We analyzed CTP images from the International Stroke Perfusion Imaging Registry (INSPIRE), which is a database of acute stroke perfusion imaging and associated clinical information. For this study we used consecutive patients presenting with acute ischemic stroke who had whole brain CTP and who were recruited into INSPIRE between 2010 and 2017 at the John Hunter Hospital, Newcastle, Australia. For standardization, only one site was used at this stage. As is routine in INSPIRE, patients all underwent baseline multimodal CT imaging with non-contrast CT, CTA, and CTP. Written informed consent was obtained from all participants, and the INSPIRE study was approved by the site's ethics committee (23 (link)).
To obtain the perfusion images, a total of 19 acquisitions occurred over 60 s. The CTP data were processed by commercial software MIStar (Apollo Medical Imaging Technology, Melbourne, VIC, Australia). CTP parameters were generated by applying the mathematical algorithm of singular value decomposition with delay and dispersion correction (24 (link)). The following four CTP parameters were generated: cerebral blood flow (CBF), cerebral blood volume (CBV), mean transit time (MTT), and delay time (DT). The penumbra and core volumes were defined with dual thresholds: DT at the threshold of 3 s for total ischemic lesion volume and CBF at the threshold setting of 30% for acute core volume (8 (link), 16 (link), 25 (link)). After single-value thresholding, core/penumbra areas were limited to a single lesion and artifactual or erroneous regions were removed. The resulting map was used as the ground truth (GT). Core/penumbra were reviewed by experts to ensure they were accurate.
To develop the model, we used 86 acute ischemic stroke patients with a large vessel occlusion (LVO): M1 segment of the middle cerebral artery (MCA) or internal carotid artery (ICA). To provide additional testing and external validation, 25 patients were used, with both LVO and non-LVO occlusions. This was done to observe whether a model trained only on lesions resulting from an occlusion of large vessel will perform as well when testing on a variety of occlusion sites. Each patient in the test set underwent follow-up MR diffusion-weighted imaging (DWI) between 24 and 72 h after onset. The volume (mL) of the infarct core, as estimated by MR-DWI, was recorded and used for external validation. On follow-up imaging, all patients had a thrombolysis in cerebral infarction (TICI) score of at least 2b, indicating relatively complete reperfusion of initially hypoperfused regions. In these cases, the volume of the acute CTP core should more closely match that of the follow-up infarct core and could therefore be used to validate the predictions.
To obtain the perfusion images, a total of 19 acquisitions occurred over 60 s. The CTP data were processed by commercial software MIStar (Apollo Medical Imaging Technology, Melbourne, VIC, Australia). CTP parameters were generated by applying the mathematical algorithm of singular value decomposition with delay and dispersion correction (24 (link)). The following four CTP parameters were generated: cerebral blood flow (CBF), cerebral blood volume (CBV), mean transit time (MTT), and delay time (DT). The penumbra and core volumes were defined with dual thresholds: DT at the threshold of 3 s for total ischemic lesion volume and CBF at the threshold setting of 30% for acute core volume (8 (link), 16 (link), 25 (link)). After single-value thresholding, core/penumbra areas were limited to a single lesion and artifactual or erroneous regions were removed. The resulting map was used as the ground truth (GT). Core/penumbra were reviewed by experts to ensure they were accurate.
To develop the model, we used 86 acute ischemic stroke patients with a large vessel occlusion (LVO): M1 segment of the middle cerebral artery (MCA) or internal carotid artery (ICA). To provide additional testing and external validation, 25 patients were used, with both LVO and non-LVO occlusions. This was done to observe whether a model trained only on lesions resulting from an occlusion of large vessel will perform as well when testing on a variety of occlusion sites. Each patient in the test set underwent follow-up MR diffusion-weighted imaging (DWI) between 24 and 72 h after onset. The volume (mL) of the infarct core, as estimated by MR-DWI, was recorded and used for external validation. On follow-up imaging, all patients had a thrombolysis in cerebral infarction (TICI) score of at least 2b, indicating relatively complete reperfusion of initially hypoperfused regions. In these cases, the volume of the acute CTP core should more closely match that of the follow-up infarct core and could therefore be used to validate the predictions.
Free full text: Click here
