102

Methodology: perfi~sion and diffusion

was used for documentation of the results. All patients underwent neurological examination after 3 and 6 weeks and a lumbar spine MRI after 6 weeks. Results: The procedure was finished successfully in all patients. Localization of the needle position and placement within the spinal disc was possible. Despite moderate pain no other complications occurred. Additionally used shifted echo gradientecho provided the display of thermal changes but required postprocessing. Improvement of up to 80% reduction of symptoms have been reported by the patients (Fig. 1).

pulse is used followed by the acquisition of another 16 snapshot FLASH images. For both series of snapshot F L A S H images T~ parameter maps are calculated on a pixel by pixel base using a 3-parameter fit and correcting for progressive saturation effects [5]. Subsequently, following Eq. 1 from both T~ parameter maps a f/2-map can be calculated. The described spin labeling technique was applied to patients with acute stroke who at our hospital routinely undergo dynamic susceptibility contrast enhanced MR imaging (DSC) protocols as an initial diagnostic tool in order to quantify cerebral blood volume (CBV), cerebral blood flow (CBF) and mean transit time (MTT). The measurement protocol used for DSC imaging was already described earlier [6]. Results: By comparing CBV, CBF and MTT images obtained with DSC imaging and perfusion images obtained with spin labeling only poor correlation between the obtained f/2 maps and CBV and CBF maps was observed. However, a relationship between a (f/A) - l map and MTT maps was found, indicating that the used spin labeling technique does not enable the direct measurement of regional blood flow but instead allows the determination of MTT in the respective tissue of interest. Conclusions: For spin labeling techniques using the T I perfusion model it is usually stated that following Eq. 1 regional blood flow can be measured [1] based on the assumption that the blood-brain partition coefficient (A) is independent of the respective tissue. There are, however, indications from literature [7] that ,l may vary significantly for different types of tissue. Based on the definition of 2 as introduced by Kety [8] it seems justified to imagine ,1 as being closely correlated to the regional cerebral blood volume (rCBV). Since in DSC imaging MTT is usually calculated via the ratio CBV/CBF this line of argument would explain for the fact that ( f / 2 ) - i parameter maps as obtained with our spin labeling sequence are closely correlated to MTT maps as known from DSC imaging.

References

Fig. 1.

Conclusion: Laser decompression of lumbar spinal disk herniation under MR is feasible but particularly the slice positioning and temperature mapping requires further improvement. PC post processing of the temperature images are currently under development. Opto-electronic navigation for interactive slice orientation is a significant improvement that eases localization of the correct image plane.

[I] [2] [3] [4] [5] [6] [7] [8]

Detre JA, et al. MRM 1992;23:37. K K Kwong, et al. MRM 1995;34:878. Schwarzbauer C, et al. MRM 1996;35:540. Deichmann R, et al. J Magn Reson 1992;96:608. Nekolla S, et al. J Comp Assist T o m o g r 1992;16:327. Ellinger R, et al. Proe. 1SMR 1998; 1206. Yokoi T, et al. J Nucl Med 1993;34:498. Kety SS, Pharmacol Rev 1951;3:1.

New approach of non-invasive perfusion imaging using arterial spin labeling: inflow turbo-sampling E P I - F A I R (ITS-FAIR) Bock, L.R. Schad. Forschungsschwerpunkt Radiologie. Deutsches Krebsforschungszentrum (DKFZ), D-69120 Heidelberg,Germany

M. Giinther, M.

Methodology: perfusion and diffusion ~

rCBF, rCBV or MTT? What do we measure with spin labeling techniques based on the T 1 perfusion model?

C. Kremser, M. Schocke, R. Ellinger, S. Felber, Department of Magnetic

Resonance, Universtiy of Innsbruck, Austria Purpose/Introduetion: Based on the T I perfusion model given by Detre et al. [I] a spin labeling technique for noninvasive quantitative measurement of brain perfusion has recently been proposed by Kwong et al. [2] and Schwarzbaner et al. [3]. These techniques thereby rely on the acquisition of T, maps after selective and nonelective inversion pulses, respectively, allowing the calculation of regional blood flow following

