Structure Preserving Cycle-Gan for Unsupervised Medical Image Domain Adaptation.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Domain shift significantly degrades medical image segmentation model performance on unseen data.
• Unsupervised domain adaptation (DA) is crucial for leveraging diverse medical imaging datasets.

Purpose of the Study:

• To introduce the Structure Preserving Cycle-GAN (SP Cycle-GAN) for unsupervised domain adaptation in medical image segmentation.
• To enhance medical structure preservation during image translation in Cycle-GANs.

Main Methods:

• Developed SP Cycle-GAN by incorporating a segmentation loss term into the Cycle-GAN training process.
• Evaluated SP Cycle-GAN on binary blood vessel segmentation (STARE, DRIVE) and multi-class cardiac segmentation (MM-WHS).
• Assessed structure preservation visually and quantitatively using Dice scores.

Main Results:

• SP Cycle-GAN demonstrated superior performance over baseline and standard Cycle-GAN approaches.
• Achieved state-of-the-art Myocardium segmentation Dice score (DSC) of 0.7435 for MR to CT adaptation in MM-WHS.
• Successfully preserved anatomical structures during unsupervised domain adaptation.

Conclusions:

• SP Cycle-GAN effectively addresses domain shift challenges in medical image segmentation.
• The proposed method enhances segmentation accuracy and preserves critical anatomical details.
• SP Cycle-GAN represents a significant advancement in unsupervised domain adaptation for medical imaging.