- Email: [email protected]
Semiautomated analysis of the extent and severity of perfusion defects on brain SPECT images: validation studies
Journal of Clinical Neuroscience(1999) 6(2), 121-127 © 1999 Harcourt Brace & Co. Ltd
Clinical studies
S e m i a u t o m a t e d analysis of the e x t e n t and s e v e r i t y of perfusion d e f e c t s on brain SPECT images: validation studies Alison E. Baird 1,3 FRACP, PHD, Mark C. Austin 2 DIP APPL SCI, Graeme J. O'Keefe 2 PHD, W. John McKay = FRACP, Geoffrey A. Donnan 1,3 MD FRACP 1Departmentof Neurology 2Departmentof NuclearMedicine,Austin and RepatriationMedicalCentre,Victoria,Australia 3Universityof Melbourne,Victoria,Australia
Summary Single photon emission computed tomography (SPECT) is a widely available and practical functional imaging technique with established clinical and research applications in neurological disorders such as epilepsy and stroke. SPECT images are usually analysed visually, or semiquantitativety by measuring side-to-side asymmetries. In order to evaluate perfusion change after thrombolytic therapy in patients with acute ischaemic stroke we developed a semiautomated, weighted volumetric analysis (the ischaemic index) that measured the extent and severity of tracer uptake abnormalities on brain SPECT images semiquantitatively. The unaffected cerebral hemisphere was used as the reference region of interest. The analysis was validated in a phantom brain model incorporating 'strokes' varying in size and degree. The phantom 'stroke' sizes measured with the ischaemic index analysis correlated closely with the true values (r=0.994, P < 0.01) and this correlation was maintained under low count conditions acquired to simulate the clinical setting. The overall operator dependent error of the analysis was _+3.4%. In 30 patients treated with thrombolytic therapy who were studied serially with 99rnTc_hexamethylpropyleneamine oxime (HMPAO) SPECT, the analysis was used to measure hypoperfusion volume and provide indices of perfusion change. This analysis has the advantages of semiautomation, ease of use and validation and has a potentially wide range of applications for both SPECT and PET. Keywords: SPECT, emission computed tomography, quantitification, volume quantitation
INTRODUCTION Brain single photon emission computed tomography (SPECT) is the only tomographic functional imaging technique currently accessible to the majority of physicians, and is attracting increasing interest among neurologists and neurosurgeons because of its current or potential applications in areas such as epilepsy, stroke, dementia and cerebral neoplasia. In patients with ischaemic stroke the measurement of hypoperfusion with SPECT has been shown to be sensitive in cortical lesions and may provide prognostic information. ~-3 Large severe reductions in perfusion may correlate with a high mortality rate and with poor prognosis. ~-3 SPECT has also attracted interest for its potential use in the evaluation and selection of patients for stroke therapies .4-6 Scanning can be delayed until clinically convenient as radiopharmaceuticals such as 99mTc-hexamethylpropyleneamine oxime (HMPAO) are trapped in the brain within 2 min of i.v. administration and show minimal redistribution. 7-8 In most SPECT studies images have been interpreted visually or by the measurement of side-to-side or lesion-to-cerebellar ratios. In research applications and clinically, where serial measurements are being performed on individual patients, quantitative or semiquantitative assessment of the extent and degree of tracer uptake abnormality increases the objectivity of study interpretation and reduces interobserver variability. Image analysis may also be improved by automated thresholding of voxel intensities to avoid manual region of interest placement, which is highly observer dependent.
Received 11 May 1996 Accepted 8 April 1997 Correspondence to: G. A. Donnan,Tel: + 613 9496 5529; Fax: +613 9457 2654
The few volumetric analyses which have been developed employed manual region of interest placement and have not been validated, to our knowledge. 1,2 In ischaemic stroke a measure of both volume and severity of tracer reduction is desirable, as current studies are targeted at evaluating perfusion change after therapeutic intervention. 5,6 The aims of this study were to develop a semiautomated analysis incorporating measurements of both the volume of tracer uptake abnormality as well as the degree, to validate this technique using a phantom stroke model and to evaluate its use in the clinical setting. MATERIALS
AND METHODS
Development of the ischaemic 'index analysis Using standard gamma camera software a new method was developed for measuring the extent and severity of tracer uptake reduction or increase on SPECT images, by comparing the count rate in the affected region with the count rate in the homologous region in the unaffected hemisphere. Region of interest assignment. The analysis was semiautomated in that it only required the operator to specify a region of interest (ROI) encompassing the stroke focus on one image and to state the number of transaxial images over which the abnormality extended. In previous studies 1,2 affected voxels were identified by drawing around the voxels on each transaxial SPECT image on which there was an abnormality. Thresholding. Extraneous scalp, background and white matter activity were excluded by thresholding voxel intensities. Two thresholds were used: 1.
A threshold to segment extraneous background activity such as from the scalp and subcortical low flow areas. A threshold level of 50% of the maximal count in the study allowed only cortical and subcortical grey matter activity to be analysed. 121
122
2.
Baird et aL
A second threshold value of 15% to locate the abnormal voxels in the affected grey matter structures. This value was obtained from studies of the side-to-side asymmetry in 66 healthy volunteers in our institution where a normal side-toside asymmetry of +14.5% was found in volunteers age matched to our stroke population. 9
Ischaemic index analysis. The ischaemic index analysis was performed on reconstructed one pixel (3.1 ram) thick transaxial images; pixel by pixel analysis was performed over a selected volume. The midline symmetry axis (sagittal) across which regions of interest were mirrored was defined manually. The number of transaxial slices over which the perfusion defect extended were defined. The transaxial slice with the largest perfusion defect was identified and an elliptical ROI was placed generously over the defect so that it overlapped the perfusion defect and surrounding grey matter. The ROI was mirrored onto the homologous region in the unaffected hemisphere and was used as the 'reference ROI in the unaffected hemisphere' (Fig. 1). In order to avoid replacing the ROI on each slice, the same ROI was automatically used as the reference ROI on the remaining transaxial slices over which the perfusion defect extended. To segment grey matter activity from 'scalp' and low count subcortical activity the cut off value of 50% of the maximal count in the study was used on the unaffected cerebral hemisphere, segmenting the white matter in the process in addition to the external activity from the scalp (Fig. 1). On each slice the count rate was compared between homologous voxels in the grey matter in the affected and unaffected hemispheres; a count difference of more than 15% was defined as abnormal. Each abnormal voxel was weighted in proportion to the difference in count rate using the formula: 1 Ischaemic Index (cm 3) =
~]i Ci[n]C i -[n]C i [a] AV
Fig. 1 Ischaemic index analysis of a region of right middle cerebral artery territory hypoperfusion (R) in a patient suffering from ischaemic stroke (SPECT study using the perfusion tracer 99mTc-hexamethylpropyleneamine oxime, image on the left side of the figure). For the ischaemic index analysis (right image), an elliptical region of interest was placed over the transaxial slice on which the perfusion defect appeared largest, and was mirrored onto the homologous region in the unaffected cerebral hemisphere (see text for methodology).
Table I
Dimensions of the phantom strokes
Total volume (cm 3) (a + c = v) 93.5 55.2 42.0 32.4
Internal stroke volume (cm 3) (a)
External dJameter (mm)
Internal diameter (mm)
77.7 46.3 31.8 23.1
40 40 30 30
36 36 26 26
Dimensions of strokes used in the phantom experiments, a = internal volume of perspex container; a + c = v = total volume of perspex container; c = volume of perspex.
