3]. Therefore, it is desirable to overcome the errors due to the slice profile without increasing its thickness. Methods: The imperfect slice profile of the RF inversion pulse can be expressed as an imperfect degree of inversion. This is expressed by having different values for the longitudinal magnetization immediately after inversion, i.e. the slice-selective (ss) and the non-selective (ns) magnetizations just after inversion, M~(0) and MnS(0), are different. We have performed FAIR measurements (32 repetitions) with interleaved ss and ns images and T~ measurements (17 values of TI) in a homogeneous phantom with different ss inversion pulse thicknesses. The inversion pulse was a hyperbolic secant RF pulse. The parameters M o, M~(0), MnS(0), T]s and T'~s were estimated using least squares fitting. Perfusion was calculated using the FAIR signals and the estimated parameters by the expression
M~ = Mn~ = [[Mo - M~s(O)]f" Tl + MSs(O) - Mns(O)] e - Tt/r~
:
i
20 40 60 Slice thickness (FWHM) [mm]
80
Fig. 1. The offset, MS~(0) - Mn~(0), in % of the equilibrium magnetization. 34o, as a function of the thickness of the inversion pulse. (B) The calculated perfusion as a function of the slice thickness. It is clear that the calculated values are welt below the level of grey matter perfusion.
Conclusion: We have found that the effect of inversion slice thickness in FAIR imaging can be separated from perfusion. This allows us to decrease the thickness of the inversion slab thereby increasing the perfusion sensitivity of the FAIR measurements without introducing offset effects. References [1] Frank LR et al. Magn Reson Med 1997;38:558-564. [2] Kwong KK et al. Magn Reson Med 1995:34:878-887. [3] Buxton RB et al. Magn Reson Med 1998;40:383-396.
•
Comparison of FAIR and dynamic bolus tracking perfusion measurement in patients with acute cerebrovascular disease
S. Hunsche, D. Sauner ~, W.G. Schreiber2, S. Boor, G. Vucurevic, P. Stoeter.
Department of Neuroradiology and 2Radlblogy, Johannes Gutenberg University Main'., Germany: 2MRC HMBU, Institute of Neurology, Queen Square, London. UK
Introduction:Arterial spin-labelling-techniques are an alternative to T*-weighted gadolinium-enhanced imaging of cerebral perfusion. The flow-sensitive alternating inversion recovery-technique (FAIR) is one variant of these labellingtechniques [1,2]. The purpose of this study is to compare perfusion-images obtained with FAIR-technique with rCBF images obtained with dynamic susceptibility weighted MR imaging and deconvolution analysis [3,4] in patients with acute cerebrovascular disease.
166
Posters: Perfusion-diffusion (brain)
Subjects and Methods: Ten patients in the acute state of stroke (imaging within the first 12 h after onset of symptoms) underwent MR-imaging with the following protocol: diffusion-weighted-sequence: single shot gradient-echo EPI sequence, TR/TE 4000/104 ms, 128 x 128 matrix. FOV = 240 ram, slice thickness=5 mm and b = 1000 s/mm 2 in all three directions. FAIR-sequence: turb-FLASHsequence with slice-selective and non-sticesetective inversion alternating every 3 s, TR/TE 11/4 ms, T I = 1200 ms, :~= 15~ 64x 64 matrix, FOV=240 mm, slice tickness = 5 mm. A T*-weighted-sequence for bolus tracking: single shot gradient-echo EPI sequence, which makes different TE for different slices [5], T R = 1.3 s, TEl = 18 ms, TE2 = 60 ms, c c = 9 0 ~ 64x 128 matrix, FOV = 300 mm, slice thickness = 5 mm. Measurements were performed on a Siemens Vision whole-body MR scanner at 1.5 T, Subtraction of slice-selective and non-sliceselective turbo-FLASH-images results in a flowweighted FAIR-image. Motion correction was applied to the FAIR images using AFNI 2.0 software [6]. rCBF-images were obtained with dynamic susceptibility weighted MR imaging and deconvotution analysis [4]. Evaluation is done with matching DWI-, FAIRand rCBF-images and defining of regions of interest (ROI) on DWl-image (area of infarction and surrounding and homologue area on contralateral side) and transfer in congruent FAIR and rCBF-image. Signalratios (ipsilateral/contralateral) were calculated in FAIRand rCBF Images and were compared Results: FAIR- and rCBF-images showed high correlations in all patients. Perfusion abnormalities were observed in 8/10 patients. 2/10 patients with small-lacunar infarction demonstrated no perfusion abnormalities. Fig. l shows a patient who was scanned 4 h and 5 days after onset of symptoms. The arrow points to an area which shows reduced perfusion and small changes of diffusion after 4 h and indicates a tissue at risk. On day 5 this area was infarcted.
formes. 4 meningiomas. 3 metastatic tumors, 1 anaplastic ependymoma, 1 anaplastic astrocytoma, 1 hemangioblastoma, and I pilocytic astrocytoma. First we acquired conventional MR images and additional perfusion images consecutively. Then we repeated perfusion M R I 2-10 days after the treatment of steroid injection intravenously(10-20 nag/day). The GE 1.5 Tesla, Horizon, Echospeed scanner equipped with EPI was used for this study. The acquired data were transferred to GE Advantage Workstation for making color perfusion map images. The same data transferred to an independent Sun workstation for further processing. To produce relative regional blood volume map and the average pixel intensities of the map an image processing software, IDL (Interactive Data Language). was used. Visual evaluation of rCBV maps were performed by comparing the relative perfusions in the brain tumor and peritumoral edema with those of contralateral white matter. Objective evaluations were performed by comparing the perfusion ratio of brain tumors and peritumoral edemas. Results: In the visual evaluation of rCBV maps, most brain tumors showed highperfusion (66%, 10/15) before the steroid treatment and showed decreasedperfusion (80%, 12/15) after the steroid treatment. In the objective evaluations all brain tumors showed decreased perfusion In visual evaluations of perfusion change in peritumoral edema, no changes were perceived, but significant decreases in perfusions were noted by objective evaluation in all cases (seven cases). Conclusion The rCBV maps acquired by perfusion MR can provide hemodynamic information in brain tumors, and peritumoral edemas. The rCBV maps could be helpful in the preoperative planning phase of brain tumors and in the monitoring of steroid effects for the conservative treatment.
