Segmentation Style Discovery: Application to Skin Lesion Images

TL;DR

This paper introduces StyleSeg, a novel segmentation method that discovers diverse and semantically consistent segmentation styles from image-mask pairs without needing annotator correspondence, improving skin lesion segmentation accuracy.

Contribution

StyleSeg is the first approach to learn segmentation styles without annotator correspondence, enabling modeling of variability in medical image segmentation.

Findings

Outperforms existing methods on four skin lesion datasets

Achieves strong alignment with annotator preferences on curated dataset

Provides a new measure, AS2, for style-annotator correspondence

Abstract

Variability in medical image segmentation, arising from annotator preferences, expertise, and their choice of tools, has been well documented. While the majority of multi-annotator segmentation approaches focus on modeling annotator-specific preferences, they require annotator-segmentation correspondence. In this work, we introduce the problem of segmentation style discovery, and propose StyleSeg, a segmentation method that learns plausible, diverse, and semantically consistent segmentation styles from a corpus of image-mask pairs without any knowledge of annotator correspondence. StyleSeg consistently outperforms competing methods on four publicly available skin lesion segmentation (SLS) datasets. We also curate ISIC-MultiAnnot, the largest multi-annotator SLS dataset with annotator correspondence, and our results show a strong alignment, using our newly proposed measure AS2, between the predicted styles and annotator preferences. The code and the dataset are available at https://github.com/sfu-mial/StyleSeg.