1

1 +f

where T~, corresponds to the longitudinal relaxation for selective inversion, f corresponds to blood flow and 2 is the blood-brain partition coefficient. It was the purpose of this study to implement a spin labeling sequence following the method described by Schwarzbauer et al. [3] on a commercial whole body MR scanner and to evaluate the clinical usefulness of this method on stroke patients by comparing the achieved results with a CM based perfusion method. Subjects and Methods: The spin labeling sequence was based on a fast inversion recovery snapshot FLASH T~ mapping sequence as first described by Deichmann et al. [4] and was implemented on a 1.5 T whole body scanner (Magnetom VISION, Siemens, Germany). The imaging sequence thereby basically consists of two parts. First the magnetization is inverted by applying a slice-selective inversion pulse. The longitudinal relaxation of the magnetization is then detected by a series of 16 snapshot FLASH images (TR = 3.9 ms, T E = 2.1 ms, ~ = 4 " , scan matrix: 64 • 128). After a delay of 13 s to ensure a complete relaxation of the longitudinal magnetization a nonselective inversion

Introduction: Recently, arterial spin labeling techniques have shown their potential for non-invasive quantitative perfusiota imaging [I-3]. For quantification it is necessary to acquire a series of images at different inflow delay times after inversion (TI). A new approach is presented for an efficient technique of sampling the blood inflow with a temporal resolution of 100 ms. It uses a technique similar to the Look-Locker technique and so the measurement time is under 5 min. Material and Methods: The principle of o u r technique is similar to the LookLocker EPI T : m a p p i n g sequence presented by P. Gowland and P. Mansfield [4]. The sequence consists of a single inversion pulse, either slice-selective or non-selective, followed by multiple EPl-readouts at low flip angles (~ = 30~ with an inter-image delay of TI z (e.g. 100 ms). The measured blood signal is disturbed b y the repeated perturbation of the readout pulses. The general kinetic model of Buxton et al. [5] is used. Using appropriate assumptions the difference or" the longitudinal magnetization at the capillary exchange site can be approxim~ited to AM(t)=0, for0
for za < t

where '1 = 0.9 ml/g is the blood-tissue partition coeff• Ma,o is the equilibrium magnetization of a voxel only containing blood spins; z~ is the arterial transit time of the tagged bolus, ~ is the flip angle; dR = Rla - Rlar,v,efr is the difference of the longitudinal relaxation rates; Rlaov,eee= R l +f/2 - ln(cosa)/ TI, is the effective apparent relaxation r a t e of tissue in presence of flow using Look-Locker. R~apv.efr can be calculated b y fitting the slice-selective inversionrecovery dataset to a modified inversion recovery curve. We use a 64 x 64-EPl readout on a standard 1.5 T scanner (MAGNETOM Vision, Siemens, Germany) with TE = 17.7 ms, TR = 2500 ms, 6 0 - 8 0 averages, slice thickness 8 ram, total measurement time about 5 min. After each EPIreadout the transversal magnetization is spoiled to avoid image artifacts. To achieve better results the gradient amplitude is varied [6].

Methodology: perfusion and diffusion Since the equation does not describe the intraarterial blood spins, the signal of these spins has to be suppressed by using crusher gradients in all spatial directions directly after readout preparation to dephase flowing spins [4]. Results and Conclusions: Fig. 1 shows the results of an healthy volunteer using ITS-FAIR. A single slice experiment was performed by sampling the inflow of the blood in the range from 0 to 2200 ms with a temporal resolution of 100 ms. Measurement time was about 9 min. Fig. 1 shows five out of 23 perfusion weighted difference images and the quantitative perfusion map calculated by using the equation given above.

103

flow-weighted 3D gradient echo pulse sequence, with an TE = 1.0 ms and TR = l heart cycle ~ 200 ms. 962 • 128 complex data points were acquired in approx. 30 min. Data were zero-filled before Fourier transformation to 1283 data points (spatial resolution 140 p.m). The signal of myocardial tissue was additionally supressed by a magnetization transfer experiment (Fig. 1).

Fig. 1. 3D angiogramms of the isolated rat heart: (a) before; (b) after acute stenosis.