References [1] Vecht CJ. J Neurol [998;245:127-31. [2] Leenders KL, Beaney RP. Brooks DJ, Lammertsma AA, Heather JD,~ Mckenzie CG. Neurology 1985;35:1610-1616. [3] Siegal T, Rubinstein R, Tzuk-Shinal T, Gomori JM. I Neurosurg 1997:86:22-27. [4] Shoshan Y. Siegal T. Neurosurgery 1996;39:1206-1214.
Assessment of the clinical benefit of diffusion weighted imaging at 1.0 T J.E. Foster ~, M.R. Waiters 2, M. Cockburn 3, K.R. Lees-', N.C. McMillan s.
tDepartment of Clinical Physics, 2Department of Medicine and Therapeutics; 3Department of Radiology. Western Infirmary, Glasgow, UK Fig. l. A patient with tissue at risk in the right MCA territory (arrow) going to infarction: (a, b, c): FAIR-, rCBF-, DWI-image 4 hours after onset of symptoms; (d): DWl-image on day 5. Discussion and Conclusions: The results demonstrate, that FAIR shows the same perfusion abnormalities as contrast enhanced perfusion imaging. High reliability, simple and short postprocessing and no need for contrast enhancement agents are strong arguments for using FAIR in assessment of acute cerebrovascular diseases.
References [1] Kwong KK, Magn Reson Med 1994;34;878. [2] Kim SG, Magn Reson Med 1995;34:293. [3] Rempp, Radiology 1995;193;637. [4] Schreiber WG, JCBFM 1998;18;1143. [5] Hunsche Set al. ISMRM 1999. [6] Cox RW, Comput Biomed Res 1996;29:162.
Comparison of perfusion in brain tumor and peritumoral edema before and after steroid t r e a t m e n t by regional cerebral blood volume maps Haejin Kang I, Jusung Sun I, Sun Yong Kim I, Jung Ho Suh l, Eun-Kee Jeong2.
1Department of Diagnostic Radiology, Ajou University Hospital, Suwon, Korea; "Department of Diagnostic Radiology, Yonsei University Hospital, Seoul, South Korea Purpose: Based on the suggested hemodynamic change of the antiedematous effects by steroid treatment in brain tumors, we observed hemodynamic change of brain tumors and peritumoral edemas after steroid treatment, then investigated the clinical usefulness of perfusion MRI. Subjects and Methods: Fifteen patients, 6 males and 9 females aged from 6 to 84, with diagnosed brain tumors by conventional MR images and followed surgicopathological confirmation except metastatic brain tumors were retrospectively reviewed. Fifteen brain tumors consisted of 4 glioblastoma multi-
Introduction: Previous work [1] using a PSIF sequence [2,3,4] showed the potential for this approach to produce diffusion weighted images with less motion sensitivity than traditional methods but considerably higher resolution than EPI based methods on a standard clinical scar ~ ner. This approach has undergone a full Clinical Assessment. Method: In addition to standard T 2 weighted axial imaging, each patient was scanned using the PSIF sequence. 29 patients who presented with acute ischaemic stroke were scanned: 9 within 8 h of the onset of symptoms and 20 were scanned between 10-48 h of symptom onset. A Siemens 1.0T Impact Expert Scanner was used with a PSIF based diffusion weighted sequence programmed using the Pargen Sequence Editor. Parameters were: TR 29ms, TE 12ms, Flip Angle 50 degrees, FOV 240 mm, Slice Thick 6mm with a matrix of 192 • 256. Three acquisitions were taken to ensure good signal to noise in the image and minimise any remaining artefacts thereby giving a time of around 12 s per slice. Results: 29 patients were scanned with the DW-PSIF sequence. 9 In 9 cases lesions were seen with DWI that were not seen on the T 2 weighted images. The time from symptom onset to scan in these patients varied from 2 h 35 mins to 7 h 10 rains. 9 In 5 cases an established infarct was depicted on the T2W sequence only. In 8 cases there were concordant abnormalities on both T2W and DW-MRI. 9 In 4 cases more than one lesion was present in each individual. The diffusion weighted images allowed the new lesion to be discriminated from older lesions. 9 In 2 of the 3 patients presenting with transient global amnesia the DWMRI sequence demonstrated abnormal signal not evident on the standard T2W images. 9 In 2 cases the examination was rendered non-diagnostic due to motion sensitivity with the DW MRI. Conclusions: Diffusion-weighted MRI has been applied in a clinical environment using a 1.0T scanner for the diagnosis, quantification and topographical localisation of early ischaemic stroke. In this study, the addition of DW-MRI to the imaging protocol assisted medical management in at least 69% of cases since both a positive and negative DW-MRI result is clinically valuable. We have therefore shown that this technique is applicable on a routine basis in centres with conventional MRI scanners. Although there is a time penalty of approximately 8 minutes with this sequence compared to EPlbased methods the resulting high resolution image allows more confident reporting by a general radiologist.
Posters: Motion, Artifacts, Quality Control - Poster Walking Tour
167
Conclusion: Our study shows, that multiple navigators are a powerful tool to study respiratory motion both spatially and temporally resolved. Our results indicate, that a pure kinematical model of respiratory motion neglecting dynamic effects may lead to considerable errors when used for adaptive motion correction. Acknowledgements: This work is supported by a grant from the Deutsche Forschungsgemeinschaft.
