Domain-Adversarial Transformer Network for Multiphase Liver Tumor Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate liver tumor segmentation is vital for data-driven analysis.
• Multiphase CT scans are crucial for diagnosis but present segmentation challenges due to variations.
• Existing algorithms struggle with inter-phase inconsistencies and feature integration.

Purpose of the Study:

• To develop a robust method for liver tumor segmentation from multiphase CT images.
• To overcome limitations of current segmentation algorithms caused by variations in CT phases.
• To improve the generalization and performance of segmentation models.

Main Methods:

• A Domain-Adversarial Transformer (DA-Tran) network was proposed.
• A Domain-Adversarial (DA) module was designed to adapt features across NC, ART, PV, and DP CT phases.
• 3D transformer blocks were utilized for patch similarity and global context attention.

Main Results:

• The DA-Tran network achieved state-of-the-art results in liver tumor segmentation.
• The method demonstrated superior performance in handling variations across multiphase CT images.
• Effective feature integration across different CT phases was achieved.

Conclusions:

• DA-Tran offers a promising solution for accurate liver tumor segmentation.
• The network effectively addresses the challenge of segmenting tumors in multiphase CT scans.
• This approach is well-suited for co-segmentation tasks in liver cancer analysis.