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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Real-world medical image segmentation has tremendous long-tailed complexityof objects, among which tail conditions correlate with relatively rare diseasesand are clinically significant. A trustworthy medical AI algorithm shoulddemonstrate its effectiveness on tail conditions to avoid clinically dangerousdamage in these out-of-distribution (OOD) cases. In this paper, we adopt theconcept of object queries in Mask Transformers to formulate semanticsegmentation as a soft cluster assignment. The queries fit the feature-levelcluster centers of inliers during training. Therefore, when performinginference on a medical image in real-world scenarios, the similarity betweenpixels and the queries detects and localizes OOD regions. We term this OODlocalization as MaxQuery. Furthermore, the foregrounds of real-world medicalimages, whether OOD objects or inliers, are lesions. The difference betweenthem is less than that between the foreground and background, possiblymisleading the object queries to focus redundantly on the background. Thus, wepropose a query-distribution (QD) loss to enforce clear boundaries betweensegmentation targets and other regions at the query level, improving the inliersegmentation and OOD indication. Our proposed framework is tested on tworeal-world segmentation tasks, i.e., segmentation of pancreatic and livertumors, outperforming previous state-of-the-art algorithms by an average of7.39% on AUROC, 14.69% on AUPR, and 13.79% on FPR95 for OOD localization. Onthe other hand, our framework improves the performance of inlier segmentationby an average of 5.27% DSC when compared with the leading baseline nnUNet.

Devil is in the Queries: Advancing Mask Transformers for Real-world Medical Image Segmentation and Out-of-Distribution Localization