Influence of Partial Volume on Venous Output and Arterial Input Function

Theoretical Background

All methods quantifying cerebral perfusion from CT measure temporal changes in signal intensity during the first pass of a bolus of an iodinated contrast agent. Changes in signal intensity from the baseline signal intensity (that is, before the arrival of the contrast agent) are linearly related to the concentration of the contrast agent in that voxel. The CBV is then calculated as the ratio of the AUC of the first passage through the tissue to that of the AUC of the first passage through an artery 6,7 with a correction factor, H, to account for the difference in the hematocrit in small versus large vessels: 1.

Calculation of CBV, therefore, requires knowledge of contrast enhancement in an arterial pixel devoid of PVEs. When imaging at the level of the basal ganglia, however, no arterial pixel can be found that is unaffected by PVEs. A pixel in the center of one of the great sinuses (eg, the SSS) is therefore often preferred as reference because the AUC of the first passage of contrast agent through a vein is equal to the AUC of an artery when the blood-brain barrier is intact. Effectively this means that the CBV is inversely related to the AUC of the VOF: 2.

The MTT is calculated by means of deconvolution analysis.6,7 Convolution describes the tissue contrast enhancement curve (C_TISSUE) at time t as a function of the arterial input curve (C_ART), the tissue residue function R(t), and the cerebral blood flow (CBF; blood volume per unit time): 3.

Conversely, knowledge of the tissue and arterial curve allows for the inverse operation, or deconvolution, which results in R(t), the “residue function.” From the residue function, the MTT can be derived.8 In practice, the MTT can be thought of as the difference between the width of the tissue curve and the width of the AIF.

The calculation of the CBF is based on the central volume principle,9-11 which states that the CBF is equal to the ratio of CBV and MTT: 4.

Consequently, the CBF is inversely related to the AUC of the VOF and directly related to the widths of the AIF and the tissue curve.

From equations (2) and (4) it can be seen that an underestimation of the AUC of the VOF as caused by PVEs would translate to an overestimation of the CBV and thus also to an overestimation of the CBF. In this perspective, a section thickness that is thin enough to guarantee an accurate measurement of the AUF_VOF has to be chosen. A thinner section thickness, however, would decrease the SNR, thus degrading image quality and a compromise on the optimal section thickness has to be found. Theoretically, only negligible effects due to PVEs are expected on the width of the AIF.12

Materials and Methods

CTP Acquisition, Reconstruction, and Measurements

The imaging studies were performed on a 16-section spiral CT scanner (Philips Mx8000 LDT).

CTP source data were derived from sequential scans covering a 2.4-cm-thick section selected 4 cm above the sella turcica and angulated parallel to the meato-orbital line to contain the upper parts of the lateral ventricles and the basal ganglia. Forty milliliters of nonionic contrast agent (Iopromide, Ultravist, 300 mg iodine/mL; Schering, Berlin, Germany) was injected into the cubital vein (18-gauge needle) at a rate of 5 mL/s followed by a 40-mL saline flush at a rate of 5 mL/s by using a dual power injector (Stellant Dual CT injector; Medrad Europe, Beek, the Netherlands). The following parameters were used: 90 kVp, 150 mAs, 8 × 3 mm collimation, 512 × 512 matrix, 200 mm field of view, 1-second cycle time, UB filter, and standard resolution. To limit the analysis to the first passage of the contrast agent and thus eliminate recirculation effects, it is common to fit an analytical function through the concentration curves. In this study, a gamma variate function was used to fit the AIF and VOF.

Data were transferred to a Philips workstation for postprocessing. (We used Philips prototype software that was the predecessor of the extended Brilliance Workstation Perfusion software.) CTP scans were reconstructed at, respectively, 3-, 6-, and 12-mm-thick contiguous axial images, thus leading to, respectively, 8, 4, and 2 sections. Thereafter, a motion-correction algorithm was applied to correct for patient motion, and the results were visually inspected to exclude the possibility of severe residual motion artifacts.

