Robust Brain Extraction Tool for Nonenhanced CT and CT Angiography: CTA-BET.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate brain segmentation is crucial for analyzing CT scans.
• Existing brain extraction tools may have limitations with specific CT modalities like CTA and NCCT.

Purpose of the Study:

• To develop and evaluate CTA-BET, a deep learning model for precise brain segmentation.
• To assess CTA-BET's performance on both CT angiography (CTA) and non-contrast CT (NCCT) images.

Main Methods:

• CTA-BET was trained on multi-institutional CTA data and validated on external CTA and NCCT datasets.
• Performance was evaluated using Dice score, Hausdorff distance, and z-score normalized histograms.
• The model was compared against five benchmark noncommercial brain extraction tools.

Main Results:

• CTA-BET achieved superior performance, outperforming all benchmark models.
• Mean Dice scores were 0.99 for CTA and 0.98 for NCCT images.
• CTA-BET demonstrated significantly better Hausdorff distance on CTA images compared to benchmarks.

Conclusions:

• CTA-BET provides a robust and accurate solution for brain extraction on both CTA and NCCT.
• The model has the potential to improve automated imaging analysis in clinical and research settings.
• Deep learning offers a promising approach for advanced medical image segmentation.