Machine Learning Models for 3-Month Outcome Prediction Using Radiomics of Intracerebral Hemorrhage and Perihematomal...

Area of Science:

• Neurology
• Radiology
• Data Science

Background:

• Intracerebral hemorrhage (ICH) and perihematomal edema (PHE) are key indicators of brain injury in stroke.
• Accurate prognostication is crucial for managing ICH patients.

Purpose of the Study:

• To evaluate the added value of PHE radiomic features for predicting outcomes in ICH patients.
• To compare machine learning models incorporating radiomics with existing clinical predictors.

Main Methods:

• Utilized a multicentric cohort of 852 acute supratentorial ICH patients.
• Extracted radiomic features from ICH and PHE lesions on CT scans.
• Trained and tested machine learning models using various input strategies, including radiomics and clinical variables.

Main Results:

• Addition of PHE radiomics improved individual risk assessment (IDI, NRI) when combined with ICH radiomics.
• Combining radiomics (ICH and PHE) with clinical variables enhanced risk classification (IDI, NRI).
• Machine learning models demonstrated accuracy comparable to or exceeding the standard ICH score.

Conclusions:

• PHE radiomics enhance individual risk assessment in ICH, though prognostic accuracy shows marginal improvement.
• Machine learning models provide quantitative, immediate risk stratification for potential intervention guidance.