References [1] Foster JE, ESMRMB 1998:293.
[2] Bruning R, Wu R, Deimling M, Porn U. Haberl R, Matmilian R. Invest Radio] 1996;31(I1): 709-715. [3] Buxton RB. Magn Reson Med I993;29:235-243. [4] Zur YuvaL Magn Reson Med 1997:37:716-722.
References [1] Ehman RL, Felmlee JP. Radiology 1989:173:255. [2] Wang et at. MRM 1995;33:713-719. [3] Taylor et al. 6th ISMR, 1998:322
Posters: Motion, Artifacts, Quality Control - Poster Walking Tour Study of spatial correlations of respiratory motion using multiple navigator pulses
~--~
Prediction and interpolation of motion states for r e a l - t i m e respiratory g a t i n g
K. Nehrke ~, P. Brrnert 2, D. Manke 2, D. Holz 2. J. Smink 3, J. C. B6ck t.
1Robert-Bosch-Hospital, Department of Radiology, Stuttgart, Germany; 2philips Research Laboratories, Hamburg, Gerrnany, "~Philips Medical Systems, Best, Netherlands
Kay Nehrke 1, Peter B6rnert 2, Johannes C. Bock I. tRobert-Boseh-HospitaL Department of Radiology, Stuttgart, Germany; 2Philips Research Laboratories, Hamburg, German)'
Introduction: Adaptive motion correction based on real-time navigator echo respiratory monitoring is a promising approach to reduce motion artefacts in MRI [I]. The information how to map the diaphragm motion which is monitored by a navigator echo, onto the respiratory motion of the region of interest, is usuall, obtained from breath-hold MR images acquired in different respiratory phases [2], However, thi method becomes inconvenient, if patient dependent mapping has to be performed [3]. In addition, it ha to be proven if the motion states existing during continuous respiration can be frozen by breath-holding For the present study multiple navigator echoes were used to study spatial correlations of respirator motion during continuous respiration. Subjects and Methods: In vivo experiments with several healthy volunteers were performed on a 1.5 T whole body seanne (GYROSCAN ACS NT, Philips Medical Systems) with self-shielded gradients (21 mT in 0.2 ms). The scan software was extended to provide three independent pencil beam navigator pulses, which can be positioned and angulated free in space. A pure navigator sequence allowing high temporal resolution (TR = I00 ms, i.e. 33 ms for each of the succeeding pencil beams) was implemented, The shift of the navigator profiles with respect to the corresponding reference profiles is determined in real-time and stored for subsequent analysis. The respiratory motion of heart and diaphragm was recorded over 10 min using pencil beams with a diameter of 25 mm. The first navigator was placed through the dome of the right hemidiaphragm, the second through the left hemidiaphragm behind the heart and the third through the left-superior margin of the heart (Fig. la). The delay of 33 ms between the three respiratory curves was corrected by appropriate interpolation of neighbouring data points. The contribution of cardiac motion was filtered out from the oversampled data. The correlation between the different n a v i g a t o r s was analysed in 2D histogram plots (Fig. 1b,c).
Introduction: Navigator echoes represent a powerful approach to monitor respiratory motion in a real-time gated acquisition [I]. However, in most MRI protocols only a few navigator echoes can be used per respiratory cycle to avoid destructive interference with the MRI sequence. Hence, it may be difficult to extract the actual distribution of motion states from the poorly sampled navigator data to find the most frequent motion state or to detect slow drifts. In addition, the gating algorithm may take the wrong decision, if the navigator pulse is too distant from the MRI sequence due to magnetization preparation pulses applied immediately before MRI. To overcome this problem the capability of different approaches to interpolate and predict respiratory motion in real-time is studied in the present work. Subjects and Methods: Navigator experiments were performed on a 1.5 T whole body scanner (GYROSCAN ACS NT, Philips Medical Systems) to monitor the diaphragm motion of several volunteers with high temporal resolution (AT = . 0.1 s). The respiratory data of the different volunteers were Fourier-analyzed to estimate the required sampling rates according to the sampling theorem. To mimic coarse sampling as, e.g. the case in a ECG triggered acquisition the number of sampling points was reduced. These reduced data sets were interpolated (sine interpolation and cubic spline) and compared to the original data with respect to the histograms of motion states. To estimate future motion states (up to 1 s ahead) linear prediction theory [2] was used. The predicted values were compared to the real motion states. The algorithms used for interpolation and prediction were implemented on the scanner to check their capability for real-time applications. Results: The Fourier analysis of the respiratory data showed, that a sampling rate -,~2 s - ~ is sufficient to satisfy the sampling theorem. The actual distribution of motion states can be recovered even from a few respiratory cycles of such a coarsely sampled curve, if appropriate interpolation is performed. Cubic
10E E
10-
b)
e)
O,
O.
r
-10
-1'o
6
1'o
-10
6
1'o
right hemidiaphragm(mm) Fig. 1. (a) Coronal scout indicating the pencil beam positions. (b,c) 2D-histogram plots of the correlation heart-diaphragm for two volunteers. While b) shows a linear relation, a hysteresis is deafly visible for c).
Results: In accordance with the literature a roughly linear relation between the position of heart and diaphragm was observed for all volunteers. However, for some volunteers a delay up to 0.3 s of the heart motion with respect to the diaphragm was observed (see Fig. lc), similar to the propagation of a wave. This maybe due to the elastic properties of the body. The gap between inspiration end expiration is up to 3 mm which may be considerable for adapted motion correction schemes.
spline interpolation is more robust against non-equidistant sampling (cardiac protocols) and missing data points (bad ECG signal) than sine interpolation. Linear prediction yields very good results for short term extrapolation (see Fig. 1). The required computation time on the scanner to interpolate between 40 data points and to predict l0 future values is small ( < 1 ms). Therefore, this approach can be used together with elaborated real-time gating methods, that rely on a careful analysis of motion statistics [3].
