Optimal inversion time for acquiring flow-sensitive alternating ...

Eur Radiol (2002) 12:2950–2956 DOI 10.1007/s00330-002-1364-8

Ali Syed Arbab Shigeki Aoki Keiji Toyama Tsutomu Araki Nobuhiko Miyazawa Hiroshi Kumagai Takako Umeda Takao Arai Toshiyuki Okubo Hiroyuki Kabasawa Yoshiyuki Takahashi

Received: 10 July 2001 Revised: 27 December 2001 Accepted: 18 January 2002 Published online: 30 April 2002 © Springer-Verlag 2002

A.S. Arbab · S. Aoki · K. Toyama T. Araki (✉) · H. Kumagai · T. Umeda T. Arai · T. Okubo Department of Radiology, Yamanashi Medical University, Shimokato 1110, Tamaho-cho, Nakakoma-gun, Yamanashi 409-3898, Japan Tel.: +81-55-2736744 Fax: +81-55-2736744 N. Miyazawa Department of Neurosurgery, Yamanashi Medical University, Shimokato 1110, Tamaho-cho, Nakakoma-gun, Yamanashi 409-3898, Japan H. Kabasawa · Y. Takahashi GE-YMS, Asahigaoka, Hino-shi, Tokyo, Japan Present address: A.S. Arbab, LDRR/CC, Building 10, Room no. B1N256, National Institutes of Health, Bethesda, MD 20892, USA

NEURO

Optimal inversion time for acquiring flow-sensitive alternating inversion recovery images to quantify regional cerebral blood flow

Abstract The purpose of this study was to correlate regional cerebral blood flow (rCBF) measured by I-123-IMP and flow-sensitive alternating inversion recovery (FAIR) studies with different inversion time (TI) values to find out the optimum TI that gives comparable rCBF and images on FAIR study. Nine patients with symptoms and signs of internal carotid or major cerebral arterial stenosis were enrolled in this study. Both I-123-IMP single photon emission computed tomography (SPECT) and FAIR images were acquired in all patients. Single-slice FAIR images (with different TI values) were acquired using a 1.5-T MRI unit. The rCBF was calculated from all I-123-IMP and FAIR images. Receiver operating characteristics (ROC) analysis was performed to detect hypoperfused segments on FAIR images. The rCBF calculated from FAIR and I-123-IMP studies were compared and correlated with each other. The ROC analysis showed no significant differences among the readers or TI

Introduction Assessment of cerebral perfusion by nuclear medicine techniques is well established, especially by single photon emission computed tomography (SPECT) and positron emission tomography (PET). Moreover, measurement of regional cerebral blood flow (rCBF) by various nuclear medicine brain imaging agents showed good correlation with the gold standard of rCBF measurement ei-

values, but a trend of higher sensitivity, specificity, and accuracy was observed with TI of 1400 ms. The rCBF values of FAIR and I-123-IMP studies significantly correlated with each other. The FAIR images with TI value of 1400 ms gave more comparable CBF. A TI value of 1400 ms might be optimum for 1.5-T MR strength to get high quality FAIR images and comparable CBF. Keywords Cerebral perfusion · MRI · FAIR · SPECT · I-123-IMP

ther by radioactive microspheres or xenon-133 (Xe-133) studies [1, 2, 3, 4, 5]. Measurements of rCBF by iodine123-iodoamphetamin (I-123-IMP) using autoradiographic (ARG) and a modified ARG method (table look up) have shown very good correlation with PET, microspheric assessment of CBF, and Xe-133 studies [3, 6, 7, 8]. With advancement in MR hardware and software, it is now possible to acquire various new faster sequences easily. In recent years, several MR imaging pulse se-