Fig. 1. Five out of 23 difference images {Tl[ms] = 200, 600, 1000, 1400, 1800} of the blood inflow sampled with a temporal resolution of I00 m. The image at the bottom right shows the calculated perfusion map. A pulsation artifact can be seen clearly. In conclusion, we have demonstrated that it is possible to sample the blood inflow non invasively in about 5 min with a temporal resolution of about 100 ms using the new IT; FAIR technique. This technique is applicable to other inversion schemes such as EPISTAR [1]. Multi-slice measurements are also possible with this technique but the temporal resolution is lowered by the number of slices. This drawback can be overcome by using 3D-EPI giving the full temporal resolution for multiple partitions of slab, which will be the focus of future work.

Results: The changes of regional perfusion after acute stenosis determined by T~ measurements showed a similar decrease in the right and left ventricular myocardium (-29.83_+15.14~ and -32.94_+6.57%+SEM). In the ischaemic myocardium below the stenotic coronary vessel perfusion decreased dramatically ( - 66.76_+ 7.52%). During nitroglycerin stress right ventricular perfusion increased (10.47_+4.74%) whereas left ventricular perfusion showed no significant alteration (0.22 + 2.81%). The ischaemic myocardium however, showed a further slight decrease of perfusion ( - 0.73 + 1.92%). The relative changes in perfusion were refered to perfusion calculated by the ultrasonic measured coronary flow and heart weight at the starting point. Calculations of global perfusion changes by T 1 and ultrasonic measured coronary flow were in good agreement. The 3D angiogramms visualized the acute induced nonocclusive constriction of the left coronary artery. Also the diminished ramification of coronary vessels below stenosis became evident. Conclusion: It could be shown that it is possible to obtain accurate values of perfusion changes with the slice selective spin inversion NMR technique. The combination of the two NMR imaging techniques showed the possibility of detection of coronary stenosis and study of microcirculatory 'effects.

References [1] Hiller K.-H, et al. ESMRMB (abstract) 1997, p. 91. [2] Ruder F, et aL JMRI 1997"7:316-320. [3] Capasso JM, et al. Am J Physiol 1989;257:H1983-H1993.

References [1] [2] [3] [4] [5] [6]

~-]

Edelman RR, et al. Rad 1994;192. 513-520. Kwong KK., et aL MRM 34, 1995;878-887. Kim S-G, MRM 1995;34:293-301. Gowland P, et al. MRM 1993;30:351-354. Buxton RB. et al. MRM 1998;40:383-396. Gowland P, et al. MRM 1992;26, 79-881.

Assessment of variations of myocardial perfusion and 3D angiography under acute coronary stenosis in the isolated rat heart by NMR imaging

Karl-Heinz Hiller 1, Christiane Waller2, Matthias Nahrendorf1, Kai Hue 2, Sabine Voll ~, Cornelia Heindl ~, Georg Ertl 2, Axel Haase 1, Wolfgang R. Bauer2.

IPhysikalisches Institut, Universitat Wilt=burg, 97074 Wiirzburg, Germany: 2Medizinische Universitatsklinik Wiir-_burg,97080 Wiir=burg, Germany

Introduction: The aim of this study was to combine a non-contrast-agent NMR method of measuring perfusion and a NMR-microscopy method which reflects the state of coronary vessels to investigate the effects of acute stenosis under rest and stress. Both methods were developed by our group [1,2]. The principle of the perfusion measurement is that spins of a selected slice in the short axis view of the heart are inverted and a perfusion sensitive T~ relaxation of these spins is observed. According to model calculations variations of perfusion P are determined from:A(1/Tl) = AP/A (A = tissue/perfusion partition coefficient of water). Methods: Ten isolated rat hearts were studied (perfusion with Krebs Henseleit buffer in the Langendorff mode, on-line registration of coronary flow, left ventricular pressure). Measurements of perfusion changes and 3D imaging of the coronary arteries were performed before, 30 min after induction of the coronary stenosis (200 lain in diameter) [3] at rest and at stress during infusion of the vasodilator nitroglycerin (0.5 rag/rain). NMR-imaging was performed on an 11.75 T magnet (AMX 500, Bruker). Perfusion measurement: Spins of a slice (short axis view) 4 - 6 mm below the valvular plane were inverted (slice thickness = 3 mm) and T I maps were gained in this slice by 16 x Snapshot FLASH images (spatial resolution 140 pan in plane, slice thickness = 1.5 ram, TR = 3.6 ms). 3D angiography: Coronary vessels were imaged by middiastolic triggered

I25--6-~ M u l t i p l e slice, p e r f u s i o n w e i g h t e d M R I of the l u n g s R.A. Jones, E. Laffon, F. Laurent, C.T.W. Moonen. RMSB, Unwersitb Bor-

deaux 2 and H6pital Haut-Leveque, Bordeaux, France.