Posters: Motion, Artifacts, Quality C o n t r o l - Poster Walking Tour
168
b>iA-~
E v
~
9
q
-5
~ -5
time (s)
o
300
I
0
time (s)
300
0Fig. 1. Deviation of the predicted diaphragm position (0.1 s ahead) from the real position during a period of 300 s. The average prediction error (A) is considerable. if the last measured data point is taken as predicted value (a). A can be decreased by a factor of 7, if linear prediction is used (b).
Conclusion: Our study shows, that interpolation and prediction of respiration curves improves the evaluation of motion statistics and the accuracy of gating decisions. Because of the low computational expense this approach can be used on a scanner to increase the performance of real-time gating. Acknowledgements: This work is supported by a grant from the Deutsche Forschungsgemeinschaft. References [I] Ehman RL, Felmlee JP. Radiology 1989;173:255. [2] Papoulis A. Probability, Random Variables and Stochastic Processes, McGraw-Hill, New York. [3] Sinkus R, Brrnert P. MRM 1998;1:148.
~'~
Effects of respiratory navigation on image quality in M R imaging of the living mouse
tor I and 2, which enclose the image acquisition show the same profile signifying no respiratory motion during the acquisition of a single image segment. Integration of the profile gave us a reliable parameter to distinguish echoes of different respiratory phases automatically (Fig. a). Figure b represents the uncorrected 5-fold average of the experiment. The navigator corrected image is shown in Figure c. The quality of the heart image is only slightly improved whereas the vessels of the liver show a clearly reduced blurring. (a) Eight respiration cycles during one image acquired within 32 segments; the acquired image segments above the dashed line were chosen for the navigator corrected image: (b) averaged image (c) navigator corrected image. Conclusion: Respiratory navigation of MRI in the living mouse is feasible and easy to realize. Navigator corrected images of the heart showed only minor improvement of edge definition and image contrast, whereas there was a clea~, reduction of motion artifacts for the surrounding vessels. Therefore, respiratory navigation bears the potential of significant in thoracoabdominal MRI adapted to the mouse model.
References J. Ruff I, F. Wiesmann z, A. Haase t. XPhysikalisches lnstitut, EP5, Universitiit
Wiirzburg; -~Med. Universitiitsklinik. 97074 Wiirzburg; Germany
Introduction: Magnetic resonance imaging (MRI) of thoracoabdominal structures is often linked with image artifacts caused by respiratory motion. In recent years prospective and retrospective navigated sequences could improve quality of human cardiac MR [12]. The purpose of our study was to examine the feasibility of respiratory navigation for MRI in the mouse. We compared navigator corrected images with non-corrected images of the mouse heart and liver vessels. Methods: Anesthetized mice (body mass 20 g) were explored on a 7 T NMR spectrometer. An ECG-triggered segmented FLASH sequence with two additional navigator echoes was implemented. One navigator echo was set immediately prior to the image segment of 8 k-lines, the other was set directly after it. The navigator was designed as a profile of ~ 1 mm thick sagittal oriented slice acquired with gradient echo. A coronal scout image was used slice. The gradient echo provides two advantages to spin echo navigators: (1) A much shorter repetition time can be realized which is useful for ECG-triggered in vivo mouse imaging due to its intrinsically high heart rate: (2) A small flip angle avoids the saturated stripe crossing the image plane. A midventricular short axis slice was imaged with the navigator sequence. To provide enough equal k-lines for retrospective navigator correction the experiment was repeated five times. Navigator profiles were evaluated with the software IDL in order to select image segments acquired during expiration. Results: We realized an easy implementable and reliable navigated and ECGtriggered sequence for image acquisition of organs of the living mouse. The navigator projection profiles of in and expiration were quite different. Naviga-
i~I "~ .,i-
[1] Chuang ML, Chen MH, Khasgiwala VC et al. J Magn Reson Imaging 1997;7(5):811-814. [2] Wang Y, Grist TM, Korosec F R e t al. Magn Reson Med 1995;33(43:541548. Real-time off-resonance correction using conjugate phase
reconstruction H. Eggers 1, F. Herold 2, P. Boesiger 3, R.-R. Grigat 2. IPhilips Research, Division Technical Systems Hamburg. Germany: 2Research Department Image Processing Systems, TU Hamburg-Harburg, Germany, 3Institute of Biomedical Engineering, ETH and Unit'ersity of Zurich, Switzerland
Introduction: The quality of magnetic resonance images degrades in the presence of spatial inhomogeneities of the static main field and susceptibility variations. While they cause only geometrical distortions for conventional acquisition schemes, imaging with advanced k-space trajectories, as for instance spirals. suffers from a high sensitivity to spatial off-resonance that often results in a severe blurting, Different approaches to correct for these artifacts have been reported. Notably the conjugate phase reconstruction (CPR) has been studied extensively and proven to yield satisfactory, image quality. Although fast approximations exist, its application in real-time imaging has been restricted to constant and linear terms of the field map [1], mainly due to the large amount of required computations.
~
i navigator 1 ...... navigator 2 ,-,,/
, ~ ~..,,,,
.,..~.~ .~;,. ~,'): ~,; ;/.,
Fig. 1. a) 8 respiration cycles during one image acquired within 32 segments; the acquired image segments above the dashed line were chosen for the navigator corrected images: b) averaged image: c) navigator corrected image
Posters: Motion, Artifacts, Quality Control - Poster Walking Tour We propose a dedicated reconstruction hardware [2], initially designed for fast convolution interpolation reconstruction, as a suitable hardware platform to speed up the CPR such that an integration of an off-resonance correction. which also considers higher order terms, into a real-time reconstruction becomes feasible. Methods: The influence of the local off-resonance frequency Awo(rl. assumed to be known from a separate field map acquisition, on the receive signal s(t) is described by f v m (r) e - i*.~r e -/~o,o~r~, dr.