The accuracy of the ischaemic index analysis was tested on a series of phantom strokes varying in size and degree of tracer uptake reduction relative to the unaffected cerebral hemisphere, which were scanned, their ischaemic index values measured, and the true and measured values compared.
strokes of varied sizes that have been seen in previous SPECT studies (Table 1). 3,1°,11 The containers were filled with varying concentrations of 99rnyc relative to the grey matter compartment in the brain to simulate the varying reductions in blood flow that may occur in human ischaemic stroke (Table 2). 3,1°m Tracer activity in the phantom grey matter and strokes. The 99mTc activity in the phantom grey matter and strokes (Table 2) is described in relative terms (ratio of stroke 99mTc activity to grey matter 99myc activity). A solution of approximately 320 mBq 99myc in 3000 ml of water was used to fill the grey matter compartments and diluted to the required concentration to fill the strokes. Calculation of true ischaemic index values for the phantom strokes. The true ischaemic index for each phantom stroke was calculated by multiplying the volume of the stroke by the relative 99myc activity; a two-compartmental analysis was used and the ischaemic index was calculated relative to a hypothetical zero perfusion defect, 1 described in detail in the Appendix.
Phantom model The phantom stroke model consisted of a brain phantom in which strokes varying in volume and 99mTc activity relative to the rest of the brain were placed. The Hoffman Brain Phantom container (Data Spectrum Corporation Deluxe Model 5000, Hillsborough, NC, USA) was used to case 21 identical 6 mm thick perspex slices (cut out by laser by Industrial Form Plastics, Springvale, Melbourne, Australia) which were stacked together and held in position by the Hoffman screws. Two hollow grey matter compartments filled with 99mTc were separated by a large central perspex white matter compartment. The internal volume of the grey matter compartments was 1750 ml. Four perspex cylindrical containers ranging in volume from 32.4 to 93.5 cm 3 were used to simulate
Scanning procedures Twenty-eight phantom strokes of varied volume and tracer activity were scanned (Table 2). The phantom stroke was placed in the right hemisphere compartment of the brain phantom and held in position by a small piece of sponge which absorbed the activity contained in the surrounding fluid. The phantom was scanned using a short acquisition (5 s/frame yielding about 35K per frame) to simulate the count rates obtained in the clinical setting, followed by a long acquisition (30 s/frame, yielding about 200 K per frame), to allow systematic effects to be studied in the presence of reduced interference from statistical effects. The SPECT data was acquired on a large field-of-view rotating gamma camera (Starcam 400AC, General Electric Medical
where
C i [n] C i [a] AV
= counts in voxel in unaffected hemisphere = counts in voxel in affected hemisphere = voxel size = (0.31) 3 cm 3
As seen from the above formula, a normal study would have an ischaemic index value of zero and the ischaemic index value of a study of maximal tracer reduction would approach the physical volume. Validation
studies
Journal of Clinical Neuroscience (1999) 6(2), 121-127
© 1999 Harcourt Brace & Co. Ltd
Semiautomated analysis of brain SPECT images 123 Table 2 Relative activity concentrations in the phantom strokes Relative activity of [99mTc]in stroke = [a] (relative to [~gmTc]in grey matter compartment of brain phantom) (%) 0.000 0.052 0.104 0.202 0.245 0.312 0,501 0.701 0.791
(0) (5.2) (10.4) (20.2) (24.5) (31.2) (50.1) (70.1) (79.1)
Relative 99mTcactivities in Strokes relative to the brain (grey matter compartments) studied in the phantom experiments [a]-relative activity in stroke (see Appendix).
Systems Milwaukee, Wisconsin) which was equipped with a low energy, high resolution, parallel hole collimator. Sixty-four planar images were acquired over 360 o in an elliptical orbit. The image matrix used was a 128 x 128 word dataset. A dual-energy dataset was acquired for scatter correction, lz This protocol is used for routine studies in our institution. Image resolution measured in a resolution phantom was 12 mm at full width half maximum (FWHM).
Data processing and measurement of the phantom stroke ischaemic index values The SPECT datasets were preprocessed using scatter correction (a modified low window subtraction method) and decay correction which were employed to improve both quantitative accuracy and image contrast. 12 Butterworth prefiltering was used and the filter parameters of critical frequency and power factor were calculated based on the maximal count in the study, la Edge detection was performed manually and attenuation correction was performed using a postreconstruction method. 13 One pixel thick transaxial slices were reconstructed on a 64 x 64 matrix with a pixel size of 3.1 mm and the ischaemic index values measured using the analysis. Variation in ischaemic index measurements resulting from operator dependent factors The accuracy of the ischaemic index analysis was tested in semiautomated operator simulated studies in which the reference region of interest was displaced by +1 pixel on the extreme vertices defining the region of interest (Fig. 2). The ischaemic index value for each phantom stroke was measured after each of the eighteen displacements, using a close fitting or small reference region of interest and a large reference region of interest. The larger reference region of interest was obtained by moving the top corner back by 3 pixels in the x, y direction and the bottom corner forward by 3 pixels in the x, y direction to give an increase in area of 84%. The variation in ischaemic index measurements was used to determine the accuracy of the analysis.
Statistical analyses Linear regression analysis was used to correlate the true and measured ischaemic index values. The gradient (slope of the regression line), gradient offset and bias were determined so that the measured ischaemic index values could be adjusted to the true values, if required. Analysis of variance was used to calculate the percentage error (repeatability) of the method. The percentage error was calculated using the formula:
© 1999 Harcourt Brace & Co. Ltd
Percentage error = ~ grand mean
x 100 1
Levine's test 14 was used to examine the hypotheses: that the variance was equal for large and small reference regions of interest; that the variance was equal for the 5 and 30 s acquisitions; and that the variance was equal for the two small volumes and the two large volumes. A P value below 0.05 was considered statistically significant.
Clinical studies Between 1992 and 1994 30 patients were studied with SPECT before and after thrombolytic administration during the pilot phase of the Australian Streptokinase Trial (ASK). 4,~5-17 The details of recruitment are described elsewhere. 4 750 MBq of 99mZc HMPAO was administered intravenously prior to thrombolytic administration and scanning was performed when clinically convenient, within 3 h. Acquisition parameters were described above. Repeat 99mTc-HMPAO administration and scanning were performed 24 h after the time of the initial study. The volume of hypoperfusion was measured with the ischaemic index analysis and from the serial measurements, the percentage reperfusion was calculated for each patient ([volume day 1-volume day 2] / volume day 1) x 100 and the change in perfusion volume (volume day 1-volume day 2). Mann-Whitney U-tests were used for statistical analyses.
RESULTS Correlation b e t w e e n m e a s u r e d and true i s c h a e m i c index v a l u e s The measured ischaemic index values showed a close correlation with the actual values but consistently underestimated the value (Fig. 3). The regression coefficient was r=0.994 (P < 0.001). The regression equation was y=l.109 x + 5.304. As the bias of the regression equation increased linearly with the true ischaemic index values, the gradient value of 1.109 and the offset value of 5.304 may be used to adjust to the true ischaemic index values using the linear regression model. The overall percentage error of the analysis was +3.4%.
Effect of region of interest size on the a c c u r a c y of the i s c h a e m i c index m e a s u r e m e n t s The ischaemic index measurements performed using the large reference ROI were significantly more accurate than when a small reference ROI was used. The percentage error of the large ROI measurements was 2.7% compared to 4.0% for the small ROI (P < 0.01). Therefore generous ROI sizing is more accurate in the clinical setting.
A c c u r a c y of the i s c h a e m i c index analysis u n d e r clinical conditions The reliability of ischaemic index measurements under clinical conditions (low count conditions) was assessed by comparing the regression coefficients and errors of the short and long acquisition studies. The regression coefficient was slightly lower for the low count study (r=-0.994, P < 0.001) than the high count study (r=0.995, P < 0.001) but was highly significant. The error of the ischaemic index measurements was higher for the low count studies (+4.4%) than for the high count studies (+2.1%) but not significantly (P > 0.05). These results indicate that
Journal of Clinical Neuroscience (1999) 6(2), 121-127
124
Baird et al.
Corner 1 origin
C.
(-1, I) ~
,'
Cj~
Elliptical Region
RO,~..,~ ~
(-1, o)
C.
