Reducing False-Positives in CT Perfusion Infarct Core Segmentation Using Contralateral Local Normalization.

Area of Science:

• Neuroradiology
• Medical Imaging Analysis
• Cerebrovascular Diseases

Background:

• Established global thresholds for infarct core segmentation in CTP scans can yield false-positives.
• These inaccuracies stem from ignoring blood flow variations between gray matter (GM) and white matter (WM), and patient-specific factors like microangiopathy.

Purpose of the Study:

• To introduce and evaluate a novel local normalization method for infarct core segmentation.
• To mitigate overestimation of infarct core volume caused by global normalization limitations.

Main Methods:

• Retrospective analysis of 2830 CTP scans from patients with suspected ischemic stroke.
• Implementation of a local normalization technique alongside conventional global normalization.
• Local normalization involved dividing relative cerebral blood flow (rCBF) maps by their contralateral counterparts to account for GM/WM and microangiopathy differences.

Main Results:

• Local normalization significantly reduced false-positives in cases with microangiopathy (mean 3.7 mL vs. 14.7 mL).
• Core volumes were slightly smaller on average, indicating improved segmentation accuracy.
• The method demonstrated robustness against naturally low blood flow values in deep WM.

Conclusions:

• The proposed local normalization method effectively reduces infarct core overestimation, particularly in deep WM and in patients with microangiopathy.
• Minimizing false-positives in CTP infarct core segmentation can lead to more optimal therapeutic decisions.
• This technique offers a more robust approach to infarct core segmentation in CTP imaging.