PESA R-CNN: Perihematomal Edema Guided Scale Adaptive R-CNN for Hemorrhage Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Intracranial hemorrhage (ICH) presents diverse patterns, challenging accurate detection and localization.
• Accurate identification of ICH is critical due to its high mortality rate.
• Existing methods struggle with the variability in ICH shapes, sizes, and subtle presentations.

Purpose of the Study:

• To develop a novel deep learning model for accurate segmentation of diverse intracranial hemorrhage patterns.
• To minimize missed hemorrhage regions, particularly small or subtle ones.
• To enhance the localization and segmentation performance for ICH detection.

Main Methods:

• Proposed Perihematomal Edema Guided Scale Adaptive R-CNN (PESA R-CNN).
• Introduced Center Surround Difference U-Net (CSD U-Net) for Region of Interest (RoI) generation incorporating Perihematomal Edema (PHE).
• Utilized Scale Adaptive RoI Align (SARA) and Multi-Scale Segmentation Network (MSSN) for scale-specific processing and integration.

Main Results:

• Achieved significant improvements in Dice coefficient (0.697) and Hausdorff distance (12.918) compared to other models.
• Demonstrated a reduction in the false-negative rate by incorporating PHE features.
• Successfully minimized missed small hemorrhage regions and enhanced segmentation for diverse ICH patterns.

Conclusions:

• PESA R-CNN effectively segments various intracranial hemorrhages, outperforming existing methods.
• The integration of PHE and scale-adaptive alignment significantly improves detection accuracy.
• This approach offers a promising solution for improving the diagnosis and management of ICH.