Ca
(0, 1 )
(1,1)
(o,Cilo)
Ci2 . ~ ( 1 , o)
_°flnteres~ (-1, -1
Ci7
Corner 2 odgln
("0,-1)
Cts
_1
~ (1, -1)
C~9
Corner i (i = 1, 2)
In the operator simulated studies, the two corner origin positions of the reference region were displaced to one of 9 positions as illustrated in Figure 4B, thereby giving a total of 18 corner positions defining the R e g i o n of Interest. R O I a to ROIc illustrates the extent of the regions resulting from the variation in the c o m e r l ' s positioning.
For each c o m e r position, 8 additional corner positions were selected. This allows an effective error of corner placem e n t by the operator of + / - 1 pixel in the horizontal and vertical directions.
Fig, 2 Operatorsimulated region of interest displacement. ROI = region of interest. the analysis is valid for use in routine patient acquisitions but that an error of up to +4.4% may exist. A c c u r a c y of the i s c h a e m i c index analysis for small stroke v o l u m e s The reliability of the ischaemic index analysis for small stroke volumes was tested by comparing the errors of the measured and true values between the two larger stroke volumes and the two smaller stroke volumes. The error in ischaemic index measurements was lower for the smaller strokes (+3.0%) than for the two larger strokes (+3.2%) but not significantly (P > 0.05). U s e in the clinical setting In 30 patients studied during the pilot phase of the Australian Streptokinase Trial, the ischaemic index analysis was used to Journal of Cfinical Neuroscience (1999) 6(2), 121-127
measure hypoperfusion volumes on SPECT images. In all cases there was a perfusion defect which could be measured with the ischaemic index analysis. The initial hypoperfusion volume was larger in the streptokinase treated group (Table 3), but not significantly. A higher mean change in ischaemic index volume was seen in the streptokinase treated group, but the overall percentage reperfusion values were not significantly different (Table 3, Fig. 4). DISCUSSION In this study a semiautomated volumetric analysis was developed and validated in studies which demonstrated that it was accurate (to +4.4%) under clinical conditions. A semiautomated program such as this is valuable for reducing observer error and bias. In conditions such as ischaemic stroke, where hypoperfusion is more common than hyperperfusion, objective region of interest placement is particularly desirable because manual ROI placement © 1999 Harcourt Brace & Co. Ltd
Semiautomated analysis of brain SPECT images 125
100 Regression equation: y = aX + b
80
a = 1.1019
(StandardDeviation=
0.0174) p < 0.001
b = 5.304
(StandardDeviation=
0.705) p < 0.001
B
60 8
t~
40
20
0
20
40 60 True IschaemicIndex (cra~)
80
100
Fig. 3 A close correlation was seen between the measured and true ischaemic index values, although the true values were underestimated by the analysis. The measured ischaemic index value for each of the 18 measurements has been charted when the large reference region of interest was used in the analysis. The overall error of the analysis was _+3.4%.
around a region of low flow is much more difficult than in a region where there is increased tracer accumulation. 18-19 The ischaemic index analysis was used in the clinical setting of therapeutic evaluation in stroke to provide measurements of the initial volume of hypoperfused tissue, and reperfusion indices in treatment and control groups. A correction factor may be applied to measured ischaemic index values to correct to the true values if required. This may be important in multicentre studies 4-5 where correction values determined for each institution could allow ischaemic index values to be directly compared and correlated. Important features of the study included the construction of a brain model which was the same size as the human head, important for attenuation correction, and which incorporated both grey and white matter compartments as the analysis involved white matter segmentation. Acquisition and reconstruction parameters were the Table 3 Trial
same as used in clinical studies and the program was accurate and can be reliably applied on strokes as small as 23.4 ml in volume. The analysis is designed for use with currently used brain perfusion radiopharmaceuticals such as 99mTc-HMPAO and 99mTc-bicisate. One limitation of the analysis may be the exclusion of white matter activity. Many programs analyse cortical activity alone because the white matter is so poorly resolved on SPECT. 6,z° During development of the ischaemic index analysis, included activity from the white matter/ventricular compartment/scalp, usually seen as a strip of pixels along the grey matter/white matter or grey matter/scalp interface led to significant overestimation or underestimation in the extent of the stroke being analysed. The introduction of a lower cut off value (segmenting the scalp, white matter and ventricles) gave the best approximation of stroke defect size and significantly reduced these errors. Several specific problems pertain to the quantitation of strokes. The presence of transhemispheric diaschisis 21 means that perfusion defect size may be underestimated due to underperfusion in the reference region of interest. The frequency and time course of transhemispheric diaschisis is not known and it must be acknowledged as a confounding factor. The program must be applied with caution if there is a stroke in the homologous region in the opposite hemisphere or if there is bilateral flow disturbance such as bilateral carotid artery disease again potentially leading to underestimation of the perfusion defect size. In addition, the uptake and retention of radiopharmaceuticals in diseased tissue may not always be directly related to tracer delivery to the tissue. 22 If hyperperfusion is present, as may occur during the acute and subacute phases of ischaemic stroke, a negative ischaemic index value will be found. 4 If there are both areas of hypoperfusion and hyperperfusion in the same study, these could be analysed separately. The ischaemic index provides a semiquantitative measurement of the extent and severity of perfusion defects on SPECT images. True quantitative SPECT measurements are only possible with 133xenon SPECT 23 and dynamic systems with arterial sampling and octanol partitioning. 24 Using a generous region of interest in the analysis minimised the operator dependent variance of measurements to +2.7%. The ischaemie index analysis would be expected to become less accurate as stroke size falls below 2 FWHM diameter (2.4 cm in our system). 2° Future applications of the analysis include the division of results by volume and severity as well as the combined ischaemic index measurement, the full automation of the analysis and further validation studies using the Hoffman brain phantom slices. This analysis has the advantages of semiautomation, ease of use and validation and has a potentially wide range of applications for both SPECT and PET.
Perfusion change in 30 patients studied during the pilot phase of the Australian Streptokinase
Number of patients Time of 99mTc-HMPAO administration (h, mean+SEM) Admission ischaemic index volume (cm 3, mean_+SEM) Reperfusion Percent reperfusion (mean-+SEM) Change in volume (cm 3, mean-+SEM)
Streptokinase
Control
P
15 5.9+1.0
15 6.4_+1.0
0.73
78.0_+12.2
45.0-+14.0
0.08
41.5_+12.8 32.9_+10.0
45.3+12.8 12.2-+6.1
45.3+_11.6 0.8 0.08
Thirty patients underwent serial 99mTc-HMPAO SPECT studies during the pilot phase of the Australian Streptokinase Trial. The hypoperfusion volume on the SPECT images was measured with the ischaemic index analysis. The initial hypoperfusion volume was larger in the streptokinase treated group, but not significantly. A higher mean change in ischaemic index volume was seen in the streptokinase treated group, but the overall percentage reperfusion values were not significantly different.
© 1999 Harcourt Brace & Co. Ltd
Joumal of Clinical Neuroscience (1999) 6(2), 121-127
126
Baird et aL
Fig. 4 HMPAO SPECT studies obtained at 2.8 and 27 h post onset of a right hemisphere syndrome in a 62-year-old male treated with intravenous thrombolysis at 3 h. On the first HMPAO SPECT study hypoperfusion is present over the right middle cerebral artery territory (volume 67.7 cm 3) while at 27 h there has been a return of blood flow apart from persisting perfusion defect in the lenticulostriate territory (volume 9 cma, 90% reperfusion). Resolution of all symptoms occurred between 4-8 h.
ACKNOWLEDGEMENTS Dr Baird is supported by the National Health and Medical Research Council of Australia. The staff of the Neurology and Nuclear Medicine Departments at the Austin Hospital are gratefully acknowledged. Ms M Prince and Dr R McCloud of the Department of Statistics, Monash University, Clayton, Victoria, Australia are thanked for their assistance with the statistical analyses. Dr R. Hicks and Dr A. Scott are thanked for their reviews of the manuscript.
REFERENCES l.
Mountz JM, Modell JG, Foster NL, DuPree ES, Ackermann RJ, Petry NA, Bluemlein LE, KuhI DE. Prognostication of recovery following stroke using the comparison of CT and Tc-99m HM-PAO SPECT J Nucl Med 1990; 31: 61~56. 2. Davis SM, Chua M, Lichtenstein M, Rossiter SC, Binns D, Hopper JL. Cerebral hypoperfusion in stroke prognosis and brain recovery. Stroke 1993; 24: 1691-1696. 3. Limburg M, van Royen EA, Hijdra A, Verbeeten B Jr. Single-photon emission computed tomography and early death in acute ischemic stroke. Stroke 1990; 21: 1150-i155. 4. Baird AE, Donnan GA, Austin MC, Fitt GJ, Davis SM, McKay WJ. Reperfusion after thrombolytic therapy in ischemic stroke measured by single photon emission computed tomography. Stroke 1994; 25: 79-85.
Journal of Clinical Neuroscience (1999) 6(2), 121-127
5. Infeld B, Davis SM, Donnan GA et al. Streptokinase increases luxury perfusion after stroke. Stroke I996; 27: 1524-1529. 6. Hanson SK, Gmtta JC, Rhoades H, Tran HD, Lamki LM, Bah'on B J, Taylor WJ. Value of single-photon emission-computed tomography in acute stroke therapeutic trials. Stroke 1993; 24: 1322-1329. 7. Andersen AR, Friberg H, Knudsen KBM, Barry DI, Paulson OB, Schmidt JF, et at. Extraction of [99mTc]-&/-HM-PAO across the blood brain barrier. J Cereb Blood Flow Metab 1998; 8 (Suppl 1): $44-$51. 8. Sharp PF, Smith FW, Gemmell HG et al. Technetium-99m HMPAO stereoisomers as potential agents for imaging regional cerebral blood flow: human volunteer studies. J Nucl Med 1986; 27: 17I 177. 9. Baird AE, Donnan GA, Austin MC, Hennessy OF, Royle J, Mc Kay WJ. Asymmetries of cerebral perfusion in a stroke-age population. J Clin Neurosci (in press). 10. Baird AE, Donnan GA, Austin MC, McKay WI. Reperfusion in the 'spectacular shrinking deficit' measured by single photon emission computed tomography. Neurology 1995; 45: 1335-1339. 11. Baird AE, Donnan GA. Prognostic value of cerebral perfusion measurements during the first 48 hours of ischaemic stroke. Lancet 1993; 342: 236. 12. Newton MR, Austin MC, Chan JG, McKay WJ, Rowe CC, Berkovic SE Ictal SPECT using technetium-99m-HMPAO: methods for rapid preparation and optimal deployment of tracer during spontaneous seizures. J Nucl Med 1993; 34: 666-670. 13. Chang LT. A method for attenuation correction in radionuclide computed tomography. IEEE Trans Nucl Sci 1978; NS-25: 638-643. i4. Milliken GA, Johnson DE. The Analysis of Messy Data Volume 1. Belmont CA: Lifetime Learning Publications; 1984. 15. Donnan GA, Davis SM, Chambers BR et al. Australian Streptokinase Trial (ASK). In: dei Zoppo GJ, Moil E, Hacke W, (eds). Thrombolytic Therapy in Acute Ischaemic Stroke II. Bedim Germany: Springer-Verlag, i993; 80-85. 16. Fitt GJ, Farrar JJ, Baird AE et al. Pilot study of intra-arterial streptokinase in acute ischaemic stroke. Med J Aust 1993; 159: 331-334. 17. Donnan GA, Davis SM, Chambers BR et al. Streptokinase for acute ischemic stroke with relationship to time of administration. JAMA 1996; 276: 961-966. 18. The Institute of Physical Sciences in Medicine. An Introduction to Emission Computed Tomography. London: 1985; Report No. 44. 19. Jaszczak RJ, Coleman RE, Whitehead FR. Physical factors affecting quantitative measurements using camera-based single photon emission computed tomography (SPECT). IEEE Trans Nucl Sci 1981; NS-28: 69-80. 20. Mazziotta JC, Phelps ME, Plummer D, Kuhl DE. Quantitation in a positron emission computed tomography: 5. Physical - anatomical effects. J Comput Assist Tomogr 198i; 5: 734-743. 21. Feeney DM, Baron JC. Diaschisis. Stroke 1986; 17: 817-830. 22. Sperling B, Lassen NA. Hyperfixation of HMPAO during the subacute phase of stroke leading to spuriously high estimates of cerebral blood flow. Stroke 1993; 24: 193-194. 23. Murase K, Tanada S, Fujita H, Sakaki S, Hamamoto K. Kinetic behavoir of technetium-99m-HMPAO in the human brain and quantification of cerebral blood flow using dynamic SPECT. J Nucl Med 1992; 33: 135-143. 24. Andersen AR, Friberg H, Schmidt JF, Hasselbalch SG. Quantitative measurements of cerebral blood flow using SPECT and Tc99m-D, L-HMPAO compared to xenon-133. J Cereb Blood Flow Metab 1988; 8 (suppl 1): $69-S81.
© 1999 Harcourt Brace & Co. Ltd
Semiautomated analysis of brain SPECT images
where a + c =
APPENDIX Calculation of the true ischaemic index values of the phantom studies For each phantom stroke studied the calculation of the true ischaemic index value was based on the total volume of the stroke and the relative [99mTc] contained in the stroke. Assuming uniform distribution of 99mZc throughout the ROI and a hypothetical zero perfusion defect 1 the formula would have been: Ischaemie
a
=
C
=
[a]
=
[c]
:
127
V total stroke volume (perspex bottle) internal volume of perspex bottle volume of perspex (total volume-internal volume) relative [99mTc] in compartment a relative [99mWc] in compartment c 0
To calculate the true ischaemic index value
I n d e x [ROll = v (1 - [v])
where v = volume of ROI where Iv] = relative [99mTc] in ROI
True ischaemic
i n d e x - a [a] + c [c] a-I-e
Where [c] = 0, As the perspex container accounted for up to 25-30% of the total volume of the stroke (Table 1) and had zero activity when the stroke was filled with 99roTe, it was necessary to calculate the true ischaemic index values using a two compartment model (Appendix figure). The stroke volume v was divided into compartments a and e. The volume of compartment a was equivalent to the internal volume of v. Compartment e was the volume of perspex and accounted for up to 25-30% of the total volume. Compartment a was filled with varying activities of 99mTc relative to the brain phantom (Table 2, Appendix Figure) while compartment e had zero 99mTc activity.
True isehaemie
i n d e x - a [a] a-l-e
The ischaemic index in compartment a was equal to the volume of compartment a multiplied by the relative 99mTc activity in that compartment; because the ischaemic index was calculated relative to a hypothetical zero perfusion defect the ischaemic index in compartment a was calculated by multiplying the volume of compartment a by (1 - [a]). The ischaemic index of compartment e was equal to the volume of the compartment because no 99mTc could enter this compartment. Because the Ischaemic Index was calculated relative to a hypothetical zero perfusion defect: Isehaemie
i n d e x = a (1 - [a]) + e (1 - [ e l )
Where [c] = 0, Isehaemic
Appendixfigure Two compartment model used to calculate the true ischaemicindexvalues for the phantom strokes and activity concentrations in compartments a and c.
© 1999 Harcourt Brace & Co. Ltd
i n d e x = a (1 - [ a l ) + e
When water (zero activity) was added to the ROI, the predicted ischaemic index was equal to the total volume of the ROI. When [99mTc] was added to the ROI, the ischaemic index was calculated using the relative [99mTc] concentration contained in compartment a added to the total volume of compartment c (denoted as zero activity).