2951

quences using arterial spin labeling (ASL) have been developed to assess the cerebral perfusion and rCBF without contrast administration [9, 10, 11, 12]. Flow-sensitive alternating inversion recovery (FAIR) is one of the sequences that can assess cerebral perfusion without contrast administration [13]. A few preliminary reports have shown FAIR’s utility in functional brain imaging [9, 14]. Measurement of quantitative rCBF using FAIR images is possible by measuring the T1 and M0 values of brain tissue in each case [14]. Quantification of blood flow through middle cerebral arteries has been tried by phase-contrast MR imaging, but rCBF could not be ascertained [15]; however, there has been no study of a direct correlation with conventional methods of assessing cerebral perfusion or rCBF. Moreover, flow signal intensity on FAIR images from different vascular compartments depends on the inversion time (TI). It is well established that a shorter TI will give signal mostly from the flow through larger arteries and a long TI will give signal from smaller capillary at the site of blood to tissue exchange of spins [9, 14]. Until now there has been no correlative study to optimize TI to quantify rCBF from FAIR images. The purpose of this study was to correlate rCBF measured by I-123-IMP with that of FAIR studies using different TI values to find out the optimum TI that would give comparable rCBF on FAIR images.

sity hospital ethical committee. No adverse reactions or patient complaints were recorded during the study. SPECT studies The rCBF was measured by the modified ARG method (table look-up method) from I-123-IMP SPECT study according to the method of Iida et al. [7, 8]. Both early and delayed I-123-IMP studies were performed in all patients. Early SPECT images were obtained after 20 min of injecting I-123-IMP (167–222 MBq) for 20 min (mid scan time was 30 min after injection) by a Toshiba GCA3000A/DI triple-head gamma camera mounted with fan beam collimators. The images were acquired in a 128×128 matrix, by continuous rotation (10 rotations, 2 min/rotation) with 4° steps using the triple energy window method (TEW). The SPECT images were reconstructed from the back-projected data using Butterworth and Ramp filters after proper scatter correction. The thickness of axial images was 6.8 mm. Delayed images were acquired at 3 h after the injection with similar acquisition parameters. Delayed SPECT images were necessary to calculate the distribution volume (Vd) and global CBF for modified ARG method to measure rCBF. A single-point brachial arterial blood sample was collected in each patient at 10 min after injection of I-123-IMP and the radioactivity was measured by a auto gamma well counter. The blood sample was collected from contralateral side of the injection. The mid section of the cerebrum (along basal ganglia and thalami) was selected in each patient to measure the rCBF. The rCBF values of the whole cerebrum, cerebral hemispheres, thalami, basal ganglia, frontal, posterior parietal, temporal, and occipital regions on both sides were measured by creating regions of interest (ROI) over the regions. The rCBF of a total of 15 regions was measured in each patient. The size of the smallest ROI was at least 10 pixels.

Materials and methods

FAIR imaging

A total of nine patients (six men, three women; age range 48–69 years) were enrolled in this comparative study (Table 1). Seven patients had various degrees of either internal carotid artery (ICA) and/or other major cerebral artery stenosis. Two patients had transient ischemic attacks and dizziness, but neither MRI nor SPECT studies showed abnormality in the brain or cerebral perfusion, respectively. All patients underwent both I-123-IMP and FAIR studies. I-123-IMP study was performed usually 3–7 days before FAIR studies. I-123-IMP SPECT study was part of routine investigation for cerebral perfusion. The FAIR imaging protocol was approved and in accordance with the guidelines of the univer-

Single slice FAIR studies with different TI values (1200, 1400, and 1600 ms) were performed according to the method of Kwong et al. [13] by a 1.5-T superconducting MR system (Signa Horizon, GE Medical Systems, Milwaukee, Wis.) using spin-echo (SE) echo-planar imaging (EPI) sequences [TR 2000 ms, TE 20.6 ms, bandwidth 83 kHz, field of view (FOV) 24×24 cm, matrix 96×96, no. of excitations (NEX) 100 (50 pairs of images), slice thickness 10 mm] with a head coil. To calculate quantitative rCBF from FAIR images, both T1 and M0 values of gray and white matters were measured in all patients by acquiring SE EPI images with different TI (200, 400, and 800 ms). All images were acquired at

Table 1 Details of patient profile, investigations, and diagnosis. FAIR Flow-sensitive alternating inversion recovery; ICA internal carotid artery; MCA middle cerebral artery; ACA anterior cerebral

artery; MRA magnetic resonance angiography; DSA digital subtraction angiography; SPECT single photon emission computed tomography; TIA transient ischemic attack

Patient no.

