Self adversarial attack as an augmentation method for immunohistochemical stainings

TL;DR

This paper explores how unpaired image translation models hide stain-specific features as imperceptible noise, which can be perturbed to generate diverse plausible images, improving supervised segmentation performance in histopathology.

Contribution

It reveals the hidden stain-specific information in cycle-consistent image translation and introduces a novel augmentation method by perturbing this noise for better segmentation results.

Findings

Perturbing hidden noise yields diverse plausible stain translations.

Augmentation improves supervised glomeruli segmentation performance.

Hidden features relate to stain-specific characteristics in histopathology images.

Abstract

It has been shown that unpaired image-to-image translation methods constrained by cycle-consistency hide the information necessary for accurate input reconstruction as imperceptible noise. We demonstrate that, when applied to histopathology data, this hidden noise appears to be related to stain specific features and show that this is the case with two immunohistochemical stainings during translation to Periodic acid- Schiff (PAS), a histochemical staining method commonly applied in renal pathology. Moreover, by perturbing this hidden information, the translation models produce different, plausible outputs. We demonstrate that this property can be used as an augmentation method which, in a case of supervised glomeruli segmentation, leads to improved performance.