A Deep Learning-Based Framework for Predicting Intracerebral Hematoma Expansion Using Head Non-contrast CT Scan.

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Hematoma expansion (HE) is a key factor in intracerebral hemorrhage (ICH) outcomes.
• Current clinical tools for predicting HE are insufficient.
• Accurate HE prediction is crucial for timely intervention.

Purpose of the Study:

• To develop a fully automated deep learning framework for HE prediction.
• To utilize only non-contrast CT (NCCT) scans for prediction.
• To improve the accuracy of HE risk assessment in ICH patients.

Main Methods:

• A two-stage deep learning framework was trained on a large retrospective dataset (n=2484).
• The framework was evaluated on a validation set (n=608) and a prospective dataset (n=500).
• Performance was measured using Area Under the Curve (AUC), sensitivity, and specificity.

Main Results:

• The framework achieved an AUC of 0.760 on the retrospective validation set.
• The prospective dataset evaluation yielded an AUC of 0.806.
• This significantly outperformed the BAT score (AUCs 0.582 and 0.699).

Conclusions:

• The developed framework accurately and automatically identifies ICH patients at high risk of HE.
• It provides more precise HE predictions compared to the conventional BAT score.
• This tool has the potential to enhance clinical decision-making for ICH management.