External Attention Assisted Multi-Phase Splenic Vascular Injury Segmentation With Limited Data

Area of Science:

• Medical imaging
• Radiology
• Computer-aided diagnosis

Background:

• The spleen is frequently injured in blunt abdominal trauma.
• Accurate segmentation of splenic vascular injury aids in severity grading, clinical decision support, and outcome prediction.
• Challenges in splenic vascular injury segmentation include high variability and difficulty in acquiring large, annotated datasets.

Purpose of the Study:

• To develop a novel framework for multi-phase splenic vascular injury segmentation, particularly for limited data scenarios.
• To enhance the accuracy and efficiency of automated splenic injury assessment.

Main Methods:

• Leveraging external data to generate pseudo splenic masks for spatial attention (external attention).
• Developing a synthetic phase augmentation module using generative adversarial networks (GANs) to expand internal data.
• Jointly applying external attention and synthetic data augmentation during training.

Main Results:

• The proposed framework significantly outperforms competing methods.
• The method improves the DeepLab-v3+ baseline by over 7% in average Dice Similarity Coefficient (DSC).
• Demonstrated effectiveness in segmenting splenic vascular injuries with limited data.

Conclusions:

• The novel framework effectively addresses the challenges of splenic vascular injury segmentation with limited data.
• The combination of external attention and synthetic phase augmentation enhances segmentation accuracy.
• This approach holds promise for improving clinical decision support in trauma cases.