Journal of Cfinical Neuroscience (1999) 6(2), 121-127
Clinical studies
S e m i a u t o m a t e d analysis of the e x t e n t and s e v e r i t y of perfusion d e f e c t s on brain SPECT images: validation studies Alison E. Baird 1,3 FRACP, PHD, Mark C. Austin 2 DIP APPL SCI, Graeme J. O'Keefe 2 PHD, W. John McKay = FRACP, Geoffrey A. Donnan 1,3 MD FRACP 1Departmentof Neurology 2Departmentof NuclearMedicine,Austin and RepatriationMedicalCentre,Victoria,Australia 3Universityof Melbourne,Victoria,Australia
Summary Single photon emission computed tomography (SPECT) is a widely available and practical functional imaging technique with established clinical and research applications in neurological disorders such as epilepsy and stroke. SPECT images are usually analysed visually, or semiquantitativety by measuring side-to-side asymmetries. In order to evaluate perfusion change after thrombolytic therapy in patients with acute ischaemic stroke we developed a semiautomated, weighted volumetric analysis (the ischaemic index) that measured the extent and severity of tracer uptake abnormalities on brain SPECT images semiquantitatively. The unaffected cerebral hemisphere was used as the reference region of interest. The analysis was validated in a phantom brain model incorporating 'strokes' varying in size and degree. The phantom 'stroke' sizes measured with the ischaemic index analysis correlated closely with the true values (r=0.994, P < 0.01) and this correlation was maintained under low count conditions acquired to simulate the clinical setting. The overall operator dependent error of the analysis was _+3.4%. In 30 patients treated with thrombolytic therapy who were studied serially with 99rnTc_hexamethylpropyleneamine oxime (HMPAO) SPECT, the analysis was used to measure hypoperfusion volume and provide indices of perfusion change. This analysis has the advantages of semiautomation, ease of use and validation and has a potentially wide range of applications for both SPECT and PET. Keywords: SPECT, emission computed tomography, quantitification, volume quantitation
INTRODUCTION Brain single photon emission computed tomography (SPECT) is the only tomographic functional imaging technique currently accessible to the majority of physicians, and is attracting increasing interest among neurologists and neurosurgeons because of its current or potential applications in areas such as epilepsy, stroke, dementia and cerebral neoplasia. In patients with ischaemic stroke the measurement of hypoperfusion with SPECT has been shown to be sensitive in cortical lesions and may provide prognostic information. ~-3 Large severe reductions in perfusion may correlate with a high mortality rate and with poor prognosis. ~-3 SPECT has also attracted interest for its potential use in the evaluation and selection of patients for stroke therapies .4-6 Scanning can be delayed until clinically convenient as radiopharmaceuticals such as 99mTc-hexamethylpropyleneamine oxime (HMPAO) are trapped in the brain within 2 min of i.v. administration and show minimal redistribution. 7-8 In most SPECT studies images have been interpreted visually or by the measurement of side-to-side or lesion-to-cerebellar ratios. In research applications and clinically, where serial measurements are being performed on individual patients, quantitative or semiquantitative assessment of the extent and degree of tracer uptake abnormality increases the objectivity of study interpretation and reduces interobserver variability. Image analysis may also be improved by automated thresholding of voxel intensities to avoid manual region of interest placement, which is highly observer dependent.
Received 11 May 1996 Accepted 8 April 1997 Correspondence to: G. A. Donnan,Tel: + 613 9496 5529; Fax: +613 9457 2654
The few volumetric analyses which have been developed employed manual region of interest placement and have not been validated, to our knowledge. 1,2 In ischaemic stroke a measure of both volume and severity of tracer reduction is desirable, as current studies are targeted at evaluating perfusion change after therapeutic intervention. 5,6 The aims of this study were to develop a semiautomated analysis incorporating measurements of both the volume of tracer uptake abnormality as well as the degree, to validate this technique using a phantom stroke model and to evaluate its use in the clinical setting. MATERIALS
AND METHODS
Development of the ischaemic 'index analysis Using standard gamma camera software a new method was developed for measuring the extent and severity of tracer uptake reduction or increase on SPECT images, by comparing the count rate in the affected region with the count rate in the homologous region in the unaffected hemisphere. Region of interest assignment. The analysis was semiautomated in that it only required the operator to specify a region of interest (ROI) encompassing the stroke focus on one image and to state the number of transaxial images over which the abnormality extended. In previous studies 1,2 affected voxels were identified by drawing around the voxels on each transaxial SPECT image on which there was an abnormality. Thresholding. Extraneous scalp, background and white matter activity were excluded by thresholding voxel intensities. Two thresholds were used: 1.
A threshold to segment extraneous background activity such as from the scalp and subcortical low flow areas. A threshold level of 50% of the maximal count in the study allowed only cortical and subcortical grey matter activity to be analysed. 121
122
2.
Baird et aL
A second threshold value of 15% to locate the abnormal voxels in the affected grey matter structures. This value was obtained from studies of the side-to-side asymmetry in 66 healthy volunteers in our institution where a normal side-toside asymmetry of +14.5% was found in volunteers age matched to our stroke population. 9
Ischaemic index analysis. The ischaemic index analysis was performed on reconstructed one pixel (3.1 ram) thick transaxial images; pixel by pixel analysis was performed over a selected volume. The midline symmetry axis (sagittal) across which regions of interest were mirrored was defined manually. The number of transaxial slices over which the perfusion defect extended were defined. The transaxial slice with the largest perfusion defect was identified and an elliptical ROI was placed generously over the defect so that it overlapped the perfusion defect and surrounding grey matter. The ROI was mirrored onto the homologous region in the unaffected hemisphere and was used as the 'reference ROI in the unaffected hemisphere' (Fig. 1). In order to avoid replacing the ROI on each slice, the same ROI was automatically used as the reference ROI on the remaining transaxial slices over which the perfusion defect extended. To segment grey matter activity from 'scalp' and low count subcortical activity the cut off value of 50% of the maximal count in the study was used on the unaffected cerebral hemisphere, segmenting the white matter in the process in addition to the external activity from the scalp (Fig. 1). On each slice the count rate was compared between homologous voxels in the grey matter in the affected and unaffected hemispheres; a count difference of more than 15% was defined as abnormal. Each abnormal voxel was weighted in proportion to the difference in count rate using the formula: 1 Ischaemic Index (cm 3) =
~]i Ci[n]C i -[n]C i [a] AV
Fig. 1 Ischaemic index analysis of a region of right middle cerebral artery territory hypoperfusion (R) in a patient suffering from ischaemic stroke (SPECT study using the perfusion tracer 99mTc-hexamethylpropyleneamine oxime, image on the left side of the figure). For the ischaemic index analysis (right image), an elliptical region of interest was placed over the transaxial slice on which the perfusion defect appeared largest, and was mirrored onto the homologous region in the unaffected cerebral hemisphere (see text for methodology).
Table I
Dimensions of the phantom strokes
Total volume (cm 3) (a + c = v) 93.5 55.2 42.0 32.4
Internal stroke volume (cm 3) (a)
External dJameter (mm)
Internal diameter (mm)
77.7 46.3 31.8 23.1
40 40 30 30
36 36 26 26
Dimensions of strokes used in the phantom experiments, a = internal volume of perspex container; a + c = v = total volume of perspex container; c = volume of perspex.
