Machine learning approach for hemorrhagic transformation prediction: Capturing predictors' interaction

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning in Medicine

Background:

• Hemorrhagic transformation (HT) is a common complication of ischemic stroke.
• HT can negatively impact patient prognosis and outcomes.
• Predicting and understanding HT is crucial for stroke management.

Purpose of the Study:

• To determine the incidence of HT in ischemic stroke patients.
• To identify key predictors of HT.
• To evaluate and compare the performance of various machine learning models in predicting HT and analyze predictor interactions.

Main Methods:

• A prospective study of 354 ischemic stroke patients.
• Utilized MRI for HT detection and predictor identification.
• Employed machine learning algorithms (LRC, SVC, RFC, GBC, MLPC) for prediction model development and validation.
• Investigated predictor interactions using generalized additive modeling (GAM).

Main Results:

• The incidence of HT was 19.8%.
• Infarction size, cerebral microbleeds (CMB), and NIHSS were significant predictors of HT.
• RFC and GBC models achieved the highest predictive performance (AUC: 0.91).
• Significant linear and non-linear interactions were found between NIHSS, CMB, and infarction size.

Conclusions:

• Infarction size, CMB, and NIHSS are key predictors of HT in ischemic stroke.
• RFC and GBC models effectively capture non-linear predictor interactions for improved HT prediction.
• Predictor interactions suggest dynamic risk assessment rather than fixed cutoffs for HT.