Hematoma expansion prediction based on SMOTE and XGBoost algorithm.

Area of Science:

• Medical Imaging and Diagnostics
• Machine Learning in Healthcare
• Neurology

Background:

• Hematoma expansion (HE) is a critical complication following spontaneous intracerebral hemorrhage (ICH).
• Existing prediction models often focus on the initial 6 hours, neglecting significant HE events occurring between 6 and 24 hours post-ICH.
• Accurate prediction of HE is vital for timely medical intervention and improved patient outcomes.

Purpose of the Study:

• To develop and evaluate a predictive model for hematoma expansion (HE) within 24 hours of spontaneous intracerebral hemorrhage (ICH).
• To predict HE occurrence at 6-hour intervals within the 24-hour window post-ICH.
• To address data imbalance issues common in medical datasets using data augmentation techniques.

Main Methods:

• Utilized patient demographics and computed tomography (CT) image features for prediction.
• Employed the XGBoost machine learning algorithm for predictive modeling.
• Applied the SMOTE (Synthetic Minority Over-sampling Technique) algorithm to handle imbalanced datasets.

Main Results:

• The XGBoost model, enhanced with SMOTE, achieved an accuracy of 0.82 and an F1-score of 0.82 for predicting HE within 24 hours.
• The model demonstrated high predictive accuracy at 6-hour intervals: 0.89 (6h), 0.82 (12h), 0.87 (18h), and 0.94 (24h).
• The proposed method outperformed other machine learning models evaluated on the dataset.

Conclusions:

• The XGBoost model, combined with SMOTE data augmentation, provides an accurate and reliable method for predicting hematoma expansion within 24 hours after ICH.
• The model's ability to predict HE at 6-hour intervals offers valuable insights for clinical decision-making.
• This approach holds significant potential for improving the management of patients with spontaneous intracerebral hemorrhage.