Artificial intelligence and cost-effectiveness analyses of radiological imaging in acute ischemic stroke

Acute ischemic stroke is caused by an occlusion of an artery in the brain. Current treatment options for acute ischemic stroke are intravenous thrombolysis and endovascular treatment. In this thesis, three parts describe varying analytical approaches to improve acute stroke care. Part I provides model-based health economic analyses of treatment decisions to improve acute stroke care. Specifically, the benefits of expedited endovascular treatment delivery and the use of CT perfusion for patient selection and occlusion detection are described. In Part II, prognostic imaging markers are studied. Deep learning-based quantification of white matter lesion volume in CT is compared to the radiologist-lead Fazekas scale for prognosticating functional outcome and intracranial hemorrhage occurrence. We studied if intravenous thrombolysis before endovascular treatment might be withheld based on increased risks for poor outcome and intracranial hemorrhage related to white matter lesion load. Furthermore, we evaluated thrombus volume, thrombus length, and thrombus radiomics as patient functional and endovascular treatment procedural outcome predictors. In Part III, we used generative adversarial networks to perform image-to-image translation. We translated CT scans with follow-up hemorrhagic or ischemic stroke lesions to baseline CT scans without. Furthermore, we remove contrast in CTA by translating a CTA to a non-contrast CT. Based on these translations we extract lesion segmentations in follow-up CT and vessel segmentations in CTA.