Introduction: An assessment of lung perfusion is important for the evaluation of patients with suspected pulmonary embolism. Despite the difficulties of imaging the lungs with MRI due to the rapid dephasing of transverse magnetisation several groups have shown that perfusion related information can be obtained with dynamic MR imaging sequences covering the first passage of a bolus of contrast agent. To date these studies have typically used either single slice techniques [1] or multiple slices with relatively poor temporal resolution [2], in this abstract we present our initial results using a multiple slice sequence with a relatively high temporal resolution and subsequent gamma variate fitting of the first pass data. Methods: A saturation prepared rapid gradient echo sequence (TE = 0.8 ms, TR = 2.7 ms, z = 16~ was implemented on a 1.5 T Philips Gyroscan ACS NT system using the body coil for transmission and reception. This sequence was used to measure seven 15 mm thick slices with a FOV of 420 mm and a temporal resolution of 1.9 s. After five images had been acquired the patient was instructed to perform a breathold with the contrast agent being injected once the patient had attained maximum inspiration, most patients were able to maintain the breathold for at least seven images, a dose of 0.075 mmol/kg Omniscan was typically used. The image data was then transferred to a UNIX workstation and evaluated using an in-house programme written in IDL which calculated a subtraction projection angiogram prior to segmenting the lungs and calculating subtraction images using the last image prior to the arrival of the contrast as a reference. The data were then spatially smoothed using a 3 • 3 median filter prior to gamma variate fitting on a pixel to pixel basis. Results: Of the 12 patients studied to date subtraction angiograms were obtained in all cases. In one case segmentation was not possible due to severe fibrosis of the lungs, in the other 11 cases gamma variate fitting was performed. In the cases where perfusion deficits were present these were seen in both the subtraction and gamma variate images. The arterial and venous phases of the first pass were observable and the gamma variate fitting provided consistent values for parameters such as the apparent mean transit time despite the poor SNR in the lung. An inter-patient comparison of the gamma variate parameters is complicated by the fact that large variations in the input haemodynamic parameters (as measured in the pulmonary artery) were present and deconvolu-

Methodology: perfusion and diffusion

104

tion techniques are rather unreliable at low SNR. In patients where the breathhold was poor the individual image quality was not degraded due to the very short acquisition time but the gamma varaite fitting was compromised in the inferior lobes, however, the gamma variate and subtraction images were still interpretable and perfusion deficits were still observable. The technique offers higher resolution than perfusion scintigraphy and appears to be a promising tool for the assessment of pulmonary perfusion.

References [1] Hatubo H, et al. Magn Reson Med 1996;36:503-508. [2] Amundsen T, et al. Radiology 1997;203:181-185.

~ 5 - ~ ] Diffusion tensor imaging: correction of anisotropic diffusion index maps J.G. Hirsch, J. Gaa, M. Hennerici, A. Gass. NMR Research Neurology~Radiology, Klinikum Mannheim/University of Heidelberg, Germany

Introduction: Fast DW MRI [1] has become an important tool in the examination of brain lesions. Furthermore, the quantification of the apparent diffusion coefficient (ADC) and its anisotropy in brain tissue is necessary for a more detailed characterization of lesions and evaluation of white matter structure changes during follow-up studies. Independent from patient positioning, diffusion tensor imaging (DTI) [2,3] offers the calculation of ADC maps and different indices describing directional information of diffusion. However, simulations and phantom experiments show a dependence of the RA index on the tensor elements' accuracy. With in vivo measurements, this may lead to misinterpretations of tissue alterations. We propose a correction of RA index maps for statistical influences. Methods: Simulations were done with IDL (Research Systems) generating Gaussian distributed coefficients with varying standard deviation o"and differing mean values Di= i-6. Measurements were performed on a 1.5 T scanner (VISION). The scheme consisted of a DW SEEPI acquisition (TIVTE 3000/110 ms, FOV 220 mm 2, MAT 128 x 128) with gradient lobes for DW ( b = 0 - 1 1 2 0 s/mm 2) following the scheme for DTI proposed in [4]. Twenty-five DW images were acquired including one without DW. Amplitude images were averaged from two measurements. Distortions due to residual eddy currents were corrected afterwards [6]. Six elements D: obtained by a pixelwise least-squares fit were combined to determine the tensor's elements D,j, i,j= x,y,:. Maps of the trace D and the relative anisotropy index RA were estimated according to [4.5] 1D O = ~ i = 1,6 ;

