Background
Quantitative measurement of the regional cerebral blood flow [rCBF] is a fundamental physiological parameter for characterizing the status of the brain tissue. RCBF measurements have important clinical implications in defining tissue ischemia [
1,
2], in diagnosing neurodegenerative diseases [
3], and in locating and monitoring angiogenically active tumour tissues [
4,
5]. Positron-emission tomography [PET] measurements using a freely diffusible tissue tracer, oxygen-15-labelled water [
15O-H
2O], is regarded as one of the rCBF gold standards [
6‐
8]. This cumbersome technique requires on-line tracer production from a cyclotron and continuous arterial blood sampling. It is expensive, technically demanding, and traumatic for the patient, and is rarely found outside specialized hospital units. An attractive alternative to PET would be dynamic contrast-enhanced computed tomography [CT] or perfusion CT. Perfusion CT methodology has improved profoundly in recent years. It is now a practical technique for the clinical environment due to the development and broad introduction of fast multidetector row CT systems, increasing computer power, new image acquisition protocols, and commercially available perfusion softwares [
1]. It is also relatively cheap, easy, and rapid to perform. Furthermore, in many acute medical and surgical conditions, such as stroke, head injury, and subarachnoid haemorrhage, and in radiotherapy planning, cerebral CT scanning is often the primary imaging modality of choice. This makes perfusion CT particularly applicable for additional tissue characterization [
9]. Another interesting application has arisen with the advent of hybrid imaging techniques such as PET/CT and single photon emission computed tomography/CT. In the same examination, and with only minor additional scan time, perfusion CT can be considered as an adjunct to further visualize the vascular physiology of relevant lesions. This complements and extends the physiological tissue information obtained by PET, e.g. the glucose metabolism using
18F-fluorodeoxyglucose [
10] or hypoxia using
18F-fluoromisonidazole [
11].
However, the underlying kinetic models for rCBF measured by perfusion CT and PET with15O-H2O differ fundamentally. Perfusion CT is based on the dynamic behaviour of an intravascular contrast agent, while15O-H2O PET models a freely diffusible tissue tracer. To investigate this further, we compared the rCBF measurements of the two techniques directly in the same healthy subjects on the same day within a few hours.
Discussion
In this study, we have compared the quantitative rCBF values that can be obtained by two imaging techniques,
15O-H
2O PET and perfusion CT. To date, studies that directly validate perfusion CT in healthy subjects have been very scarce. The majority of studies concern patients with cerebrovascular disease, and validation has been against either the stable xenon-CT method [
20,
21] or
15O-H
2O PET [
22]. Although rCBF can be derived from the non-ischemic hemisphere in stroke patients, the design is suboptimal. It is quite possible that regional perfusion in the undamaged hemisphere is influenced to some degree by either a subclinical tissue pathology, a generalised micro- or macrovascular disease, remote functional effects of neural damage (diaschisis), or a co-morbidity (cardiac function, pulmonary disease). Particularly the quality of the bolus input gives errors in the rCBF determination, e.g. bolus buffering in the lungs. Therefore, a reduced cardiac output will systematically decrease rCBF [
23]. Similarly, in patients with carotid occlusion, selection of a single arterial input function will cause increased delay and dispersion of the contrast agent to the ischemic areas and thus, underestimate rCBF by 15% to 20% [
24].
As the greater part of the validation studies is aimed at cerebrovascular diseases, the regions used focus on the major cerebral artery territories or whole hemispheres. Thus, the rCBF measures are a heterogeneous mixture derived from both the white and grey matter tissues, vascular volumes, and cerebrospinal fluid spaces, but the relative weights of the individual tissue components are unknown [
21,
22,
24‐
26]. This will render direct comparison between our quantitative grey matter rCBF CT values and these studies difficult.
In studies where the white and grey matter rCBF CT values are available, these range from 14 to 30 mL min
-1 100 g
-1 and 40 to 70 mL min
-1 100 g
-1, respectively, with a COV of 25% to 30% [
6,
20,
23,
27]. Although mostly derived from patient studies, these results are quite similar to the results we found in normal healthy subjects. We found average volume-weighted rCBF CT measurements of 21.8 ± 3.4 mL min
-1 100 g
-1 for the white matter and 71.8 ± 8.0 mL min
-1 100 g
-1 for the grey matter. The relative regional between-subject COV ranged from 11% to 28%. The rCBF CT values were significantly larger than the average volume-weighted rCBF PET measurements by 25% in the white matter and 47% in the grey matter. The absolute values were 17.4 ± 2.0 mL min
-1 100 g
-1 for the white matter and 48.7 ± 5.0 mL min
-1 100 g
-1 for the grey matter with a COV in the 10% to 17% range. Our findings correspond to the values previously reported in the literature using this technique [
7,
8,
13,
28,
29]. In a larger Japanese study encompassing 70 healthy subjects spanning 11 institutions, the overall average rCBF for cerebral cortical regions were 42.7 ± 6.3 mL min
-1 100 g
-1 with a COV of 14.6% [
30].
When normalizing to the white matter, the relative regional grey matter COV was nominally lower with PET compared to CT in 10 of 14 ROIs. One explanation for the lower COV with PET was that we used the average of two measurements. This was not done for perfusion CT to keep the radiation dose within acceptable limits. To our knowledge, a test-retest study of baseline rCBF CT has not been reported on healthy subjects.