Age (years)/gender

Date of IMP SPECT

Date of FAIR

Diagnosis

1 2 3 4 5 6 7 8 9

68/M 63/M 48/M 51/M 53/F 69/F 57/F 55/M 69/M

99.10.26 99.11.02 00.01.14 00.01.25 00.01.31 00.02.25 00.02.29 00.03.31 00.06.09

99.11.02 99.11.05 00.01.18 00.01.28 00.02.08 00.03.03 00.03.07 00.04.04 00.06.13

Right ICA severe stenosis on MRA and DSA Right ICA severe stenosis (90%), left ICA stenosis (40%) on MRA and DSA Right MCA (M1) occlusion with collateral circulation on DSA TIA but normal on MRI and SPECT Right ICA, occlusion, left ICA stenosis, left ACA occlusion on DSA Bilateral mild ICA stenosis with lacunar infarctions on MRA and MRI Dizziness but normal flow and perfusion on MRA and SPECT Right ICA, MCA, and ACA stenosis on MRA and decreased perfusion on SPECT Right ICA stenosis on MRA and mild decreased perfusion on SPECT

2952

the mid section of the cerebrum (along the basal ganglia and thalami) in each patient. The total time required to acquire one image was 3 min 36 s. After transferring the images to a workstation, 15 ROIs were created over the whole cerebrum, cerebral hemispheres, thalami, basal ganglia, frontal, posterior parietal, temporal, and occipital regions on both sides to record the average signal intensity for measuring rCBF of these regions. The size of the smallest ROI was at least 10 voxels. Much care was taken not to include the high intensity area in the venous sinuses and peripheral vessels.

Image selection and analysis All FAIR images were obtained during the same session without moving the patient from the headrest. Although image co-registration was not applied to select the CBF images of I-123-IMP study, a matching slice along the mid section of cerebrum was selected from the I-123-IMP study in all patients. One investigator selected all these image sections and the same person also drew the ROIs to minimize false registration of ROIs.

ROC analysis A total of 108 segments from nine patients (bilateral occipital, parietal, temporal, frontal, basal ganglia, and thalami) were entered into the ROC analysis. Eighty of the 108 segments showed normal perfusion on I-123-IMP SPECT (any segment that showed ≥30 ml/100 g min–1 rCBF on rest with increase of flow after diamox was considered normal). Three independent readers analyzed all FAIR images. None of the three readers were aware of any patient’s history, and had not participated in consensus readings of the images. The readers were asked to score the segments in each patient with a confidence level of 1–5 based on their observations (1 definitely normal, when signal intensity of the segment is similar to those of other segments; 2=probably normal, when overall signal intensity in the segment is similar but slightly different from those of other segments; 3=equivocal, when signal intensity in the segment is overall decreased but present; 4=probably hypo, when signal intensity is overall decreased with areas of absent signal within the segment compared with those of other segments; and 5=definitely hypo, when signal intensity is absent in the segment on FAIR images). For FAIR images with each TI, a binormal ROC curve was fitted to each reader’s confidence rating data using maximum likelihood estimation [16]. The diagnostic accuracy of FAIR images with different TI was determined by calculating the area under each reader-specific binormal ROC curve (Az index) plotted in the previously designated square [17]. Differences between ROC curves of individual reader were tested for significance by a bivariate method of the ROCKIT algorithm using the chisquared test [18]. Composite ROC curves for FAIR image with each TI were calculated by averaging the binormal parameter values of three readers. Differences among the FAIR images of different TI in terms of Az indices were analyzed statistically by the analysis of variance (ANOVA) post hoc (Bonferroni) test. A p-value of less than 0.0167 was considered significant for this table of data. Any segment that showed a confidence level of 4 or more was considered an abnormally perfused area on FAIR images. The sensitivity, specificity, and accuracy of detecting abnormally perfused segments using FAIR images with each TI were calculated for each reader. A post hoc ANOVA test (Bonferroni) was applied to assess the significant differences among the readers. A p-value of less than 0.0167 was considered significant for this table of data.

