A Boundary-Enhanced Liver Segmentation Network for Multi-Phase CT Images with Unsupervised Domain Adaptation

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Multi-phase computed tomography (CT) is crucial for diagnosing hepatic diseases.
• Liver segmentation in multi-phase CT faces challenges including extensive annotation requirements and poor contrast in certain phases.
• Existing methods struggle with domain shift across different CT phases, necessitating multi-phase annotations.

Purpose of the Study:

• To develop an effective liver segmentation method for multi-phase CT images that addresses annotation burden and contrast issues.
• To propose a dual discriminator-based unsupervised domain adaptation (DD-UDA) framework to enable segmentation without multi-phase annotations.
• To enhance segmentation accuracy by improving boundary recognition in low-contrast CT images.

Main Methods:

• Proposed a dual discriminator-based unsupervised domain adaptation (DD-UDA) network for liver segmentation.
• Implemented feature-level and output-level discriminators to reduce domain distribution differences.
• Introduced a boundary-enhanced decoder to improve recognition of liver boundaries in challenging contrast phases.

Main Results:

• The DD-UDA method achieved superior liver segmentation performance on the target MPCT-FLLs dataset compared to baseline UDA and other state-of-the-art methods.
• Demonstrated significant improvements in Intersection over Union (IoU) scores across different phases (PV, ART, NC) without multi-phase annotations.
• Achieved IoU scores of 0.823 (PV), 0.811 (ART), and 0.800 (NC), outperforming the baseline's 0.785, 0.796, and 0.772 respectively.

Conclusions:

• The proposed DD-UDA method effectively addresses the challenges of liver segmentation in multi-phase CT images.
• Unsupervised domain adaptation combined with boundary enhancement significantly improves segmentation accuracy and reduces annotation labor.
• The method shows strong potential for clinical application in hepatic disease diagnosis using multi-phase CT scans.