New Semi-Supervised Method Enhances Medical Image Segmentation with Cross-Modal Alignment

Researchers have developed Semi-MedRef, a novel semi-supervised learning framework designed for medical referring image segmentation. This method addresses the high cost of acquiring paired pixel-level annotations and referring texts by effectively utilizing unlabeled data. Semi-MedRef employs three key components: T-PatchMix for alignment-preserving cross-modal augmentation, PosAug for position-aware text augmentation, and Positional Affinity Contrastive Learning (PACL) to encourage anatomically grounded cross-modal representation learning. Experiments on the QaTa-COV19 and MosMedData+ datasets show that Semi-MedRef surpasses existing fully supervised and semi-supervised methods across various label regimes. AI

IMPACT Enhances medical image segmentation accuracy by leveraging unlabeled data and cross-modal alignment, potentially improving diagnostic tools.

RANK_REASON The cluster describes a new research paper detailing a novel method for medical image segmentation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• MosMedData+
• PosAug
• Positional Affinity Contrastive Learning
• QaTa-COV19
• Semi-MedRef
• T-PatchMix
• Yuchen Li

• paper
• other

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