Dual Co-Train framework boosts ultrasound tongue segmentation with limited data

Researchers have developed a novel framework called Dual Co-Train for improving ultrasound tongue segmentation, particularly in scenarios with extremely limited labeled data. This source-free domain adaptation method utilizes a lightweight UltraUNet backbone and refines pseudo-labels for unlabeled target data. The system also generates synthetic target-style image-mask pairs using a conditional GAN, enabling closed-loop adaptation without requiring access to the original source data. Evaluations across multiple datasets and transfer pairs demonstrate that Dual Co-Train significantly enhances segmentation overlap and contour accuracy compared to existing baseline methods. AI

IMPACT Enhances segmentation accuracy in medical imaging with minimal labeled data, potentially improving diagnostic tools.

RANK_REASON This is a research paper detailing a new method for a specific computer vision task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Alisher Myrgyyassov
• arXiv
• Dual Co-Train
• UltraUNet

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →