We analyzed brain MRI examinations of consecutive patients with clinical symptoms of acute stroke and LVO of the anterior circulation between 2012 and 2018 jointly managed or treated by the Department of Neurology and the Division of Neuroradiology, Vascular and Interventional Radiology at the Medical University of Graz. Patients were retrospectively identified using the internal picture archiving and communication system (PACS). We excluded patients not receiving MR perfusion imaging as well as patients without LVO, stroke of the vertebrobasilar circulation, and stroke mimics without detected DWI lesion or vessel occlusion. Eighty-seven patients met those criteria. In six patients, automatic evaluation of perfusion-diffusion mismatch failed either in one or both software applications. A failed calculation was considered when either perfusion data or diffusion data were not provided in one or both of the two software packages due to severe motion artifacts. Finally, a total of 81 patients (38 females, mean age 68.8 ± 12.7 years, range 27–88 years, median NIHSS 11) were included in our analysis. Of these, 53 had occlusions of the middle cerebral artery (M1 or M2 segment), 21 of the internal carotid artery, five of the terminal internal carotid artery, and two patients had tandem occlusions of the internal carotid artery and the middle cerebral artery, respectively.
This study was approved by the local ethics committee in accordance with the declaration of Helsinki.
Magnetic resonance imaging data acquisition
MRI was performed on a single 1.5-Tesla clinical scanner (SIEMENS MAGNETOM Espree, Siemens Healthineers) according to the standard protocol used at our hospital for the workup of patients with acute stroke. This protocol includes the following sequences: axial diffusion-weighted single-shot echo planar imaging (TR/TE 4000/88 ms; isotropic diffusion weighting; b values 0 and 1000 s/mm2, matrix 128 × 128, FOV 240 × 240 mm2), axial perfusion imaging (TR/TE 1800 ms, 40 ms, slice thickness 5 mm, matrix 128 × 128, FOV 240 × 240 mm2, flip angle 60), axial fluid-attenuated inversion recovery (TR/TE 8000 ms/99 ms, slice thickness 5 mm) or optional axial fluid-attenuated inversion recovery blade (TR/TE 8500 ms/98 ms, slice thickness 5 mm), axial T2-weighted (TR/TE 49 ms, 40 ms, slice thickness 2.5 mm), before 2015 axial T2*-weighted GE (TR/TE 936 ms, 26 ms, slice thickness 5 mm), axial 3D TOF angiography (TR/TE 26 ms, 7.00 ms, slice thickness 0.8 mm), axial T1-weighted post contrast media (TR/TE 690 ms, 17 ms, slice thickness 5 mm), and optional angiography of the extra cranial vessels using fast low angle shot magnetic resonance imaging. Contrast media Gadovist (Bayer Vital GmbH), ProHance (Bracco Imaging), and Dotarem (Guerbet) were administered intravenously via a power injector (Spectris; Medrad Inc.) with a dose of 0.1 mmol/kg body weight at 4 ml/s flow followed by 20 ml of NaCl 0.9% at the same flow. ProHance is the standard contrast medium that is substituted by Gadovist or Dotarem in the rare cases of known intolerance. The total acquisition time took 14 min and 18 min, respectively, with the additional extra cranial angiography.
Automated perfusion-diffusion mismatch calculation
The RAPID® software was installed in our hospital 2012 as prerequisite for participation in the SWIFT PRIME stroke trial . The Olea Sphere® DSC plug-in for automated calculation of diffusion-perfusion mismatch was installed 2018 as part of the Olea Sphere® post processing package (including volumetric tumor assessment tractography etc.). Both commercially available software packages are used in the daily clinical routine and there is no personal or financial relationship to the vendors. Quantitative automated perfusion-diffusion mismatch calculation performed by RAPID® was recalculated with Olea Sphere® software for all patients in this study. RAPID® and Olea Sphere® are two fully automated post processing software applications used for quantitative stroke analysis. Both applications provide perfusion and diffusion maps such as relative cerebral blood flow, relative blood volume, and relative mean transit time. Furthermore, both software applications quantify the volume of hypoperfused tissue and ADC lesion and automatically calculate the volume of perfusion-diffusion mismatch. Decreased perfusion is quantified on Tmax maps, an estimate of delayed blood supply whereby a delay of 6 s (Tmax > 6 s) [18, 19] of the time to the maximum of the residue function was chosen as a threshold for hypoperfused tissue for both software applications (see Fig. 1). Threshold for ADC (620 × 10−6 mm2/s) volume estimation was also identical for both software applications . Perfusion-diffusion mismatch, the penumbra, is defined as the difference in the volume of segmented hypoperfused tissue volume and the segmented volume of the ADC lesion. All volumes are given in milliliter (ml).
All MRI source images along with perfusion-diffusion mismatch calculation summaries from RAPID® and Olea Sphere® were independently reviewed and checked for validity by two experienced neuroradiologists (N.H. and U.W.). The neuroradiologists were not blinded to the software. However, the reports were evaluated in terms of correct placement of the arterial input function, and misregistration of regions outside the territory of the occluded artery, respectively. Moreover, studies with distinctive motion artifacts that rendered the results of the automated segmentation useless had been excluded. The morphological MRIs were assessed for T2-FLAIR lesions, hemorrhage, and vessel occlusions.
Statistical analyses were performed using the software SPSS 24 (IBM) and OriginPro 2018 (OriginLab Corporation). For the comparison of parameters evaluated by the two software applications RAPID® and Olea Sphere®, scatterplots were used to investigate the correlation between both methods. Bland-Altman plots were used to detect systematic biases in one of the two methods. For all volume differences in the Bland-Altman plots, we calculated mean ADC volume differences (VolumeRAPID®–VolumeOlea Sphere®). Descriptive statistics was used to quantify the volume differences between both software packages for hypoperfused tissue, ADC, and perfusion-diffusion mismatch (mean values for normal distributed values and median values for non-normal distribution). The Shapiro-Wilk test was used to test for normal distribution of data. Not surprisingly, the estimated volumes were not normal distributed since occlusions of different vascular sections lead to a hypoperfusion of certain cerebral areas. A test for significant differences was therefore performed with the Wilcoxon signed-rank test. Results were reported with z-score (z) and Pearson’s correlation coefficient (r) to report effect size.