Submission Title
Stain Augmentation Palette
Abstract
This piece was created using ISBI paper 913 and a single H&E histological image. Staining variations among different laboratories hinder the effectiveness of computer-aided diagnosis systems. To address this issue, stain colors and concentrations can be synthetically separated, enabling the modification of color information without disrupting the image’s structure. In this piece, stain augmentation is used to generate new random variations of the image, a common practice in training Deep Learning models. Here, the color variance is intentionally high, resulting in vibrant color palettes. Presenting multiple images within a canvas immediately evokes the kaleidoscopic essence of iconic pop-art pieces.
This piece resonates with the pop-art movement ideal of never staying too far from reality. Even if they present unrealistic staining, they still remain suitable for training purposes and have the potential to enhance cancer detection performance.
Authors
- Fernando Pérez Bueno (Universidad de Granada)
- Lucia Manso Ortega ( Basque Center on Cognition, Brain and Language )