The AIF and the VOF were obtained semiautomatically with the user selecting vascular structures of interest and the computer algorithm choosing the optimal AIF and VOF based on the AUC and the time of arrival (only for the AIF). In each patient, the VOF was selected from the same vein for the different section thicknesses. Because the inclusion of structures with high attenuation (eg, bone) would result in a baseline increase and thus an erroneously high AUC, the baseline attenuation as calculated as the mean attenuation value before contrast agent arrival was subtracted for each voxel from the passage curves when calculating the AUC. For all patients, the AUCs of the AIF and VOF and the width of the AIF were obtained for all sections of the 3-, 6-, and 12-mm series by drawing regions of interest in the ACA and one of the major sinuses (SSS, straight sinus, sigmoid sinus). The width of the AIF was defined as the full width at half maximum of the fitted gamma variate function. Finally, circular regions of interest with a standard size of 2.5 cm in diameter were drawn for each section, in the peripheral anterior flow territory at the level of the basal ganglia on a standard location for all patients and each section thickness (Fig 1), containing both white and gray matter. All regions of interest were placed manually by one experienced observer (I.v.d.S.). In all patients, the perfusion in this area was not affected by infarction or blood.

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

A circular region of interest with a standard size of 2.5 cm in diameter was drawn in the peripheral anterior flow territory at the level of the basal ganglia on a standard location containing both white and gray matter. The location of the measurement of the AIF (artery) and the VOF (vein) are also indicated.

Results

The mean AUC of the VOF and AIF are shown in Fig 1 and Table 1, for 12-, 6-, and 3-mm section thicknesses. With a smaller section thickness, the AUC of the VOF and the AIF increased, and Fig 2 shows that for all patients a smaller section thickness resulted in an equal or larger AUC for both the AIF and VOF. This increase was statistically significant for the mean AUC of 6- versus 12-mm, 3- versus 12-mm and 3- versus 6-mm section thickness of both the VOF and AIF; however, the difference between 6 and 3 mm became borderline significant when the optimal VOF was compared (Table 1, right columns). Figure 3 shows the raw attenuation curves of the optimal VOF for 3-, 6-, and 12-mm section thickness. The differences in AUC between the 3 section thicknesses were larger for the AIF than for the VOF. With a smaller section thickness, the width of the AIF decreased slightly. This difference is statistically significant for the mean width of 3- versus 12-mm-thick sections. When the widths of the optimal AIF were compared, however, this difference was no longer significant.

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Mean AUC for the VOF (A), AIF (B) and mean widths of the AIF (C) for 12-mm (○), 6-mm (▪), and 3-mm (•) section thickness in 10 patients. In all patients, a smaller section thickness results in an equal or larger AUC of the VOF (A) and AIF (B). The width of the AIF decreases slightly (C) with a smaller section thickness in most patients.

• Download figure
• Open in new tab
• Download powerpoint

Fig 3.

Raw attenuation curves of the optimal VOF for 12-mm (○), 6-mm (▪), and 3-mm (•) section thickness (same patient as in Fig 1).

Table 2 shows the mean values of perfusion parameters for different section thicknesses in the 10 patients. The mean CBV and CBF decreased, and the mean MTT increased for smaller section thickness.

Discussion

CTP is considered a valuable method for the assessment of patients with acute stroke and has also been applied to other brain pathologies.11, 12,14 For these clinical implications, quantitative perfusion parameters are essential. A requirement for the determination of quantitative values is that the AUC_AIF is unhampered by PVEs. Because this requirement is not met in most CTP experiments, the AIF is normalized with respect to the VOF, and it is assumed that the VOF is unhampered by PVEs. In a previous article on the influence of AIF and VOF properties on perfusion parameters, the determination of the AIF and VOF showed to be a source of variability in the estimation of perfusion values.15 We evaluated the influence of PVEs on the measurement of AIF and VOF and quantified the effect on absolute perfusion parameters. We observed that, for a section thickness of 6 mm or larger, PVEs affect not only the AIF, but also the VOF, leading to an overestimation of CBF and CBV.

To investigate the effect of section thickness on the AIF and VOF measurements and, consequently, on the quantitative perfusion parameters, the mean AUC of the AIF and VOF were determined for 3 reconstructions of, respectively, 3-, 6-, and 12-mm-thick sections in 10 patients. We observed in all subjects an increase in the AUC of both the AIF and VOF with decreasing section thickness. Selecting the optimal AUC_VOF and AUC_AIF, as occurs in clinical practice, yielded the same findings, except for the AUC_VOF of 3 versus 6 mm, which became borderline significant. Although more prominent for the AIF, these differences were also significant for the VOF. This implies that the selection of a great vein instead of an artery diminishes, but does not solve, the problem of PVEs. The perfusion slab is commonly positioned at the level of the basal ganglia. When the perfusion slab is angulated parallel to the meato-orbital line, the SSS passes diagonally through the slab in most patients.

