Style Augmentation improves Medical Image Segmentation

TL;DR

This paper demonstrates that style augmentation, previously used in classification, effectively reduces texture bias and enhances medical image segmentation performance, especially with limited labeled data.

Contribution

It introduces style augmentation for segmentation tasks, addressing texture bias and improving results on the MoNuSeg dataset.

Findings

Style augmentation reduces texture overfitting.

Segmentation performance improves with style augmentation.

Effective with limited labeled data.

Abstract

Due to the limitation of available labeled data, medical image segmentation is a challenging task for deep learning. Traditional data augmentation techniques have been shown to improve segmentation network performances by optimizing the usage of few training examples. However, current augmentation approaches for segmentation do not tackle the strong texture bias of convolutional neural networks, observed in several studies. This work shows on the MoNuSeg dataset that style augmentation, which is already used in classification tasks, helps reducing texture over-fitting and improves segmentation performance.