S(/)
The CPR approximates the inverse of this transformation by simply inverting the signs of the exponents. Various suggestions on how to accelerate its calculation have been made: time segmented reconstruction [3], frequency segmented reconstruction [4] and polynomial approximation of the additional exponential term [5]. Due to their similarity, we restrict our investigation to the polynomial approximation. It requires essentially the fast evaluation of the sum:
m(r) _~ ~ ciAeflo (r) ~ -
I (ti s(t)),
i=0
where c~ denotes complex interpolation coefficients and N depends on the difference of the maximum and minimum off-resonance frequency. This calculation has been mapped to the dedicated reconstruction hardware described in [2], making use of pipelined and parallel processing to speed up the computation. Despite some differences, the required operations and the resulting data flow exhibit a structure very similar to that of convolution interpolation reconstruction. Therefore, an identical hardware architecture is suitable to efficiently execute both. Results: The performance of the CPR implementation has been analyzed integrated into a real-time reconstruction on a 1.5 T Gyroscan ACS-NT system (Philips Medical Systems, Best, The Netherlands). For a typical value for N of 12, a frame rate of 10 images/s is attaine& Conclusion: It has been demonstrated that the CPR can be performed in real-time. A further increase in the achieved frame rate is expected from an extraction of linear terms of the field map to improve the shimming and thus to decrease N. Besides. a proper integration of the field map acquisition into the running measurement is necessary to extend the applicability to interactive imaging.
~b(j)=aj2 + bj + c
(2)
where a, b and c are constants that contain information about the read gradient strength Gn. the 2 pulse separation, t. and the velocity along the read direction. Method and Results: We acquired a set of SE Burst images on a SMIS 2.0T small bore horizontal system. With a device built by SMIS (Guildford, England) we could translate a gelatine phantom, along the axis of the magnet at different constant velocities. Figure I shows the images acquired with GRII: and v = 8, 12 and 16 mm s - I. Using the phase information extracted from the same experiment, but run with the phase encoding gradient off. it is possible to correct the original data and reconstruct images that are free from artefacts. The corrected images are displayed below the corresponding original images. When GRLr no artefacts are shown in the images: see the coronal (GR• and the transverse views in Figure 1. Discussion and Conclusions: Burst-type sequences can be very sensitive to the translation of the spins along the large read gradient applied during the experiment. Phantom studies show that the phase imposed on k-space data causes image artefacts that can be corrected if the phase map is known or can be calculated from [2]. A combination of this work with that in [2] should allow a correction to be made for general rigid body motion, although deformational motion is always likely to remain a problem for Burst.
v =8
mm s~
v=12 mms I
v=16 mms
"1
Original Coronal Views GRIIv Corrected Coronal Views v=16 mm f
v=16 mm st
Original Coronal and Transverse Views GR-I-v
References [I] [2] [3] [4] [5]
169
Kerr AB, Pauly JM et al. Magn Reson Med 1997:38:355-367. Eggers H, Proksa R. Proc ISMRM, 1999. Noll DC. Meyer CH et al. EKE Trans Med Imag 1991;10:629-637. Man LC. Pauly JM et al. Magn Reson Med 1999;37:785-792 Schomberg H. submitted to EKE Trans Med lmag 1999, References
~-~
T r a n s l a t i o n a r t e f a c t s in b u r s t i m a g e s
[1] Hennig J e t al, MAGMA 1993:1:39-48. [2] Wheeler-Kingshott CA et al. Proc. 7th Meet ISMRM, 1986
C.A. Wheeter-Kingshot0 '2, Y. Cremillieux 3, S.J. Doran I. ISchool of Physical Sciences, University e f Surrey, GuildJord G ~ SXH, UK: 2Now at NMR Unit. Department of Clinical Nenrology: Institute of Neurology, UCL, London WCIN 3BG UK: ~Laboratoire de RNM. UniversitO Claude Bernard Lyon I, France
Introduction: Burst [1] is an ultra-fast technique that uses long and very, large read gradients during excitation and acquisition. Hence, it is very sensitive to motion along the read direction. The effect of rotating the sample during Burst acquisition has been previously discussed [2]. Here we present the problem of the sample moving during the experiment at a constant known speed. The images are affected by a complicated artefact that results from a phase modulation of the k-space data, typical of Burst experiments. Knowledge of the phase map associated with the sample translation can be used to correct the images. Theory: It is well known that moving spins gain a phase:
~= - foG'r(t)dt = -
G.rodt-
G'v(t)tdt
(1)
assuming that acceleration and higher derivatives may be neglected. Two features of Burst lead to distinctive phase patterns in the echoes: (i) the n echoes that cover k-space to generate the image are created in one shot by applying a pulse train of n independent cr-pulses, together with a frequency encoding read gradient, The magnetisation that generates echoj evolves under a read gradient of the same amplitude, but different length from that generating any other echo; (ii) the read gradient is very large compared to the other imaging gradients. As a result of this gradient scheme, the phase of the echo train assumes different values for different echoes and is a polynomial of the second order in the pulse number j:
A new method for estimating phase values in projection reconstruction M R imaging G. Lateur 1"2, R. Van de Walle 1"2 I. Lemahieuk ~Department of Electronics and
Information Systems, University of Ghent, Ghent, Belgium: 2Magnetic Resonance Department, Ghent University Hospital Ghent. Belgium Introduction: In projection reconstruction (PR) magnetic resonance imaging (MRI), an unknown phase factor is superimposed on the acquired signals. In this paper, we present a new method for estimating this phase factor. Our results are compared to those obtained with standard methods, using in vivo MR data. Subjects and methods: We will restrict this discussion to 2D ~maging. By traversing the spatial-frequency space (k-space) along a radial line through the origin at an angle 0 relative to the k-axis, we can sample the Fourier transform of the orthogonal projection of the image function onto an axis at the same angle 0 relative to the x-axis. Since the image function is real, so are these projections. We, therefore, expect our signals to be Hermitian, i.e, S o ( - k ) = So(k)*, where k is the coordinate along the axis in k-space and * denotes complex conjugation. However. because MR signals are measured in quadrature relative to an a priori unknown reference-axis, an unknown phase factor is superimposed on the signals. This is the main reason real PR data are not Hermitian. The most obvious way to estimate this factor is to look at the phase of the DC-component of the signal, i.e. the sample at k = 0. A better method consists of calculating the phase of the mean signal. This has the advantage that the noise in the individual samples is partially averaged out. We note that using only a small number of samples around the DC-component yields the best results, as the high-frequency components have relatively small signal-to-noise ratios.
