English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Contrastive learning has shown great promise over annotation scarcityproblems in the context of medical image segmentation. Existing approachestypically assume a balanced class distribution for both labeled and unlabeledmedical images. However, medical image data in reality is commonly imbalanced(i.e., multi-class label imbalance), which naturally yields blurry contours andusually incorrectly labels rare objects. Moreover, it remains unclear whetherall negative samples are equally negative. In this work, we present ACTION, anAnatomical-aware ConTrastive dIstillatiON framework, for semi-supervisedmedical image segmentation. Specifically, we first develop an iterativecontrastive distillation algorithm by softly labeling the negatives rather thanbinary supervision between positive and negative pairs. We also capture moresemantically similar features from the randomly chosen negative set compared tothe positives to enforce the diversity of the sampled data. Second, we raise amore important question: Can we really handle imbalanced samples to yieldbetter performance? Hence, the key innovation in ACTION is to learn globalsemantic relationship across the entire dataset and local anatomical featuresamong the neighbouring pixels with minimal additional memory footprint. Duringthe training, we introduce anatomical contrast by actively sampling a sparseset of hard negative pixels, which can generate smoother segmentationboundaries and more accurate predictions. Extensive experiments across twobenchmark datasets and different unlabeled settings show that ACTIONsignificantly outperforms the current state-of-the-art semi-supervised methods.

Bootstrapping Semi-supervised Medical Image Segmentation with Anatomical-aware Contrastive Distillation