Spatial CT perfusion data helpful in automatically locating vessel occlusions for acute ischemic stroke patients

Area of Science:

• Neuroimaging
• Cerebrovascular Diseases
• Medical Decision Support

Background:

• Accurate localization of vessel occlusions is critical for effective stroke treatment.
• Endovascular thrombectomy is expanding, necessitating advanced methods for identifying occlusion sites.
• CT perfusion (CTP) imaging offers potential for automated vessel occlusion detection.

Purpose of the Study:

• To investigate the efficacy of CT perfusion (CTP) data in automatically locating cerebral vessel occlusions.
• To develop and validate an atlas-based model for identifying occlusion locations using CTP.
• To compare CTP-based localization with expert annotations from CT angiography (CTA).

Main Methods:

• An atlas was created mapping downstream regions to specific vessel segments.
• Hypoperfused regions in CTP data were identified and correlated with the atlas.
• A cohort of 596 patients from the DUtch acute STroke (DUST) study was analyzed.
• The model's accuracy was assessed against expert CTA annotations using Cohen's kappa coefficient.

Main Results:

• The atlas-based model achieved a mean accuracy of 86% in locating vessel occlusions.
• Substantial agreement (kappa up to 0.68) was found between CTP-based localization and CTA annotations.
• Anterior large vessel occlusions and posterior circulation occlusions were located with 80% and 92% accuracy, respectively.

Conclusions:

• The spatial patterns of hypoperfusion on CTP can effectively indicate vessel occlusion locations in acute ischemic stroke.
• The model demonstrates promise for automated clinical decision support in stroke care.
• Inter-patient variations in cerebral vasculature may limit the precision of CTP in differentiating specific occlusion locations.