The accuracy of the ischaemic index analysis was tested on a series of phantom strokes varying in size and degree of tracer uptake reduction relative to the unaffected cerebral hemisphere, which were scanned, their ischaemic index values measured, and the true and measured values compared.
strokes of varied sizes that have been seen in previous SPECT studies (Table 1). 3,1°,11 The containers were filled with varying concentrations of 99rnyc relative to the grey matter compartment in the brain to simulate the varying reductions in blood flow that may occur in human ischaemic stroke (Table 2). 3,1°m Tracer activity in the phantom grey matter and strokes. The 99mTc activity in the phantom grey matter and strokes (Table 2) is described in relative terms (ratio of stroke 99mTc activity to grey matter 99myc activity). A solution of approximately 320 mBq 99myc in 3000 ml of water was used to fill the grey matter compartments and diluted to the required concentration to fill the strokes. Calculation of true ischaemic index values for the phantom strokes. The true ischaemic index for each phantom stroke was calculated by multiplying the volume of the stroke by the relative 99myc activity; a two-compartmental analysis was used and the ischaemic index was calculated relative to a hypothetical zero perfusion defect, 1 described in detail in the Appendix.
Phantom model The phantom stroke model consisted of a brain phantom in which strokes varying in volume and 99mTc activity relative to the rest of the brain were placed. The Hoffman Brain Phantom container (Data Spectrum Corporation Deluxe Model 5000, Hillsborough, NC, USA) was used to case 21 identical 6 mm thick perspex slices (cut out by laser by Industrial Form Plastics, Springvale, Melbourne, Australia) which were stacked together and held in position by the Hoffman screws. Two hollow grey matter compartments filled with 99mTc were separated by a large central perspex white matter compartment. The internal volume of the grey matter compartments was 1750 ml. Four perspex cylindrical containers ranging in volume from 32.4 to 93.5 cm 3 were used to simulate
Scanning procedures Twenty-eight phantom strokes of varied volume and tracer activity were scanned (Table 2). The phantom stroke was placed in the right hemisphere compartment of the brain phantom and held in position by a small piece of sponge which absorbed the activity contained in the surrounding fluid. The phantom was scanned using a short acquisition (5 s/frame yielding about 35K per frame) to simulate the count rates obtained in the clinical setting, followed by a long acquisition (30 s/frame, yielding about 200 K per frame), to allow systematic effects to be studied in the presence of reduced interference from statistical effects. The SPECT data was acquired on a large field-of-view rotating gamma camera (Starcam 400AC, General Electric Medical
where
C i [n] C i [a] AV
= counts in voxel in unaffected hemisphere = counts in voxel in affected hemisphere = voxel size = (0.31) 3 cm 3
As seen from the above formula, a normal study would have an ischaemic index value of zero and the ischaemic index value of a study of maximal tracer reduction would approach the physical volume. Validation
studies
Journal of Clinical Neuroscience (1999) 6(2), 121-127
© 1999 Harcourt Brace & Co. Ltd
Semiautomated analysis of brain SPECT images 123 Table 2 Relative activity concentrations in the phantom strokes Relative activity of [99mTc]in stroke = [a] (relative to [~gmTc]in grey matter compartment of brain phantom) (%) 0.000 0.052 0.104 0.202 0.245 0.312 0,501 0.701 0.791
(0) (5.2) (10.4) (20.2) (24.5) (31.2) (50.1) (70.1) (79.1)
Relative 99mTcactivities in Strokes relative to the brain (grey matter compartments) studied in the phantom experiments [a]-relative activity in stroke (see Appendix).
Systems Milwaukee, Wisconsin) which was equipped with a low energy, high resolution, parallel hole collimator. Sixty-four planar images were acquired over 360 o in an elliptical orbit. The image matrix used was a 128 x 128 word dataset. A dual-energy dataset was acquired for scatter correction, lz This protocol is used for routine studies in our institution. Image resolution measured in a resolution phantom was 12 mm at full width half maximum (FWHM).
Data processing and measurement of the phantom stroke ischaemic index values The SPECT datasets were preprocessed using scatter correction (a modified low window subtraction method) and decay correction which were employed to improve both quantitative accuracy and image contrast. 12 Butterworth prefiltering was used and the filter parameters of critical frequency and power factor were calculated based on the maximal count in the study, la Edge detection was performed manually and attenuation correction was performed using a postreconstruction method. 13 One pixel thick transaxial slices were reconstructed on a 64 x 64 matrix with a pixel size of 3.1 mm and the ischaemic index values measured using the analysis. Variation in ischaemic index measurements resulting from operator dependent factors The accuracy of the ischaemic index analysis was tested in semiautomated operator simulated studies in which the reference region of interest was displaced by +1 pixel on the extreme vertices defining the region of interest (Fig. 2). The ischaemic index value for each phantom stroke was measured after each of the eighteen displacements, using a close fitting or small reference region of interest and a large reference region of interest. The larger reference region of interest was obtained by moving the top corner back by 3 pixels in the x, y direction and the bottom corner forward by 3 pixels in the x, y direction to give an increase in area of 84%. The variation in ischaemic index measurements was used to determine the accuracy of the analysis.
Statistical analyses Linear regression analysis was used to correlate the true and measured ischaemic index values. The gradient (slope of the regression line), gradient offset and bias were determined so that the measured ischaemic index values could be adjusted to the true values, if required. Analysis of variance was used to calculate the percentage error (repeatability) of the method. The percentage error was calculated using the formula:
© 1999 Harcourt Brace & Co. Ltd
Percentage error = ~ grand mean
x 100 1
Levine's test 14 was used to examine the hypotheses: that the variance was equal for large and small reference regions of interest; that the variance was equal for the 5 and 30 s acquisitions; and that the variance was equal for the two small volumes and the two large volumes. A P value below 0.05 was considered statistically significant.
Clinical studies Between 1992 and 1994 30 patients were studied with SPECT before and after thrombolytic administration during the pilot phase of the Australian Streptokinase Trial (ASK). 4,~5-17 The details of recruitment are described elsewhere. 4 750 MBq of 99mZc HMPAO was administered intravenously prior to thrombolytic administration and scanning was performed when clinically convenient, within 3 h. Acquisition parameters were described above. Repeat 99mTc-HMPAO administration and scanning were performed 24 h after the time of the initial study. The volume of hypoperfusion was measured with the ischaemic index analysis and from the serial measurements, the percentage reperfusion was calculated for each patient ([volume day 1-volume day 2] / volume day 1) x 100 and the change in perfusion volume (volume day 1-volume day 2). Mann-Whitney U-tests were used for statistical analyses.
RESULTS Correlation b e t w e e n m e a s u r e d and true i s c h a e m i c index v a l u e s The measured ischaemic index values showed a close correlation with the actual values but consistently underestimated the value (Fig. 3). The regression coefficient was r=0.994 (P < 0.001). The regression equation was y=l.109 x + 5.304. As the bias of the regression equation increased linearly with the true ischaemic index values, the gradient value of 1.109 and the offset value of 5.304 may be used to adjust to the true ischaemic index values using the linear regression model. The overall percentage error of the analysis was +3.4%.
Effect of region of interest size on the a c c u r a c y of the i s c h a e m i c index m e a s u r e m e n t s The ischaemic index measurements performed using the large reference ROI were significantly more accurate than when a small reference ROI was used. The percentage error of the large ROI measurements was 2.7% compared to 4.0% for the small ROI (P < 0.01). Therefore generous ROI sizing is more accurate in the clinical setting.
A c c u r a c y of the i s c h a e m i c index analysis u n d e r clinical conditions The reliability of ischaemic index measurements under clinical conditions (low count conditions) was assessed by comparing the regression coefficients and errors of the short and long acquisition studies. The regression coefficient was slightly lower for the low count study (r=-0.994, P < 0.001) than the high count study (r=0.995, P < 0.001) but was highly significant. The error of the ischaemic index measurements was higher for the low count studies (+4.4%) than for the high count studies (+2.1%) but not significantly (P > 0.05). These results indicate that
Journal of Clinical Neuroscience (1999) 6(2), 121-127
124
Baird et al.
Corner 1 origin
C.
(-1, I) ~
,'
Cj~
Elliptical Region
RO,~..,~ ~
(-1, o)
C.