Posters: Motion, Artifacts, Quality Control - Poster Walking Tour
170
As the set s of all Hermitian signals is convex, we can, given any non-Hermitian signal S(k), determine a unique Hermitian signal Sr~(k)~,S~r at minimal s of S(k). Sn(k) is called the projection of S(k) onto Z n. For the measured signal S(k), we can then consider all signals S,(k) = ejos(k), and calculate that particular phase ~b,, minimizing the distance between Sob(k) and its projection onto Xn We then obtain e - ~ ' ~ as an estimate of the unknown phase factor. We have
1
I 2~1 ( y T. x +. + y .+ x Z. ) + 2 X Y i = .
]
,
~bm= ~ a t a n ~21-~I ( y T x : + y i + x , - ) + y 2 - X 2 J
where X + j Y is the DC-component and x, +- +jy~ are the samples at spatial frequencies +i,Jk. As with the previous method, we should not use too many samples in this calculation. Results: We estimated the phase factor in real. fully acquired signals using the various methods described above. Based on the hermiticity of the signals, we then used a part of the data to estimate the remaining part and reconstructed the images. We found our method to lead to the best approximation of the images obtained using the actual measurements. Conclusion: A new method for the estimation of the unknown phase factor in PR MR data was proposed. The method was shown to perform better than existing ones.
T h e use of an off-resonance reference signal to correct for
variation in coil loading D.J. Herlihy, I.R. Young. Robert Steiner MR Unit, Imperial College School of Medicine, Hammersmith Hospital, London, UK Purpose: To use variation in coil Q at an off resonance reference frequency to detect and correct for motion at the imaging frequency. Introduction: Motion, such as respiration, results in changes in the loading and therefore Q of a coil with consequent fluctuation in its output. Coil motion relative to the volume to be imaged is particularly hard to prevent in, for example, small flexible internal coils such as those used with g~troscopes or colonoscopes. Q varies as the inverse of the effective resistance of the coil plus the load. The load is the dominant factor at all except the lowest fields. The impact of changes in resistance as the coil moves is broadband, and can be assumed to vary quadratically with frequency (though this is only an approximation if the frequency range is very large). A low powered RF source, the output of which can be precisely controlled, and which is located as far from the body as is practicable, may be used to inject an RF signal. This signal may be detected by an appropriately tuned coil that is coincidental in position with an imaging coil. Variation in Q will result in variation of the signal detected by the reference coil, and the inverse ratio of this signal to a reference derived at the start of the study may be used as a correction for the imaging signal. Methods: In principle any out of band frequency could be used as the reference, with appropriate allowance for the relative sensitivities of that frequency to the imaging one. A coil was built and dual tuned to 42.3 MHz (the imaging frequency) and to the reference frequency which was chosen to be about 10 MHz above in order to minimize the effect on the imaging mode. A second coil was triple tuned to 42.3 MHz, to a much higher frequency (about 100 MHz), and to a much lower one (about 10 MHz). The third mode (10 MHz) was intended to monitor possible external noise spikes. A 52 MHz sinusoidal signal was injected via a probe located some distance from the coil as a phantom was moved from 1 to 2 cm away from the coil. Data were obtained from the coils, though the machine used (a prototype 1.0 T neonatal system) did not allow for automation of the correction process, and this would have to be implemented off-line. In practice, machines could be fairly readily modified to include the correction process described here. Results: The experiments demonstrate how data may be collected from which correction could be achieved for variation of the load on the imaging coil, and how wide-band noise spikes can be monitored. Discussion: Keeping the reference frequency close to that of the resonance is philosophically attractive, making the behaviour of the reference more like that of the imaging detector, but the reference mode may have a detrimental interaction with the imaging mode. Using a high frequency reference means that the reference system sensitivity is high. This is useful at low field where, although the extrapolation is greater, over the fairly limited range of variation that is actually encountered, the larger variations in reference signal allow for more precise adjustments. Once this concept is accepted, the idea of using a third otherwise passive channel to help detect wide-band external noise spikes is obvious. Conclusions: The use of off resonance reference data to measure coil Q and to correct for variation has been investigated. The full evaluation of this approach is still being undertaken.
Elimination of fold-over artefacts by using multiple rf coils
Robert Steiner MR Unit Clinical Science Centre, Imperial College School of.Medicine, Hammersmith Hospital, London. W I2 0HS, UK A.H. Herlihy, D.J. Larkman, G.A. Coutts, J.V. Hajnal.
