The accuracy of CTP measurements has been assessed by comparison with experimental imaging and phantom evidence. CTP measurements of blood flow have been compared with and highly correlate to experimental ex vivo microsphere and 2,3,5-triphenyltetrazolium chloride staining measurements in laboratory animals.^2,3 CTP measurements have also been compared with and generally correlate to stable xenon CT, positron emission tomography (PET), and magnetic resonance imaging (MRI) measurements in human patients.^4,5

The term accuracy has several meanings regarding CTP results. Accuracy refers to the (absolute or relative) quantitative perfusion results calculated for a voxel of imaged tissue. For example, how accurately does CTP calculate the true blood flow in a given voxel? Accuracy also refers to the ability of CTP to identify a region of tissue. For example, how accurately does CTP calculate the volume of an infarct lesion? These definitions of accuracy are related yet distinct: the first definition describes how well CTP can measure true blood flow, while the second definition defines how well tissue status or viability can be classified via CTP.

At the voxel level, CTP benefits from the direct linear relation that exists between tracer concentration and x-ray attenuation, because this relation facilitates absolute, quantitative calculation of results.¹ In contrast, bolus-based MR perfusion, the primary alternative clinical perfusion imaging modality, depends on the indirect T₂^* or T₂ gadolinium effect, which is based on a more complicated relation.⁶ (For more details, the see Chapter 6.) The appeal of quantitative results is their promise to identify absolute perfusion thresholds, representing the physiologic basis for clinical decisions.⁷ For many clinical applications (e.g., stroke), hemispheric ratios have become increasingly popular. The latter is because of the paucity of clear absolute thresholds (e.g., because of age or tissue type [gray vs. white matter]) or discrepancies in calibration and accuracy.

During data processing, voxels may be differentiated by their perfusion results and grouped into different tissue zones, visually demarcated by different colors. For instance, in acute stroke, an algorithm might color voxels with blood flow less than X with color 1 (these voxels will
constitute the infarct lesion), while voxels with flow equal or greater than X with color 2 (marking noninfarcted tissue). Such an algorithm enables visual discrimination between tissue types (e.g., infarct lesion vs. noninfarcted tissue) and may be used to calculate the volume of a lesion. However, no universal algorithm for demarcating voxels into tissue zones exists. It is also important to stress that CTP-based tissue classification (e.g., irreversibly damaged tissue vs. normal tissue) relies on the assumption that the hemodynamic parameters can truly predict tissue fate, which is still a subject of intense research.

Returning to ischemic stroke, one of the aims of CTP is to identify and discriminate between different tissue states, including normal tissue, ischemic penumbra, and infarct core (i.e., irreversibly damaged tissue). At the present time, there are a variety of hemodynamic profiles that have been reported to describe these different tissue types. A review found multiple profile definitions for penumbra, including the entire region of hypoperfusion seen on CTP using MTT alone, using CBF alone, using MTT and CBV, using time to peak (TTP) and CBV, using CBF, CBV, and TTP, among others.⁸ Overall, the review identified four different profile definitions for the core, 17 different profile definitions for the penumbra, and two different profile definitions for normal tissue. Tissue discrimination is further complicated by disagreement about which quantitative thresholds optimally describe different tissue states. For core, thresholds for mixed gray-white matter include CBV: 2.0 mL/100 g, MTT: 6.05 seconds, regional CBF (rCBF): 0.34 mL/100 g/min; thresholds for gray matter only include rCBF: 0.20 mL/100 g/min, and CBF*CBV: 31.3; thresholds for white matter only include CBF: 9 mL/100 g/min, CBV: 0.82 mL/100 g, and CBF*CBV: 8.14. For penumbra, thresholds for mixed gray-white matter include CBF: 27.9 mL/100 g/min, CBV: 1.69 mL/100 g, MTT: 6.53 seconds, rCBF: 0.50 mL/100 g/min, regional CBV (rCBV): 0.85 mL/100 g, relative MTT (rMTT): 1.45 seconds; thresholds for gray matter only include CBF: 25 mL/100 g/min, rCBF: 0.48 mL/100 g/min, rCBV: 0.60 mL/100 g, rMTT: 1.60 seconds; no thresholds for white matter only were reported.^8,9

Thus, heterogeneous tissue profile definitions and heterogeneous perfusion thresholds undermine tissue state discrimination consensus. This variability is related to the fact that different CTP data acquisition and data processing techniques (described below) generate different quantitative hemodynamic values at the pixel level and, consequently, at the regional tissue and lesion levels.

