Assessment of intracranial arterial collateralisation and venous microperfusion profile to predict infarct evolution in patients with early ischemic stroke and endovascular treatment using machine learning models
Final Report Abstract
Acute ischemic stroke due to large vessel occlusion is still one of the leading causes for disability and morbidity worldwide, and triage of acute stroke patients remains challenging. Many factors come into play to optimally triage stroke patients and to select optimal treatment candidates that benefit from minimally invasive stroke treatment called endovascular thrombectomy. One of the key factors that needs to be evaluated during treatment triage is the robustness of intracerebral blood flow by means of medical imaging. Usually, intracranial collaterals are assessed on medical imaging at patients’ admission and are regarded as important biomarkers indicative for the quality of cerebral microperfusion in ischemic brains. Most commonly, arterial collaterals are assessed on pre-treatment CT angiography images. However, arterial collateral assessment on CT angiography does not measure tissue perfusion, as it only represents blood flow to but not through ischemic tissue. Recent studies suggested that venous collaterals may be more sensitive for the assessment of cerebral microperfusion since they represent blood flow after permeating ischemic brain tissue. However, studies on venous outflow profiles and their potential utilization as collateral biomarkers are scarce and comprehensive clinical studies were needed to investigate this matter. Moreover, collateral assessment (whether arterial or venous) is inherently subjective, and to date, no standardized, reliable, and automated procedure exists to determine collateral blood flow pathways on multiple levels of the collateral cascade in ischemic stroke patients. We conducted a series of clinical trials using a multicenter stroke database including patients from the University Medical Center Hamburg-Eppendorf and Stanford University. 740 stroke patients with large vessel occlusion anticipated for thrombectomy treatment were enrolled. In a first step, we found that favorable venous outflow profiles are strongly correlated to perfusion imaging biomarkers of tissue microperfusion prior to treatment, and that both perfusion biomarkers and venous outflow profiles strongly predicted functional clinical outcomes 90 days after successful thrombectomy treatment. In two additional studies, we investigated the impact of venous outflow and perfusion imaging biomarkers on cerebral edema progression prior to treatment, which is known to be a key parameter that is closely related to procedural and clinical outcomes. We found that favorable venous outflow profiles and perfusion imaging parameters were strongly linked to reduced early edema formation on baseline medical imaging and to favorable clinical outcomes following thrombectomy treatment. In another study, we evaluated the impact of favorable venous collateral profiles on procedural outcomes during thrombectomy and found that patients with robust venous drainage exhibited better vessel reperfusion rates compared to those with unfavorable venous profiles. Lastly, we found venous collateral robustness to be strongly dependent on the distinct arterial vessel occlusion location in stroke patients. We were the first group to conduct comprehensive clinical studies to investigate venous outflow profiles and their association with radiological, procedural, and clinical outcomes. We were surprised, to what extent venous outflow information seems to correlate with tissue fate, clinical recovery, and procedural success, which further encouraged us to develop an automated and standardized procedure for the delineation of intracranial collaterals on baseline imaging. Our next aim is to use the gathered information about venous outflow profiles and implement it into the anticipated machine learning algorithm. We are currently testing some diagnostic algorithms for the detection and extraction of arterial and venous time curve information on baseline CT angiography images. We manually labeled arteries and veins on CT angiography images and acquired their specific temporal contrast concentration curves. We will then register the respective labeled CT angiography images to a standardized template space. Next, we aim to train a machine learning algorithm to automatically extract this information from angiography images. Following this step, we strive to improve the diagnostic accuracy of the algorithm by including more data and aim to reduce image artefacts. Thus far, the promising results of our clinical studies paired with the first steps made towards the development of an automated algorithm to detect collateral vessels encourage us to conduct further research in this field.
Publications
-
Association of Venous Outflow Profiles and successful Vessel Reperfusion after Thrombectomy. Neurology. 2021
Faizy TD, Kabiri R, Christensen S, Mlynash M, Kuraitis GM et al.
-
Favorable Venous Outflow Profiles Correlate With Favorable Tissue-Level Collaterals and Clinical Outcome. Stroke. 2021 Mar 8
Faizy TD, Kabiri R, Christensen S, Mlynash M, Kuraitis GM et al.