Introduction: Fold-over artefacts arise when imaging frequencies occur at locations other than the desired location, but within the spatial region of radio-frequency (rf) coil sensitivity. These artefacts can occur on conventional scanners, but they frequently manifest more seriously on systems with short magnets or short gradient tubes. This study investigates the use of an extra rf coil and a post-processing method to identify and remove fold-over artefacts. Consider a scanner equipped with a single birdcage body coil to which is added a second receive only coil sensitive to the region from which the fold-over artefacts originate. Data is acquired from each coil to separately measure the spatial distributions of image domain signals originating from the desired slice location and the artefact bearing region. The resulting sensitivity data forms a spatially dependent 2 • 2 complex matrix IS] for each image pixet. The object to be imaged is then placed in the system and image data acquired with the same coil geometry and sequence to produce an image from each coil for each slice scanned. Corresponding pixels are represented in these two images as a complex column vector [C]. Pre-multiplying [C] by the inverse of IS] will then separate out the artefact and desired image signals, provided IS] is non-singular [1]. To avoid altering un-artefacted data only those regions of the image where the fold over artefact impinged upon the image were corrected. These regions were determined by the value of the determinant of matrix IS]. The magnitude of the corrected data is subject to an intensity normalisation with respect to the sensitivity data. The corrected region, therefore, requires re-normalisation to avoid discontinuity in the final image. Method: The above method was tested using a 1.0T Picker (Highland Height~ Cleveland, OH) prototype neonatal scanner, which has a 38 cm long bore and produces fold-over artefacts with conventional spin echo sequences. The birdcage body coil was centred at isocenter and a 7 cm loop surface coil was positioned 10 cm away from isocenter where it is known that signal is folded into the main image. Sensitivity maps of the rf coils were obtained by imaging a uniform 1 6 cm diameter copper sulphate phantom. The phantom was first located at isocenter, and was then moved to the location of the source of the artefact and imaged with both rf coils. Image data to be corrected was collected by placing a structured phantom in the bore such that the phantom extended over the region containing isocenter and the foldover location. Images were obtained with both rf coils. Anatomical data was collected by substituting an arm for the long structured phantom. Spin echo imaging parameters were: TE 20 ms, TR 200 ms, FOV 20 cm, slice thickness 5 mm, 25 s acquisition. The fold-over artefact was removed from the data images as described above. Results: The fold over artefact was removed revealing underlying structure that was obscured by the artefact. Discussion: The noise level on the corrected images is strongly dependent on the magnitude of det[S] and SNR can be degraded if IS] itself has relatively low SNR. This technique was simple to implement, avoided pulse sequence limitations and could be applied to a wide variety of situations where foldover artefacts contaminate image data.
Reference [1] Larkman DJ et al. ISMRM Proc 1991:91. --~
T h r e e years follow-up of M R I equipment by automatic quality a s s e s s m e n t protocol
P. Bourel, E. Coste, D. Gibon, V. Daanen, I. Rousselle, J. Rousseau.
University
Hospital and Centre, Lambret, Lille, France Introduction: The purpose of quality assessment protocols is to permit an objective assessment of the technical performances of the apparatus (i) to set up standards for a priori comparison of the performances of different machines; (ii) to check a machine for conformity with the manufacturer's specifications; (iii) to ensure the long-term follow-up of technical characteristics and their possible drift with time, Quality parameters are subject to normal statistical fluctuations, which means that their tendencies can only be evaluated by continual monitoring. We have devised acquisition protocols that take up only a small amount of machine time, as well as algorithms for automatically processing the images acquired. Together, they ensure a reliable, regular supply of a large volume of technical information essential for machine follow-up. Material and methods: Quality assessment was carTied out on three machines made by SIEMENSTM,a resistive magnet 0.2T OPEN~, a superconductive 1.0T EXPERT ~ and a susperconductive 1.5T VISION~ . The test object (Deluxe Model MRI/DLX-I | Data Spectrum Corporation rrw) is formed by a cylinder of acrylic material (inside diameter of 21 era), filled with an aqueous copper sulfate solution. It has a full, uniform section for measuring the signal-to-noise ratio (SNR), image uniformity and global distortion. Five sets of inserts provide access to the other parameters: local distortion, slice thickness, slice profile spatial resolution. A specific protocol is devised to minimize
Posters: Perfusion-methodology test object positioning errors. The body coil is used. The image acquisition sequences are nearly identical for the three machines: Flash2D, 4 mm thick., number of signals av. = 2. TR = 120 ms, TE = 5 ms for Vision 1.5T and Expert 1.0T, 9 ms for Open 0.2T, flip = 90", FOV = 250 mm, matrix: 256 x 256, BW = 300 Hz per pixel for Vision 1.5T and Expert 1.0T, 74 Hz per pixel for Open 0.2T). Two acquisitions are effected in succession. The images are processed using dedicated software on an IBM-PC type microcomputer. Results: The measurements presented are those obtained for the three machines installed in the MR1 department of our General Hospital. The study of their drifts with time covers the period from February 1996 to december 1998. The mean values and the variability of the different parameters measured are given. The parameters observed for the VISION 1.5T all indicate excellent stability over the observation period. The parameters measured for the EXPERT 1.0T show satisfactory stability and values that are to be expected in relation to the reference machine, except for signal uniformity. The results obtained for the OPEN 0.2T illustrate the technical limitations of this machine. This is perfectly understandable, since this imager, equipped with a low field, open, resistive magnet has been specially designed for interventional MRI and in pediatric or emergency context. Discussion: The quality assessment methods presented have proved accurate, reliable, fast and capable of being implemented on a routine basis. Each quality control requires approximately 10 min manipulation. The main value of the protocol resides in the follow-up of the quality parameters and objective analysis of any drift that they may exhibit when the eye of the radiologist detects some deterioration in the ability of the image to facilitate diagnoses, if, of course, the eye has not gradually become accustomed to it. Acknowledgment: The authors wish to thank L, Nicol from Siemens Medical System for its constructive criticisms anc comments.
