Multiparametric AI-based perfusion analysis outperforms Tmax thresholding for critically hypoperfused tissue...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Stroke Imaging

Background:

• Perfusion imaging is crucial for identifying hypoperfused tissue in acute ischemic stroke (AIS).
• Conventional threshold-based methods are commonly used but may have limitations in accuracy.
• Estimating the extent of ischemic tissue is vital for treatment decisions and outcome prediction.

Purpose of the Study:

• To compare an artificial intelligence (AI)-driven multiparametric approach with conventional thresholding methods.
• To evaluate the accuracy of AI in estimating ischemic core and hypoperfused tissue volumes in AIS patients.
• To determine if the AI approach offers improved accuracy over standard methods.

Main Methods:

• Analysis of 186 AIS patients from the HIBISCUS-STROKE cohort.
• Comparison of a deep-learning AI model with standard thresholding (ADC for core, Tmax for hypoperfusion).
• Correlation of AI and threshold-based volumes with final infarct volume on follow-up MRI.

Main Results:

• AI-derived ischemic core volume strongly correlated with ADC-based estimates (ρ = 0.82, P < 0.0001).
• Both AI and thresholding methods showed similar correlation with final infarct volume in recanalized patients.
• AI-estimated hypoperfused tissue volumes were significantly lower and showed more favorable bias than Tmax thresholding in non-recanalized patients.

Conclusions:

• The AI-driven multiparametric approach refines the estimation of critically hypoperfused tissue in AIS.
• The AI method maintains comparable performance to conventional methods for assessing the ischemic core.
• AI offers a promising advancement for more accurate perfusion imaging analysis in stroke.