RA =

2y (Di,- 0 ) 2 + 2"(D~y+ O2,:+ D~:)

(1)

Results: Fig. la shows the dependence of the RA index (mean of the resulting distribution) on the standard deviation a of 6 Gaussian distributed coefficients D i from simulated data. As approximation, we propose an hyperbolic dependence (Fig. lb)

RA(#) = x/RA o + 1.138'o"2

RA

RA - ~RA~ + 1.138"o2

0,6

0.6

I ~..----"~J~.~,,,~ I o., i ~ / ~ " I 7 ~ / ~ 1o.2 ~__J-"_ ~ l ~o,o V-j,j / i sd~v,~i ~o .~F./~ ~"" tel o.0 o,,,

0.1

0,2

Conclusion: Inherently. indices like the relative anisotropy index in DTI depend on statistical conditions. This may lead to misinterpretation of tissue anisotropy if statistical conditions in regions differ significantly or change during follow-up examinations. The proposed correction compensates extreme statistical differences and makes comparisons of diffusion anisotropy indices more reliable. References [t] [2] [3] [4] [5] [6]

Turner R. Le Bihan D. Chesnick AS. Magn Reson Med 1991:19:247. Basser PJ, et at. J Magn Reson 1994;B 103:247. Pierpaoli C, et al. Radiology 1996;20 1:637. Basser PJ, et al. Magn Reson Med 1998:39:928. Shimony JS, et aL Proc 6th ISMRM: 1998:1241. Haseigrove JC, Moore JR. Magn Reson Med 1996:36:960.

(2)

With in vivo measurements, we estimated o"= EEj6, E~ is the statistical errors of the least-squares fit. The dependence of RA(a) may lead to misinterpretation of diffusion anisotropy in two typical brain pathologies: (A) Definite reduction of diffusion, but no hyperintensity in T2-weighted images, e.g. in hyperacute ischemia. DT elements in these regions are usually worse estimates resulting in an artificially elevated anisotropy index RA (Fig. 2). (B) Increase of diffusion and significant hyperintensity in T2-weighted, e.g. in edematous lesions (e.g. chronic infarction, MS). Within these regions, the statistical fit error is reduced considerably due to high signal intensity, therefore anisotropy might be underestimated in comparison to surrounding tissue.

0.0

Fig. 2. Patient study (58y) of hyperacute ischemia: (a) trace-ADC map showing reduced diffusion (arrows): (b) map of mean statistical fit error revealing two different regions probably of different ages: ROI 1 (hyperacute, no T2w hyperintensity) has significantly increased fit error (35%), fit error of ROI 2 (12%) - - acute, already prominent on T 2 w - - is similar to normal tissue. (c) Index map of relative anisotropy RA calculated according to Eq. 1. Considerably "increased" anisotropy is observed in ROI 1 (RA = 0.35) in comparison to ROI 2 (RA = 0.17). (d) Corrected RA index map (Eq. 2) showing reduced values in ROI 1 (RA = 0.26) and slightly reduction in ROt 2 (RA = 0.15).

0,3

0,4

0,0

0.1

0,2

0,3

0,4

Fig. l(a) Simulated data, RA calculated according to Eq.1 from Gaussian distributed (relative standard deviation a) coefficients D r, i = 1,6. Different degrees of anisotropy were simulated by choosing differing mean values of D i. (b) Plot of approximated functionality RA = RA(c).

Detection of cortico-spinai tract pathology in motor neuron disease using diffusion tensor imaging C.M. Ellis I, A. Simmons 1'2, D.K. Jones 3, J. Bland 1, J.M. Dawson 1"2. MA Horsfield3, SCR Williams h2, PN Leigh I. 1Department of Clinical Neuro-

sciences, Institute of Psychiatry and Guy's, King's and St Thomas" School of Medicine, 2Neuroimaging Department, Maudsley Hospital, and ~Dirision of Medical, University of Leicester, Leicester Royal Infirmary, UK Objective: To explore the hypothesis t h a t degeneration of the motor fibers in MND would be reflected by changes in t h e diffusion characteristics of the white matter fibers in the posterior limb of the internal capsule, and that these changes could be detected by diffusion tensor M R I . Methods: We studied 22 patients with El Escorial definite, probable or possible MND, 11 with limb onset (mean age 54.5 • 10.7 years) and 11 with bulbar onset (mean age 49.6 _+ 11.7 years), and compared them with 20 healthy, age matched controls (mean age 46.0 + 12.6 years). Clinical evaluations included measures of disease severity and extent (MRC grading for muscle strength, ALS severity scale) and measures of upper motor neuron (UMN) involvement (modified Ashworth spasticity scale and locally devised spasticity scale). Diffusion tensor MRI was carried out using a 1.5 T GE Siena system fitted with Advanced NMR hardware and software capable of producing echoplanar MR images. Data were acquired from seven coronal slices centred to include the posterior limb of the internal capsule. M a p s of the mean diffusivity, fractional anisotropy and T2-weighted signal intensity were generated.

105

Metltodology: perfitsion and diffusion Result: We found a significant increase in the mean diffusivity and reduction in fractional anisotropy along the cortico spinal tracts between the three subject groups, most marked in the bulbar onset group. The fractional anisotropy correlated with measures of disease severity and UMN involvement, whilst the mean diffusivity correlated with disease duration (Table ?).

Limb onset group Bulbar onset group Control group P-value

Mean diffusivity + SD mmZ/s ( x 10-3)

Fractional anisotropy + SD

0.77 + 0.05 0.78 +_0.03

0.76 __.0.05 0.73 + 0.05

0.73 + 0.02 0.001

0.77 + 0.02 0.007

Methods: We calculated the noise-induced bias of image intensity under the assumption of independent Gaussian noise in the real and imaginary part of the image data. In addition to the numerical evaluation [1], we determined analytical expressions of different order that approximate the solution. We verified our results with Monte Carlo simulations and MR measurements of a doped water phantom. The simulations were executed on a 128 • 128 matrix with a structured intensity pattern, superposed Gaussian noise and 10 simulated b values, resulting in SNRs between 3.2 and 1.5. A diffusion-weighted EPI sequence with strongly reduced T~ of 200 ms and b values of 0, 100, 200,..., 800 s m m - ~-was used for the MR measurements. With this sequence we performed reference measurements (high SNR) with a flip angle of 90~ and noisy measurements with a flip angle of 10~ Results: The results of the intensity correction are shown in Fig. 1; the calculated ADCs are presented in Table 1. Whereas the uncorrected values differ by 19% (simulation) and 11% (measurement) from the reference value, the difference is reduced to 1 and 2.5%, respectively after correction.

Conclusion: The results support the use of diffusion-tensor MRI in detecting pathology of the cortico spinal tracts in ALS. The correlation between fractional anisotropy and UMN involvement suggests the measure may provide an objective marker of UMN pathology, but longitudinal studies are now needed to demonstrate changes in this measure with disease progression.

~

C o r r e c t i o n o f A D C values a t small S N R for high-precision

ADC

MC Sim.

MR Meas.

Reference Not corrected Corrected

0.100 0.081 0.099

2.36 2.10 2.30

diffusion m e a s u r e m e n t s Olaf Dietrich, Sabine Heiland, Klaus Sartor. Department of Neuroradiology,

University of Heidelberg Medical School, INF 400, D-69120 Heidelberg, German), Introduction: Noise in MR image data increases the mean intensity of image regions due to the usually performed magnitude reconstruction [1-3]. In conventional imaging, the signal-to-noise ratio (SNR) can be improved by averaging the complex data of repeated excitations, and the remaining intensity bias usually can be neglected. This strategy is not applicable at in vivo diffusion imaging because the combination of motion-induced phase shifts and averaging will lead to partially annihilated image data. Hence, a higher level of noise must be generally expected in diffusion images and especially an increasing SNR with increasing diffusion weighting (b value). This varying bias of intensity causes systematic errors when calculating the apparent diffusion coefficient (ADC). We present a method to correct for these systematic errors. 1

o., s,g.a, I

corr. signal [ ~1-'-}-odg. noise J

0.9 0.8 0.7

m~ x

~'0.6

~ 0.5 0.4 0.3

+-.-I-, .+..+..+,,

+..+..-I-.+..+

0.2 0.1

0

0

2

4 6 b value (a. u.)