Calculation of rCBF from FAIR images The following formulas were applied to calculate absolute rCBF from FAIR images in each patient according to Kim and Tsekos [14].

where, S(FAIR)=average signal intensity in the ROIs on FAIR images, ƒ=rCBF (milliliters of blood per gram per second), TI= inversion time (1200, 1400, or 1600 ms), M0=unperturbed longitudinal magnetization of tissue–water proton (arbitrary unit), T1=longitudinal relaxation time of tissue–water proton (in milliseconds), TR=repetition time (2000 ms), λ=tissue–blood partition coefficient [(grams of water/grams of tissue)/(grams of water/milliliter of blood)], and it is assumed to be 0.9. Both T1 and M0 values of tissue–water proton in gray and white matters were measured by acquiring SE EPI images with different TI (200, 400, and 800 ms) using MRvision software (MRVision Co., Menlo Park, Calif., USA) in a workstation (Ultrasparc, Sun Microsystems, Mountain View, Calif.). The measured T1 of gray and white matters were 0.83±0.04 and 0.55±0.03 s, respectively (gray to white matter ratio was 1.5). The measured M0 was 1198.31±129.30 (arbitrary unit). Data analysis The rCBF measured (total 135 segments, 15×9) from FAIR and I-123-IMP studies in all patients were correlated among each other. The simple linear regression method was applied to correlate among the studies. A p-value of less than 0.05 was considered significant.

Results ROC analysis Figure 1 shows the composite ROC curves for each TI of image acquisition. No significant differences of Az val-

Fig. 1 Composite receiver operating characteristics (ROC) curves and Az values of the areas under curves in detecting abnormal segments on FAIR images with different TI. No significant difference was observed among the Az values. TPF true-positive fraction; FPF false-positive fraction

2953

Fig. 2a–d A representative case showing normal perfusion on I-123-IMP study. a–d Images of I-123-IMP, flow-sensitive alternating inversion recovery (FAIR) with 1200, 1400, and 1600 ms, respectively. Note that there is almost similar image quality seen with TI of 1200 and 1400 ms, but decreased signal intensity causing less contrast image quality is observed with that of 1600 ms. Regional perfusion on FAIR images is similar to that of I-123-IMP image. White area represents highest and blue area represents lowest perfusion on I-123-IMP image (a). Red area represents highest and blue area represents lowest perfusion on FAIR images (b–d)

Table 2 Combined sensitivity, specificity, and accuracy of detecting abnormally perfused segments by different inversion time (TI) on receiver operating characteristics analysis by three independent readers TI (ms)

Sensitivity

Specificity

Accuracy

1200 1400 1600

65.00±5.20 67.67±5.51 63.00±10.58

84.33±3.05 85.33±2.08 83.67±3.21

80.33±0.58 81.67±1.15 79.00±3.46

ues were found among the readers for any of the TI used for image acquisition. The areas under composite ROC curve (Az index) were 0.78±0.08 for 1200 ms, 0.76±0.11 for 1400 ms, and 0.75±0.07 for 1600 ms TI. No significant differences were found among the Az values of composite curves. Table 2 shows the cumulative sensitivity, specificity, and accuracy in detecting hypoperfused segments on dif-

ferent TI. Although no significant differences were observed among different TI for sensitivity, specificity, and accuracy, a trend of higher sensitivity, specificity, and accuracy was observed with TI of 1400 ms. All readers also showed a trend of higher specificity (83–87%) and accuracy (81–83%) with TI of 1400 ms in detecting hypoperfused segments on FAIR images. Figure 2 shows representative images of FAIR and I-123-IMP studies in a patient with almost normal perfusion. There is similar image quality seen with TI of 1200 and 1400 ms, however, decreased signal intensity causing less contrast image quality is observed with that of 1600 ms. Regional perfusion on FAIR images showed similarity to that of corresponding I-123-IMP image. In four patients with ICA stenosis or occlusion showed linear high signal intensity along the sulci at hypoperfused segments. The rCBF values of the normally perfused segments was 41.53±8.56 ml/100 g min–1 on I-123-IMP studies.

2954

I-123-IMP and FAIR studies with all TI values, but a relatively better correlation was found with TI 1400 ms.