Ca
(0, 1 )
(1,1)
(o,Cilo)
Ci2 . ~ ( 1 , o)
_°flnteres~ (-1, -1
Ci7
Corner 2 odgln
("0,-1)
Cts
_1
~ (1, -1)
C~9
Corner i (i = 1, 2)
In the operator simulated studies, the two corner origin positions of the reference region were displaced to one of 9 positions as illustrated in Figure 4B, thereby giving a total of 18 corner positions defining the R e g i o n of Interest. R O I a to ROIc illustrates the extent of the regions resulting from the variation in the c o m e r l ' s positioning.
For each c o m e r position, 8 additional corner positions were selected. This allows an effective error of corner placem e n t by the operator of + / - 1 pixel in the horizontal and vertical directions.
Fig, 2 Operatorsimulated region of interest displacement. ROI = region of interest. the analysis is valid for use in routine patient acquisitions but that an error of up to +4.4% may exist. A c c u r a c y of the i s c h a e m i c index analysis for small stroke v o l u m e s The reliability of the ischaemic index analysis for small stroke volumes was tested by comparing the errors of the measured and true values between the two larger stroke volumes and the two smaller stroke volumes. The error in ischaemic index measurements was lower for the smaller strokes (+3.0%) than for the two larger strokes (+3.2%) but not significantly (P > 0.05). U s e in the clinical setting In 30 patients studied during the pilot phase of the Australian Streptokinase Trial, the ischaemic index analysis was used to Journal of Cfinical Neuroscience (1999) 6(2), 121-127
measure hypoperfusion volumes on SPECT images. In all cases there was a perfusion defect which could be measured with the ischaemic index analysis. The initial hypoperfusion volume was larger in the streptokinase treated group (Table 3), but not significantly. A higher mean change in ischaemic index volume was seen in the streptokinase treated group, but the overall percentage reperfusion values were not significantly different (Table 3, Fig. 4). DISCUSSION In this study a semiautomated volumetric analysis was developed and validated in studies which demonstrated that it was accurate (to +4.4%) under clinical conditions. A semiautomated program such as this is valuable for reducing observer error and bias. In conditions such as ischaemic stroke, where hypoperfusion is more common than hyperperfusion, objective region of interest placement is particularly desirable because manual ROI placement © 1999 Harcourt Brace & Co. Ltd
Semiautomated analysis of brain SPECT images 125
100 Regression equation: y = aX + b
80
a = 1.1019
(StandardDeviation=
0.0174) p < 0.001
b = 5.304
(StandardDeviation=
0.705) p < 0.001
B
60 8
t~
40
20
0
20
40 60 True IschaemicIndex (cra~)
80
100
Fig. 3 A close correlation was seen between the measured and true ischaemic index values, although the true values were underestimated by the analysis. The measured ischaemic index value for each of the 18 measurements has been charted when the large reference region of interest was used in the analysis. The overall error of the analysis was _+3.4%.
around a region of low flow is much more difficult than in a region where there is increased tracer accumulation. 18-19 The ischaemic index analysis was used in the clinical setting of therapeutic evaluation in stroke to provide measurements of the initial volume of hypoperfused tissue, and reperfusion indices in treatment and control groups. A correction factor may be applied to measured ischaemic index values to correct to the true values if required. This may be important in multicentre studies 4-5 where correction values determined for each institution could allow ischaemic index values to be directly compared and correlated. Important features of the study included the construction of a brain model which was the same size as the human head, important for attenuation correction, and which incorporated both grey and white matter compartments as the analysis involved white matter segmentation. Acquisition and reconstruction parameters were the Table 3 Trial
same as used in clinical studies and the program was accurate and can be reliably applied on strokes as small as 23.4 ml in volume. The analysis is designed for use with currently used brain perfusion radiopharmaceuticals such as 99mTc-HMPAO and 99mTc-bicisate. One limitation of the analysis may be the exclusion of white matter activity. Many programs analyse cortical activity alone because the white matter is so poorly resolved on SPECT. 6,z° During development of the ischaemic index analysis, included activity from the white matter/ventricular compartment/scalp, usually seen as a strip of pixels along the grey matter/white matter or grey matter/scalp interface led to significant overestimation or underestimation in the extent of the stroke being analysed. The introduction of a lower cut off value (segmenting the scalp, white matter and ventricles) gave the best approximation of stroke defect size and significantly reduced these errors. Several specific problems pertain to the quantitation of strokes. The presence of transhemispheric diaschisis 21 means that perfusion defect size may be underestimated due to underperfusion in the reference region of interest. The frequency and time course of transhemispheric diaschisis is not known and it must be acknowledged as a confounding factor. The program must be applied with caution if there is a stroke in the homologous region in the opposite hemisphere or if there is bilateral flow disturbance such as bilateral carotid artery disease again potentially leading to underestimation of the perfusion defect size. In addition, the uptake and retention of radiopharmaceuticals in diseased tissue may not always be directly related to tracer delivery to the tissue. 22 If hyperperfusion is present, as may occur during the acute and subacute phases of ischaemic stroke, a negative ischaemic index value will be found. 4 If there are both areas of hypoperfusion and hyperperfusion in the same study, these could be analysed separately. The ischaemic index provides a semiquantitative measurement of the extent and severity of perfusion defects on SPECT images. True quantitative SPECT measurements are only possible with 133xenon SPECT 23 and dynamic systems with arterial sampling and octanol partitioning. 24 Using a generous region of interest in the analysis minimised the operator dependent variance of measurements to +2.7%. The ischaemie index analysis would be expected to become less accurate as stroke size falls below 2 FWHM diameter (2.4 cm in our system). 2° Future applications of the analysis include the division of results by volume and severity as well as the combined ischaemic index measurement, the full automation of the analysis and further validation studies using the Hoffman brain phantom slices. This analysis has the advantages of semiautomation, ease of use and validation and has a potentially wide range of applications for both SPECT and PET.
Perfusion change in 30 patients studied during the pilot phase of the Australian Streptokinase
Number of patients Time of 99mTc-HMPAO administration (h, mean+SEM) Admission ischaemic index volume (cm 3, mean_+SEM) Reperfusion Percent reperfusion (mean-+SEM) Change in volume (cm 3, mean-+SEM)
Streptokinase
Control
P
15 5.9+1.0
15 6.4_+1.0
0.73
78.0_+12.2
45.0-+14.0
0.08
41.5_+12.8 32.9_+10.0
45.3+12.8 12.2-+6.1
45.3+_11.6 0.8 0.08
Thirty patients underwent serial 99mTc-HMPAO SPECT studies during the pilot phase of the Australian Streptokinase Trial. The hypoperfusion volume on the SPECT images was measured with the ischaemic index analysis. The initial hypoperfusion volume was larger in the streptokinase treated group, but not significantly. A higher mean change in ischaemic index volume was seen in the streptokinase treated group, but the overall percentage reperfusion values were not significantly different.
© 1999 Harcourt Brace & Co. Ltd
Joumal of Clinical Neuroscience (1999) 6(2), 121-127
126
Baird et aL
Fig. 4 HMPAO SPECT studies obtained at 2.8 and 27 h post onset of a right hemisphere syndrome in a 62-year-old male treated with intravenous thrombolysis at 3 h. On the first HMPAO SPECT study hypoperfusion is present over the right middle cerebral artery territory (volume 67.7 cm 3) while at 27 h there has been a return of blood flow apart from persisting perfusion defect in the lenticulostriate territory (volume 9 cma, 90% reperfusion). Resolution of all symptoms occurred between 4-8 h.
ACKNOWLEDGEMENTS Dr Baird is supported by the National Health and Medical Research Council of Australia. The staff of the Neurology and Nuclear Medicine Departments at the Austin Hospital are gratefully acknowledged. Ms M Prince and Dr R McCloud of the Department of Statistics, Monash University, Clayton, Victoria, Australia are thanked for their assistance with the statistical analyses. Dr R. Hicks and Dr A. Scott are thanked for their reviews of the manuscript.