18(3 ~

8

10

IX-t< odg. signal

Discussion: The results show that noise can significantly influence the determination of ADCs. Increased signal intensities especially at large b values cause systematically reduced diffusion coefficients. This effect is important when precise absolute determinations of the ADC are required at measurements with high noise levels. We were able to correct this bias with our correction scheme. References [I] Henkelmann R. Med Phys 1985;12:232-233. [2] McGibney G. Smith MR. Med Phys 1992;20:1077-1078. [3] Gudbjartsson H, Patz S. Magn Reson Med 1995;34:910-914.

Selecting an appropriate anisotropy index for displaying

diffusion data with improved contrast and sensitivity P.A. Armitage, M.E. Bastin, I. Marshall. Department of Medical Physics and Medical Engineering, The University of Edinburgh, Western General Hospital

Crewe Road, Edinburgh EH4 2XU+ Scotland. Introduction: A review of the sensitivity of different diffusion anisotropy indices (DAIs) to changes in anisotropy was undertaken and as a result a method was derived for displaying anisotropy maps with improved contrast in human brain diffusion tensor imaging (DTI). Methods and Results: The standard deviation ( A j . volume ratio (Avr), fractional anisotropy (A,~) and eigenvalue ratio (Aer) DAIs were studied [1.2]. Fig. 1 shows the behaviour of these indices relative to As,t, as this conforms to the chemical definition of anisotropy. The gradient of these plots, dA//dAsd ( j = vr. fa, er), indicates the sensitivity of the respective anisotropy index, i.e. the change in measured anisotropy per unit change in A~d, and this is shown in Fig. 2. A histogram of typical Asd values found in the human brain (Fig. 3) was then used to devise a new anisotropy display index (A~), that has a low sensitivity to noise, but still retains a good contrast range over typical human brain anisotropies. Fig. 4 shows anisotropy maps from a typical transverse section of the human brain, corresponding to the histogram in Fig. 3. (Elscint Prestige 2.0T, b = 100 and 1000 smm - 2 EPI-DTI sequence.)

1

12(3

,

9

le+~++ee I .~+o*

0.8 0.6

6C

<-

..-...-

0.4

9 9 e+~

.%" .~/..'+ 0

0.2 200

400 600 + b value / (s/mm2l

_ "

,,~m 1 - A w . 9 9 .% 9 9 9 Aw

o~e~Pr*** #v

800

Fig. 1. Results of intensity correction: Monte Carlo simulation in left diagram, MR measurement in right diagram.Table 1: Calculated ADCs from simulation and measurements

c;

""

o.'2oi,,

A~

o~e

Fig. 1 Anisotropy variation.

o18

Pulse sequences MR1

106

References 2.7m- Oe 2F2

-*

9 e e~

~ l~_

Oe

~< / . . . . "o

IL

1

9

X

."

a~ e 9e

m"

Aul

oee

Am

w

999 A e o ~ Awv

.

**~o ae~

*%

9

[1] Conturo TE, et aL Magn. Reson. Med. 1996;35:399-412. [2] Basser PJ. N M R Blamed. 1995;8:333-344.

Pulse sequences MRI -~MR

C A T s c a n : a m o d u l a r a p p r o a c h for h y b r i d i m a g i n g

- " ; | : : - . , . . _ ", C. Hillenbrandl'2; D, Hahn 2, A, Haase I, P.M. Jakob 1. IPhysikalisches bzstitut, EP5 and 2Institut far Ri~ntgendiagnostik, Universitiit W~rzburg, Wiirzburg, Germany

Aid Fig. 2. Anisotropy sensitivity.

60(

0,2

0.4

A~

0~

0,8

Fig. 3. Histogram of brain anisotropy values.