Discussion

Fig. 3a–c Regression plots of rCBF measured from I-123-IMP and FAIR images using different TI. All plots show significant correlation between I-123-IMP and FAIR images. A higher correlation coefficient is observed between I-123-IMP and FAIR with TI of 1400 ms

On FAIR images, the average signal intensities of normally perfused segments were 4.7±1.03, 4.6±1.35, and 4.1±1.09 (p5 cm/s, arterioles 1 cm/s) [19]. In our study, use of this TI value (0.829 s) would have produced signal mostly from the macrovascular compartment, because we used a thick inverted slab (three times of 10-mm-thick slice) and fresh blood would not have traveled all the way to capillary bed by this time. It would give an image of blood flow through larger vessels rather than real perfusion. Although there are reports suggesting that a longer TI would give better image quality, optimization of TI or limitation of very long TI (considering the image thickness and MR system) has not been discussed [9, 14]. In this study a relatively longer TI (1400 ms) showed better image quality and contrast than those of 1200 and 1600 ms, although there was no significant difference observed on ROC analysis. The longest TI (1600 ms) used in this study showed a significant decrease in the signal intensity and contrast of FAIR images, although calculated rCBF was higher than those with 1200 and 1400 ms. Kim and Tsekos found better image quality at TI of 1.6 s in their MR system (4 T) where T1 of tissue was 1.3–1.4 s [9, 14]. Our results also support their findings that a longer TI (which allows blood to travel to the capillary beds and greater than T1) would give better image quality and contrast. But decline of image quality and intensity at TI of 1600 ms also indicates that a proper optimization of TI value for a given MR system is needed. This TI of 1600 ms in our system might have been too long to produce better images, because according to the foregoing formulas a longer TI (TI>T1) would give decreased signal intensity images. The higher calcu-

2955

lated rCBF on images with TI of 1600 ms might be due to the influence of TI value rather than signal intensity. Although there was correlation observed between the ratios of intensity of FAIR images and I-123-IMP study in previous study [20], absolute rCBF from FAIR images was not measured. Significant correlation among rCBF measured by FAIR and I-123-IMP studies indicated that FAIR study might be a complimentary investigation to measure rCBF during routine MRI with only additional 4 min. Moreover, higher specificity and accuracy of FAIR images with all TI values (higher trend with 1400 ms) indicated that FAIR images with proper TI would be a complimentary investigation to assess cerebral perfusion; however, small correlation coefficient might raise a question of clinical utilization of FAIR images. The smaller correlation coefficient might be due to inclusion of linear high signal intensities present at hypoperfused segments, which was not excluded from the ROI (although intensity from peripheral vessels was excluded). These linear signal intensities might have influenced the rCBF calculated from average intensity on ROI. The linear high signal intensity seen on FAIR images at hypoperfused segments might be due to delayed flow or flow through collateral vessels which corresponds to the time of acquisition (TI values). The linear intensity was more prominent on images with shorter TI (1200 ms). In all cases these linear high signal intensities

at hypoperfused segments were easily identified and could be separated from flow through small arterioles and capillary beds in tissue. Addition of a sequence with FAIR imaging to suppress flow effect might eliminate this problem. Or development of software to eliminate high signal intensity by means of threshold (upper threshold) would be useful. Significant correlation with I-123-IMP, higher specificity and accuracy in detecting hypoperfused segments, and absence of contrast administration might make the FAIR imaging a complimentary investigation on routine MR imaging to assess the brain perfusion in addition to contrast-enhanced MR perfusion and diffusion-weighted imaging. A multislice image acquisition would make FAIR imaging a more acceptable tool for assessing whole-brain perfusion without adding any cost and significant time [21]. Flow-sensitive alternating inversion recovery along with diffusion-weighted imaging in patients with acute cerebral infarction might help assessing the potential area of damage after infarction. Quantification of rCBF by FAIR might help clinician to decide method of treatment. Higher correlation with I-123-IMP of FAIR images with TI 1400 ms in our study indicates that a TI of 1400 ms would be optimum for FAIR imaging with a 1.5-T MR system. An optimized TI value for any given MR system is essential to get higher-quality FAIR images that would give comparable CBF to the established method of CBF measurement.

References 1. McHenry LC Jr, Merory J, Bass E, Stump DA, Williams R, Witcofski R, Howard G, Toole JF (1978) Xenon-133 inhalation method for regional cerebral blood flow measurements: normal values and test–retest results. Stroke 9:396–399 2. Nakano S, Kinoshita K, Jinnouchi S, Hoshi H, Watanabe K (1989) Comparative study of regional cerebral blood flow images by SPECT using xenon133, iodine-123 IMP, and technetium99 m HMPAO. J Nucl Med 30:157– 164 3. Nakano S, Kinoshita K, Jinnouchi S, Hoshi H, Watanabe K (1989) Critical cerebral blood flow thresholds studied by SPECT using xenon-133 and iodine123 iodoamphetamine. J Nucl Med 30:337–342 4. Greenberg JH, Kushner M, Rango M, Alavi A, Reivich M (1990) Validation studies of iodine-123-iodoamphetamine as a cerebral blood flow tracer using emission tomography J Nucl Med 31:1364–1369