REFERENCES l.
Mountz JM, Modell JG, Foster NL, DuPree ES, Ackermann RJ, Petry NA, Bluemlein LE, KuhI DE. Prognostication of recovery following stroke using the comparison of CT and Tc-99m HM-PAO SPECT J Nucl Med 1990; 31: 61~56. 2. Davis SM, Chua M, Lichtenstein M, Rossiter SC, Binns D, Hopper JL. Cerebral hypoperfusion in stroke prognosis and brain recovery. Stroke 1993; 24: 1691-1696. 3. Limburg M, van Royen EA, Hijdra A, Verbeeten B Jr. Single-photon emission computed tomography and early death in acute ischemic stroke. Stroke 1990; 21: 1150-i155. 4. Baird AE, Donnan GA, Austin MC, Fitt GJ, Davis SM, McKay WJ. Reperfusion after thrombolytic therapy in ischemic stroke measured by single photon emission computed tomography. Stroke 1994; 25: 79-85.
Journal of Clinical Neuroscience (1999) 6(2), 121-127
5. Infeld B, Davis SM, Donnan GA et al. Streptokinase increases luxury perfusion after stroke. Stroke I996; 27: 1524-1529. 6. Hanson SK, Gmtta JC, Rhoades H, Tran HD, Lamki LM, Bah'on B J, Taylor WJ. Value of single-photon emission-computed tomography in acute stroke therapeutic trials. Stroke 1993; 24: 1322-1329. 7. Andersen AR, Friberg H, Knudsen KBM, Barry DI, Paulson OB, Schmidt JF, et at. Extraction of [99mTc]-&/-HM-PAO across the blood brain barrier. J Cereb Blood Flow Metab 1998; 8 (Suppl 1): $44-$51. 8. Sharp PF, Smith FW, Gemmell HG et al. Technetium-99m HMPAO stereoisomers as potential agents for imaging regional cerebral blood flow: human volunteer studies. J Nucl Med 1986; 27: 17I 177. 9. Baird AE, Donnan GA, Austin MC, Hennessy OF, Royle J, Mc Kay WJ. Asymmetries of cerebral perfusion in a stroke-age population. J Clin Neurosci (in press). 10. Baird AE, Donnan GA, Austin MC, McKay WI. Reperfusion in the 'spectacular shrinking deficit' measured by single photon emission computed tomography. Neurology 1995; 45: 1335-1339. 11. Baird AE, Donnan GA. Prognostic value of cerebral perfusion measurements during the first 48 hours of ischaemic stroke. Lancet 1993; 342: 236. 12. Newton MR, Austin MC, Chan JG, McKay WJ, Rowe CC, Berkovic SE Ictal SPECT using technetium-99m-HMPAO: methods for rapid preparation and optimal deployment of tracer during spontaneous seizures. J Nucl Med 1993; 34: 666-670. 13. Chang LT. A method for attenuation correction in radionuclide computed tomography. IEEE Trans Nucl Sci 1978; NS-25: 638-643. i4. Milliken GA, Johnson DE. The Analysis of Messy Data Volume 1. Belmont CA: Lifetime Learning Publications; 1984. 15. Donnan GA, Davis SM, Chambers BR et al. Australian Streptokinase Trial (ASK). In: dei Zoppo GJ, Moil E, Hacke W, (eds). Thrombolytic Therapy in Acute Ischaemic Stroke II. Bedim Germany: Springer-Verlag, i993; 80-85. 16. Fitt GJ, Farrar JJ, Baird AE et al. Pilot study of intra-arterial streptokinase in acute ischaemic stroke. Med J Aust 1993; 159: 331-334. 17. Donnan GA, Davis SM, Chambers BR et al. Streptokinase for acute ischemic stroke with relationship to time of administration. JAMA 1996; 276: 961-966. 18. The Institute of Physical Sciences in Medicine. An Introduction to Emission Computed Tomography. London: 1985; Report No. 44. 19. Jaszczak RJ, Coleman RE, Whitehead FR. Physical factors affecting quantitative measurements using camera-based single photon emission computed tomography (SPECT). IEEE Trans Nucl Sci 1981; NS-28: 69-80. 20. Mazziotta JC, Phelps ME, Plummer D, Kuhl DE. Quantitation in a positron emission computed tomography: 5. Physical - anatomical effects. J Comput Assist Tomogr 198i; 5: 734-743. 21. Feeney DM, Baron JC. Diaschisis. Stroke 1986; 17: 817-830. 22. Sperling B, Lassen NA. Hyperfixation of HMPAO during the subacute phase of stroke leading to spuriously high estimates of cerebral blood flow. Stroke 1993; 24: 193-194. 23. Murase K, Tanada S, Fujita H, Sakaki S, Hamamoto K. Kinetic behavoir of technetium-99m-HMPAO in the human brain and quantification of cerebral blood flow using dynamic SPECT. J Nucl Med 1992; 33: 135-143. 24. Andersen AR, Friberg H, Schmidt JF, Hasselbalch SG. Quantitative measurements of cerebral blood flow using SPECT and Tc99m-D, L-HMPAO compared to xenon-133. J Cereb Blood Flow Metab 1988; 8 (suppl 1): $69-S81.
© 1999 Harcourt Brace & Co. Ltd
Semiautomated analysis of brain SPECT images
where a + c =
APPENDIX Calculation of the true ischaemic index values of the phantom studies For each phantom stroke studied the calculation of the true ischaemic index value was based on the total volume of the stroke and the relative [99mTc] contained in the stroke. Assuming uniform distribution of 99mZc throughout the ROI and a hypothetical zero perfusion defect 1 the formula would have been: Ischaemie
a
=
C
=
[a]
=
[c]
:
127
V total stroke volume (perspex bottle) internal volume of perspex bottle volume of perspex (total volume-internal volume) relative [99mTc] in compartment a relative [99mWc] in compartment c 0
To calculate the true ischaemic index value
I n d e x [ROll = v (1 - [v])
where v = volume of ROI where Iv] = relative [99mTc] in ROI
True ischaemic
i n d e x - a [a] + c [c] a-I-e
Where [c] = 0, As the perspex container accounted for up to 25-30% of the total volume of the stroke (Table 1) and had zero activity when the stroke was filled with 99roTe, it was necessary to calculate the true ischaemic index values using a two compartment model (Appendix figure). The stroke volume v was divided into compartments a and e. The volume of compartment a was equivalent to the internal volume of v. Compartment e was the volume of perspex and accounted for up to 25-30% of the total volume. Compartment a was filled with varying activities of 99mTc relative to the brain phantom (Table 2, Appendix Figure) while compartment e had zero 99mTc activity.
True isehaemie
i n d e x - a [a] a-l-e
The ischaemic index in compartment a was equal to the volume of compartment a multiplied by the relative 99mTc activity in that compartment; because the ischaemic index was calculated relative to a hypothetical zero perfusion defect the ischaemic index in compartment a was calculated by multiplying the volume of compartment a by (1 - [a]). The ischaemic index of compartment e was equal to the volume of the compartment because no 99mTc could enter this compartment. Because the Ischaemic Index was calculated relative to a hypothetical zero perfusion defect: Isehaemie
i n d e x = a (1 - [a]) + e (1 - [ e l )
Where [c] = 0, Isehaemic
Appendixfigure Two compartment model used to calculate the true ischaemicindexvalues for the phantom strokes and activity concentrations in compartments a and c.
© 1999 Harcourt Brace & Co. Ltd
i n d e x = a (1 - [ a l ) + e
When water (zero activity) was added to the ROI, the predicted ischaemic index was equal to the total volume of the ROI. When [99mTc] was added to the ROI, the ischaemic index was calculated using the relative [99mTc] concentration contained in compartment a added to the total volume of compartment c (denoted as zero activity).
Journal of Cfinical Neuroscience (1999) 6(2), 121-127