Tissue equivalent breast phantom for MRI G.P. Liney, D.J. Tozer, L.W. Turnbull. Centre for MRI, Hull Royal Infirmary Hull, UK
Introduction: The purpose of this study was to design an inexpensive tissue equivalent breast phantom, which could be used to test a variety of breast MR imaging sequences. Methods: Studies were performed using a 1.5 T GE Sigma system and a commercial breast coil, Various materials were examined for their similarity to breast tissue in terms of longitudinal relaxation time and resonant frequency. T~ measurements were performed using the variable flip angle technique, which utilises the signal intensity ratio of two different gradient echo sequences. Specifically, a 3D FSPGR sequence was acquired with 60 sections (thickness/ separation =2.5/0 mm). A proton density-weighted sequence (TE/TR/flip = 4.2/15 rns/6~ was acquired followed by a T j-weighted sequence (TR/TE/flip = 4.2/15 ms/35~ The signal intensity ratio of these two sequences was then determined for all materials studied. Absolute values of T~ were derived from this ratio by using a calibration curve established with commercial test objects of known relaxation time. The accuracy of the method was evaluated using two separate phantom oils with Tt, values of 140 and 925 ms. Materials were investigated at two ambient temperatures and in various concentrations. Additionally, ten women (22-57 years) were examined to determine the range of T 1 values found in breast tissue. The final phantom design consisted of a layer of lard, to simulate adipose tissue, surrounding a commercial jelly product to simulate normal parenchyma. 'Enhancing lesions' were incorporated into the design by suspending small capsules of Gd-DTPA doped water within the phantom. Results: T l measurements in the two oils were within 4% of actual values using this method. In vivo studies revealed mean( + SD) T 1 values for normal parenchyma of 773.8 • 183.0 ms and 226.7 + 41.6 ms for adipose tissue. T, values for the fat equivalent material were 165.7 ms at 6~ and 204.5 ms at 21~ Values for the jelly product varied with concentration and temperature (355.2-1061.2 ms at 6~ 484.4 to 1394.0 ms at 21~
171
Conclusions: The phantom effectively simulates MR images of the breast and has enabled multipurpose quality assurance tests to be performed including fat suppression.
Posters: Perfusion-methodology ~-~
M o d i f i c a t i o n s to the F A I R s e q u e n c e to o b t a i n q u a n t i t a t i v e perfusion i m a g e s
N. Karger ~, S. Lusse 1, B. Kfihn-'. J. Grimm ~, J. Biederer 1, M. Heller, C.-C. Gl(ier 1. tKlinikurn an der CA U :u Kiel. Klinik fur Diagnostische Radiologie, Michaelisstr. 9; D-24105 Kiel; 2Siemens-Medizintechnik, Henkestr. 12Z D-91052 Erlangen, Germany Introduction: MR Imaging using spin labeling of water protons has the potential to non-invasively map perfusion in the brain and other organs. The FAIR (flow-sensitive alternating inversion recovery) tagging-sequence is usually combined with an EPl-readout sequence, which is fast but sensitive to susceptibility gradients, and has a limited resolution and signal-to-noise-ratio [1]. A second problem inherent to the FAIR tagging scheme is its dependence on the value of the inversion time T l, which makes it sensitive to the different spin-lattice relaxation times T I, in tissue [2]. To overcome these problems, we combined FAIR with a UFLARE (ultrafast low angle RARE) readout sequence. We also employed a different quantification method for the perfusion images, which is based pin the different apparent Tl-values after global and slice-selective inversion. Subjects and Methods: The FAIR technique is based on the acquisition of two images: one obtained after global inversion of the proton spins, and one obtained after slice selective inversion. The conventional FAIR result is obtained by subtraction of these images. The 'T~ method' relies on the measurement of two relaxation times: T~~ after global inversion, and "17~ ~ after slice-selective inversion of the spins. These were obtained by performing a series of 10 FAIR measurements with varying T l values; the two spin lattice relaxation times were then obtained by fitting the respective signal intensities to a monoexponential decay function. Perfusion rates are calculated pixel-by-pixel according to: f = ;~(I/T] ~ - 1/T~)q" [3]. Here, f is the perfusion in mt/(g s), and ~ is the brain tissue/blood-water partition coefficient ( ,~ 0.9 ml/g), q is a correction factor and is a function of T~, and the relaxation time of arterial blood. Transverse slices through the brain of four healthy volunteers (27-36 years) were measured on a 1.5 T scanner (Siemens Magnetom Vision). The total acquisition time was close to 10 min. Gray and white matter ROIs for the calculation of perfusion were determined based on the relaxation times in the T~-maps.
to a Tl-weighted image (right) of one volunteer. The average perfusion values for 4 volunteers are ( f + SD): 72 _+ 14 rrd/(100 g/min) for grey matter, 43 + 6 ml/(100 g min) for white matter. These results are in the range of published values for brain perfusion [2,4], Conclusion: It has been shown that quantitative perfusion imaging using FAIR-UFLARE is possible by applying the Tt-technique. This method avoids the problems arising from the choice of T t in the conventional FAIR method. It is better suited to measure abdominal organs than FAIR-EPI because it is not limited by problems with susceptibility gradients and it does not require fat suppression.
References
Fig. I. Tl-weighted FSPGR images acquired in the coronal plane; (left) the left breast of a young woman acquired during a dynamic contrast-enhanced examination demonstrating an-enhancing lesion, and (right) a section through the breast phantom incorporating the Gd-DTPA doped water capsule.
[1] [2] [3] [4]
Crelier GR, Hoge RD. Munger P, Pike GB. MRM 1999:41:132-136. Ben" SS, Mai VM. JMRI 1999:9:146-150. Schwarzbauer C, Morrissey SP, Haase A. MRM 1996;35:540-546. Yang Y, Frank JA, Hou L, Ye FQ, McLaughlin AC. MRM I998:39:825832.