5. Matsuda H, Higashi S, Tsuji S, Seki H, Sumiya H, Fujii H, Oba H, Terada H, Imai K, Tonami N et al. (1987) A new noninvasive quantitative assessment of cerebral blood flow using N-isopropyl(iodine 123)p-iodoamphetamine. Am J Physiol Imaging 2:49–55 6. Ohkubo M, Odano I, Takahashi M (1997) A comparative study of simple methods to measure regional cerebral blood flow using iodine-123-IMP SPECT. J Nucl Med 38:597–601 7. Iida H, Akutsu T, Endo K, Fukuda H, Inoue T, Ito H, Koga S, Komatani A, Kuwabara Y, Momose T, Nishizawa S, Odano I, Ohkubo M, Sasaki Y, Suzuki H, Tanada S, Toyama H, Yonekura Y, Yoshida T, Uemura K (1996) A multicenter validation of regional cerebral blood flow quantitation using [123I]iodoamphetamine (IMP) and single-photon emission computed tomography (SPECT). J Cereb Blood Flow Metab 16:781–793 8. Iida H, Itoh H, Nakazawa M, Hatazawa J, Nishimura H, Onishi Y, Uemura K (1994) Quantitative mapping of regional cerebral blood flow using iodine123-IMP and SPECT. J Nucl Med 35:2019–2030

9. Kim SG (1995) Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med 34:293–301 10. Roberts DA, Detre JA, Bolinger L, Insko EK, Leigh JS Jr (1994) Quantitative magnetic resonance imaging of human brain perfusion at 1.5 T using steady-state inversion of arterial water. Proc Natl Acad Sci USA 91:33–37 11. Edelman RR, Siewert B, Darby DG, Thangaraj V, Nobre AC, Mesulam MM, Warach S (1994) Qualitative mapping of cerebral blood flow and functional localization with echo-planar MR imaging and signal targeting with alternating radio frequency. Radiology 192:513–520 12. Chen Q, Siewert B, Bly BM, Warach S, Edelman RR (1997) STAR-HASTE: perfusion imaging without magnetic susceptibility artifact. Magn Reson Med 38:404–408

2956

13. Kwong KK, Chesler DA, Weisskoff RM, Donahue KM, Davis TL, Ostergaard L, Campbell TA, Rosen BR (1995) MR perfusion studies with T1weighted echo planar imaging. Magn Reson Med 34:878–887 14. Kim SG, Tsekos NV (1997) Perfusion imaging by a flow-sensitive alternating inversion recovery (FAIR) technique: application to functional brain imaging. Magn Reson Med 37:425–435 15. Stock KW, Weizel SG, Lyrer PA, Radii EW (2000) Quantification of blood flow in the middle cerebral artery with phase-contrast MR imaging. Eur Radiol 10:1795–1800

16. Metz CE (1986) ROC methodology in radiologic imaging. Invest Radiol 21:720–733 17. Hanley JA, McNeil BJ (1982) The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143:29–36 18. Metx CE, Herman BA, Roe CA (1998) Statistical comparison of two ROCcurve estimations obtained from partially paired database. Med Decis Making 18:110–121 19. Van As H, Schaaffsma TJ (1985) Flow in nuclear magnetic resonance imaging. In: Peterson SB, Muller RN, Rinck PA (eds) Thieme, New York, pp 68–95

20. Arbab AS, Aoki S, Toyama K, Kumagai H, Arai T, Kabasawa H, Takahashi Y, Araki T (2001) Brain perfusion measured by flow-sensitivealternating-inversion-recovery (FAIR) and dynamic susceptibility contrastenhanced magnetic resonance imaging: comparison with nuclear medicine technique. Eur Radiol 11:635–641 21. Kim SG, Tsekos NV, Ashe J (1997) Multi-slice perfusion-based functional MRI using the FAIR technique: comparison of CBF and BOLD effects. NMR Biomed 10:191–196