This obviously introduces PVEs, with the extent being dependent on the individual size and anatomy of the SSS and degree of angulation of the slab; however, other factors could also explain the observed effects. When a multisection technique is used, the amount of attenuation can vary over the sections because of differences in beam hardening. This would result in an AUC of the VOF, which is consistently higher or lower for particular sections across the detector track. The influence of beam hardening is expected to be small, however, because the perfusion slab is only 2.4 cm thick and is not located near structures that are expected to cause the most severe beam hardening, such as the skull base or the protuberantia occipitalis. Another factor that could explain our findings is that the fit of the gamma variate function to the time-attenuation curve could behave differently for thinner sections because of the increased noise level in thin sections. Such differences would influence the shape and thus the AUC of the AIF and VOF. As can be seen in Fig 3, the SNR of the VOF at 3-mm-thick sections is high enough to guarantee a good fit. Therefore, it can be concluded that effects due to a low SNR in the fitting procedure are too small to explain the observed increase in the AUC of the VOF for thinner sections.

The section thickness had a slight effect on the width of the AIF, but selection of the optimal AIF, as occurs in clinical practice, resulted in insignificant differences in the width for the 3 section thicknesses. This means that the MTT is hardly influenced by the section thickness, because the MTT is directly related to the width of the AIF. In theory, a somewhat wider AIF could be explained by the incorporation of tissue in the arterial voxels because the wider tissue function combines with the AIF function; however, the gamma variate function fit, which restricts the AIF to the first pass of the bolus, would most likely eliminate these small effects. These results are in concordance with the findings of Kealey et al.15 In that article, PVEs, which were intentionally induced by a more peripheral placement of the region of interest, significantly increased CBF and CBV, whereas the MTT was not significantly changed compared with central placement of the region of interest.

The CBV and CBF values we observed for a 3-mm section thickness are comparable with the literature. Regions of interest were placed in both gray and white matter, and the mean CBV value of 3.1 and CBF value of 34.4 are in line with a series of positron-emission tomography (PET) and single-photon emission tomography studies in which CBV values ranged between 4.6 and 2.0 for gray matter and 2.6 and 2.0 for white matter and CBF values ranged between 52.1 and 40.2 for gray matter and 20.3 and 26.0 for white matter.16-20

It has been suggested that the substantial variability in perfusion values, related to manual region of interest placement and the definition of variables during postprocessing, could improve by use of a semiautomated system selecting the optimal time-attenuation curves.15,21 Despite use of a semiautomated system, as we did in this study, considerable variation in perfusion values was still observed. By using thinner sections for the VOF determination, this variability decreased, which shows that PVEs do contribute significantly to this variability.

Choosing the optimal section thickness in CTP is a trade-off between increasing PVEs with thicker sections and reducing SNR with thinner sections. The increase in noise affects both the VOF and the measurement of the bolus passage through the brain tissue; however, the effects of the increased noise on the VOF are small, as can be concluded from Fig 3. Therefore, the VOF measurements should be performed on thin sections of approximately 3 mm to minimize PVEs. When the same, small section thickness for the measurement of the bolus passage through the brain tissue is used, however, the SNR on tissue level will become too low to obtain high-quality CTP images. Therefore, the optimal approach is to obtain the AIF and VOF at a small section thickness and to perform subsequent analysis on thicker sections. Both reconstructions can be obtained from the same dataset. Because the SNR of the VOF at a small section thickness is high enough, such an approach does not need an increase in radiation dose. Eventually, the choice of the section thickness in CTP should be validated against a gold standard. This could be done by either correlating the VOF measurements to blood-sampling data or validating the entire CTP procedure against another brain perfusion imaging technique such as PET or stable xenon CT.22

Conclusion

In conclusion, both AIF and VOF are affected by PVEs in sections of 6 mm or thicker, which results in an overestimation of the CBV and CBF. The MTT is hardly influenced by section thickness. We recommend performing VOF measurement on thin sections of approximately 3 mm and conducting subsequent analysis on thicker sections (for example, 10 mm).