William A. Copen

Perfusion imaging can reveal pathology that is otherwise invisible to the neuroradiologist’s eye. In acute stroke, CT- and MR-based perfusion imaging uniquely depict the alterations in microvascular blood flow that threaten brain tissue viability, and these techniques are increasingly used to direct the therapies that attempt to keep ischemic neurons alive. However, when neuroradiologists interpret perfusion images, they may unknowingly base critical clinical decisions upon hemodynamic measurements that are severely inaccurate.

This study exposes the hidden measurement errors that occur when perfusion scan durations are too short to completely sample the passage of an injected contrast bolus through the brain, a situation that is probably quite common. Although physiologic studies have suggested that adequate sampling might require scanning for well over a minute, scans as short as 40 seconds have been used, both in clinical practice and in the published studies that form the basis of perfusion imaging–based decision making. We studied the errors that short perfusion scans can produce, by retrospectively examining perfusion data that were gathered from patients with acute stroke using an unusually long MR perfusion scan lasting 110 seconds after contrast injection. We simulated the results that shorter scans would have yielded, by progressively deleting the images that were acquired last.

We processed these data using several different commonly used postprocessing algorithms, to produce maps of the four regional hemodynamic parameters that are most often studied with perfusion imaging: cerebral blood volume (CBV), cerebral blood flow (CBF), mean transit time (MTT), and the time at which the deconvoluted response function reaches its maximum (Tmax). We studied how these measurements erroneously varied as a function of simulated scan duration, within regions of interest (ROIs) that were placed within ischemic lesions identified in diffusion-weighted images (DWI). We chose to study perfusion measurements within DWI lesions because especially severe ischemia might be presumed to exist within these lesions, and because some studies have suggested that perfusion images can substitute for DWI in identifying irreversibly injured tissue. Perfusion measurements for all parameters except Tmax were normalized, by dividing them by values that were obtained from a second ROI, which was placed within normal-appearing tissue in the contralateral hemisphere.

We found that truncation of perfusion data by short perfusion scans caused severe underestimation of all four perfusion parameters, with the exception of CBF, for which measurement errors were much smaller. For example, our data suggest that shortening a perfusion scan from 110 seconds postinjection to 39.5 seconds would result in a predicted underestimation of CBV by between 47.6% and 64.2% of the normal value, depending upon the calculation method that was used. Similarly, the shortest scan duration caused a predicted underestimation of MTT by 133–205%, and underestimation of Tmax by 6.19–8.00 seconds. However, CBF was underestimated by only 1.96–4.10%.

We also estimated the proportion of patients for whom the errors introduced by a short scan might result in “lesion reversal,” ie, the erroneous calculation of a below-normal measurement when the actual value was above-normal, or, in the case of Tmax, the erroneous calculation of a value less than or equal to 6 seconds when the actual value was greater than 6 seconds. We found that lesion reversal occurred in 37–46% of patients for CBV, suggesting that short perfusion scans might cause the appearance of a low-CBV lesion when no such lesion was actually present, in more than one-third of patients. Lesion reversal was found in 2–4% of patients for CBF, 28–54% of patients for MTT, and 42–44% of patients for Tmax.

Because the volume of tissue with Tmax greater than 6 seconds has been used in some clinical trials to determine eligibility for emergent recanalization therapy, we also studied how calculation of these volumes might vary artifactually as a function of scan duration. Using a regression model derived from all of our patients’ data, we found that reduction of scan duration from 110 to 39.5 seconds caused a predicted false reduction of Tmax lesion volume by either 71.5 mL or 93.8 mL, depending upon which of two calculation methods was used.

The truncation-related errors that we found were large enough to change individual patients’ management, or to alter the results of clinical trials. Therefore, we believe that our findings highlight the importance of avoiding truncation-related errors by using perfusion scans that are sufficiently long, and probably longer than those that many readers are using currently. We hope that our study will lead to increased attention to the algorithms that are used by postprocessing software to create perfusion maps, the details of which are often omitted from published research on perfusion imaging. We suggest that perfusion postprocessing algorithms should be tested to determine their vulnerability to scan duration-related artifacts such as those illustrated by the current study. As such testing is more difficult when vendors of proprietary postprocessing software do not disclose the details of the algorithms that they use, greater transparency may be beneficial in the development of postprocessing software that is intended for clinical use.

Read this article at AJNR.org …