In the study design, we have tried to limit the variation between the two techniques further by performing same-day measurements within 1 to 2 h. A further source of variation, which has not been considered in previous validation studies, is the impact that changes of the pre-scan arterial blood gas status might have on rCBF [
29]. The P
aCO
2 decreased significantly by 2 mmHg from PET to CT scanning, suggesting slight hyperventilation. Hyperventilation decreases rCBF by washing out P
aCO
2 by approximately 2% per millimetre of mercury [
31,
32]. This would decrease the rCBF CT measurements by 4% and further increase the difference between techniques if corrected for. The hyperventilation itself may have been caused by anticipatory anxiety in the CT scanning session possibly associated to the procedure itself. Mood states in scanning sessions have been investigated by Matthew et al. [
28], and a trend was found for anticipatory anxiety to be lowered from the first to second scans. This may have been forestalled by letting the subjects rest in the CT scanner for several minutes before scans and maybe even perform a 'sham-scan' before scans.
The bias between techniques was not constant, but increased with the increasing rCBF value (Figure
2), which indicates a logarithmic influence. It is possible that there are global effects in a measurement that influences the regional values. This could be between-subject differences, physiological fluctuations, or methodological errors pertaining to the measurement of the input function and the involved corrections [
32,
33] or in the selection of the venous ROI and the arterial input function in perfusion CT [
23,
34]. One strategy is to normalize the rCBF to the tissue that systematically co-varies with the global fluctuations and is not affected by isolated pathological processes. We examined this hypothesis using the white matter as a reference tissue. For the overall grey matter, the normalized rCBF in perfusion CT was only 20% larger than that for PET. Thus, more than half of the difference between techniques can be explained by global fluctuations affecting both tissues alike, but there is a residual effect manifest as a larger contrast between the white and grey matter tissues in perfusion CT. Interestingly, normalization to the white matter reduced the grey matter COV in PET, but increased the COV in perfusion CT, indicating that there are individual grey/white matter differences to be considered. Thus, prior to the use of tissue normalization in a clinical setting, it is essential that effects of bias and noise are well understood.
One aspect that also needs to be examined is the effect of differences in resolution, the partial volume effect [PVE]. The PVE will of course affect not only PET values, but also PCT values, as the resolution of both methods is insufficient to accurately quantify the rCBF in the cortical grey matter. The object of the paper, however, was not so much to measure 'true' cortical rCBF, but to compare two methods under clinical conditions. So the strategy was to have the PVE affect the two methods to the same degree, rather than to introduce a new level of complexity and potential bias by PVE correction through e.g. tissue-segmented MRI. This was done by securing comparable image resolutions and an accurate image registration between the two imaging modalities. Thus, any error caused by tissue heterogeneity in a given ROI would affect the sampled values to the same degree. We do not believe that the differences between methods can be related to image resolution.
The two methods,
15O-H
2O-PET and perfusion CT, are inherently different since perfusion CT relies on the dynamic behaviour of a non-diffusible intravascular iodine medium, whereas
15O-H
2O-PET relies on a tracer that is freely diffusible into the tissues. Strictly speaking, the term 'rCBF' should only be reserved to denote the volume flow rate of blood though a functional tissue that has the ability to exchange nutrients and waste products, thus, the capillary blood flow. However, a purely intravascular tracer, as iodine contrast, will distribute to all vascular segments. Thus, a fundamental flaw with perfusion CT is the presence of high-contrast signals in regions without a functional tissue and a capillary bed such as the choroid plexus, arteries, arterioles, venules, veins, and sinuses. An example can be seen in Figure
1, where a draining vein in the left occipital region has an rCBF signal increase on perfusion CT without any discernable signal on
15O-H
2O-PET, indicating the absence of a functional tissue. The most common strategy is to eliminate vascular pixels in the CT images before calculation of rCBF. A simple regional rCBV threshold of 8 mL/100 g has been suggested as the most accurate [
6]. This threshold, however, was not feasible in our study as large and irregular sections of the brain parenchyma were excluded from analyses. We, thus, chose a threshold of 14.4 mL/100 g that respected tissue integrity and kept the rCBF CT images legible for clinical use. The grey matter rCBV has been measured to 3 to 4 mL/100 g [
30,
35], so both thresholds are somewhat above the normal tissue rCBV level. In the ROI definitions, we carefully omitted obvious larger vascular structures, but there is definitely a contribution from smaller non-capillary vessels and probably also from a PVE from larger vessels. We regard that the blood volume influenced the signal as the dominant error source in the overestimation of rCBF CT values, in the increased contrast between the white and grey matter, and as an important regional noise contribution. This has been recognized previously as well [
6,
36,
37].
Although biased, we found that the rCBF CT does correlate with rCBF PET for each individual over a broad range of values from the white to the grey matter (Figure
3), but poorly if only grey matter rCBF values were considered. Previous studies in patients have found
r2 ranging from 0.5 to 0.8 with significant linear regression slopes of 0.7 to 1.4 [
20,
22,
38,
39] and in healthy subjects,
r2 from 0.4 to 0.9 with slopes 1.0 to 1.55 [
6,
25]. The significant correlations signify that rCBF CT does deliver a perfusion-weighted signal, but with a tendency to overestimate the values particularly for highly perfused regions.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
IL conceived the study. JG, LH, and IL participated in the design of the study. JG and IL coordinated the study, and JG carried out the scannings. RP performed the statistical analysis. All authors drafted the manuscript and read and approved the final manuscript.