Alternatively, in stroke qualitative visual inspection, or eyeballing, of CTP maps may identify normal tissue, penumbra, and infarct core, but this is dependent on user interpretation, which in turn is dependent on skill level and subjective thresholds. In a recent study, five stroke specialists independently reviewed CTP maps for acute ischemic stroke patients by qualitative visual inspection. Inter-rater agreement about whether or not to perform endovascular treatment was fair (Cohen’s kappa statistic = 0.29, range = 0.07 to 0.67) and intra-rater agreement was poor (kappa statistic = 0.14, range = −0.27 to 0.27).^10,11 This study, alongside similar MR perfusion studies,¹² demonstrates the considerable disagreement in treatment decisions based on solely qualitative visual inspection of perfusion studies because of subjective tissue state identification and discrimination. Of course, suboptimal tissue state discrimination may erroneously lead to the exclusion of appropriate patients (type II error) and the inclusion of inappropriate patients (type I error) for treatment, leading to the failure of clinical trials and, for example, obscuring the benefit reperfusion therapies in acute stroke. In addition, visual inspections work only in patients with unilateral disease. In patients where both brain hemispheres are affected and in the absence of focal lesions, the visual assessment without quantitative values is flawed.

Technique differences occur during the data acquisition, data processing, and image interpretation stages of a CTP study,¹³ and several known factors cause variability in results.

Volume coverage refers to the field of view achieved by the CT scanner. The larger the volume coverage, the more tissue is imaged. Larger detector widths describe lesions more completely than smaller detector widths, which may only partially describe them. Lesion volume calculations differ depending on the imaging volume coverage and the location of tissue covered by slices.^14,15,16

The data acquisition rate is the number of images obtained by the CT scanner per unit of time. It is proportional to temporal resolution but also to radiation dose, and, therefore, it must be optimized to balance adequate image quality with radiation safety. The data acquisition rate affects CTP results, but the specific relation is controversial, with different studies recommending different data acquisition rates ranging from one image per 0.5 second to one image per 3 seconds.^17,18,19,20 Similarly, data acquisition duration should allow the tracer bolus to wash completely in and out of the imaged tissue, because truncated scanning duration may impair study source data. There are also different acquisition options that one should be aware of when protocoling CTP studies: cine, burst mode, helical shuttle, and jog or shuttle mode. All of them have strengths but also weaknesses.

User inputs are the subjective selections that operators make during CTP acquisition and postprocessing. Operators select the slices that are scanned. They also
select important inputs that are used to process image data. For instance, operators draw regions of interest on top of raw source images to label the arterial input function(s) and the venous output function. There are some reports for brain studies that indicate that varying placement of the arterial input function does not appear to cause significant results variability,^21,22 with the exception that placement distal to an arterial thrombus causes significant variability.²³ These findings of course need to be interpreted with some healthy skepticism, as certain parameters, such as Tmax, do depend more on the placement of the arterial input function than others (e.g., CBV). Varying placement of the venous output function causes significant variability in CTP results.²⁴ Here, again, the parameter of study is important to consider. The venous output function is much less important for parameters, such as Tmax, than for parameters such as CBV. The exact importance, however, of the arterial input function (AIF) and venous output function placement needs to be vetted in large-population studies. Operators also may subjectively select the voxels or regions of interest on CTP maps for quantitative interrogation. Studies of operator variation are split, with several authors concluding that operator variation causes only minimal variability in CTP results,^24,25,26,27 while others report operator coefficients of variability greater than 30%, suggesting that user inputs and inter- and intraoperator variation cause substantial variability in CTP results.^28,29 Again, one should not generalize the operator variability as it is likely also dependent on the hemodynamic parameter in question. For example, there is most likely an impact on Tmax whether the AIF is picked in the anterior or posterior circulation or whether or not the AIF is picked in an artery on the ipsilateral versus contralateral side of a stroke.