New framework enhances fetal ultrasound segmentation with semi-supervised learning

Researchers have developed DACL, a novel semi-supervised framework designed to improve the segmentation of fetal ultrasound images. This method utilizes both a lightweight convolutional network and a Transformer-based network, leveraging limited labeled data and unlabeled data through consistency regularization. DACL introduces a dual-agreement consistency loss that aligns distributional predictions and uncertainty, aiming to reduce the impact of unreliable pseudo-labels. The framework also incorporates an interpolation-based consistency strategy with mixup to enhance robustness, showing significant improvements in boundary accuracy under extreme annotation scarcity. AI

IMPACT Improves accuracy in medical image segmentation, potentially leading to better diagnostic tools and reduced reliance on manual annotation.

RANK_REASON The item is an academic paper detailing a new methodology for image segmentation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• fetal ultrasound
• mixup
• semi-supervised learning
• Transformer-based network

• paper
• other

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