Machine learning-based prognostic prediction for acute ischemic stroke using whole-brain and infarct multi-PLD ASL...

Area of Science:

• Neuroimaging
• Radiology
• Machine Learning

Background:

• Accurate early prognostic prediction is crucial for personalized treatment in acute ischemic stroke (AIS).
• Assessing the predictive value of radiomics from multi-post-labeling delay arterial spin labeling (multi-PLD ASL) cerebral blood flow (CBF) images is key.
• Developing integrated prediction models combining imaging and clinical data is essential.

Purpose of the Study:

• To evaluate the predictive power of radiomics features from whole-brain and infarct CBF images obtained via multi-PLD ASL.
• To develop and compare radiomics, clinical, and comprehensive prediction models for AIS prognosis.
• To assess the clinical utility and stability of these predictive models.

Main Methods:

• Radiomics features were extracted from multi-PLD ASL CBF images of 110 AIS patients.
• Five machine learning algorithms were used to build radiomics, clinical, and comprehensive models.
• Model performance was assessed using ROC analysis, decision curves, and permutation tests.

Main Results:

• Combined radiomics models outperformed infarct-only models and showed similar performance to clinical models.
• Comprehensive models integrating radiomics and clinical features demonstrated superior predictive performance.
• A support vector machine-based comprehensive model achieved the highest AUC (0.904), driven by clinical and radiomics features.

Conclusions:

• Whole-brain CBF radiomics can substitute for infarct-specific features, simplifying analysis.
• Combined radiomics models are valuable, especially when clinical data is incomplete.
• The comprehensive model integrating multi-PLD ASL CBF radiomics and clinical data provides a robust tool for early AIS prognosis.