Introduction: A versatile concept for N M R hybrid imaging is presented. This concept essentially integrates different imaging approaches in a sequential fashion and is therefore called CAT (Combined Acquisition Technique). CAT is not a single specific measurement sequence, but rather a sequence concept whereby distinct acquisition techniques with varying imaging parameters are employed in rapid succession. One important CAT sequence optimization feature is the application of mixed bandwidth technology. Details of both the CAT methodology and possible CAT acquisition strategies are provided in this work. Methods: The essential basis of the CAT concept is to measure the central part of k-space accurately and artifact free, whereas for the rest of k-space, a reduced sequence performance is accepted. This approach relies on the facts that the overall image appearance is largely dictated by the portion of k:space confined around its center, and that the peripheral parts of k-space can therefore be measured vath reduced accuracy. With CAT. the central echoes are acquired with an imaging module of choice which dominates the overall sequence behaviour, whereas in contrast the peripheral k-space lines are acquired with a different (mostly faster) module. The optimal CAT sequence design involves the application of mixed bandwidth technology. Such a mixed bandwidth CAT approach employs e.g. a relatively long data sampling period (narrow bandwidth) for the center of k-space (to increase SNR etc.) and a relatively short data sampling period (wide bandwidth) for the higher spatial frequencies (to increase speed or to reduce artifacts from off-resonance signals). Experimental:A variety of segmented and single-shot MR-CAT scan techniques including mixed bandwidth FLASH/EPI-CAT, RARE/EPI-CAT. FLASH/ BURST-CAT, GRASE/RARE-CAT were implemented on a Siemens Vision 1.5 T whole body MR scanner using standard gradient- and RF-hardware. In order to validate the CAT concept as well as to demonstrate the inlage quality with the different CAT techniques, a number of experiments in phantoms and in humans have been completed. Results: The in viva images obtained with the different CAT scans showed promising results in brain, abdominal a n d thoracic imaging. High quality cardiac images yielding an in-plane resolution of 800 ~tm have been obtained reproducibly in short breathhold times ( 1 0 - 2 0 s) with the mixed bandwidth FLASH/EPt-CAT approach (Fig. t). G o o d quality single shot brain and cardiac images were obtained in 150-200 ms (matrix 65 • 256) with both RARE/EPI- and mixed bandwidth FLASH/EPI-CAT scans.

Fig. 4.

Discussion: The regions where the DAIs show maximum sensitivity determine the range of auisotropy values that provide most of the image contrast. Aer, and to a lesser extent Afa, show the greatest sensitivity to isotropic diffusion, as illustrated by the overall brightness of these images. As such, they are extremely sensitive to noise contamination, resulting in CSF appearing relatively bright for isotropic diffusion. Aw shows CSF to be very dark because it has a low sensitivity to isotropic diffusion and hence noise. However. the contrast is relatively poor because, as is seen in Fig. 3, Asa largely takes values of less th;~n 0.5 and so Avr has high sensitivity over a range of anisotropies that have a low pixel count. Similarly. Am retains a relatively high sensitivity to anisotropic diffusion and so produces an image that is also fairly dim. To improve contrast in the anisotropy maps, a new index (A~) was devised such that its sensitivity dA~v/dA~d takes the form of a gamma-variate function, with a peak at around A~a = 0.2, corresponding to the maximum value in Fig. 3. The properties of this function are illustrated in Figs. 1, 2 and 4. As can be seen Agv shows low sensitivity to isotropic diffusion, helping to eliminate noise bias, but has a high sensitivity for the majority of anisotropies found in the human brain, resulting in improved contrast. Conclusion: The sensitivity of a DAI enables anisotropy contrast to be chosen for a given application by selecting a DAI with maximum sensitivity over the required range of anisotropies, Other DAIs can be derived that give good contrast for a specific application, as has been demonstrated in the brain.

Fig. 1. High resolution cardiac CAT scan: Stills from a multislice segmented FLASH/EPI-CAT Scan for coronary imaging. Each slice was acquired in 20 heart beats with an in-plane resolution o f 720 x 840 Jam. the boxed areas are magnifications of the left coronary artery (diameter < 1.5 mm).

Conclusion: In this work CAT, a hybrid imaging concept, which combines aspects of different acquisition techniques, was introduced. The principal idea of CAT is that the imaging experiment is no more restricted to a single specific imaging sequence with fixed parameters. Therefore, CAT opens a completely new field of hybrid imaging expertments, t h a t allows a flexible trade-off between