Sie befinden sich hier
Inhalt
Pathology deals with the analysis of tissue samples mainly using microscopic images. Since the entire process from tissue preparation to diagnosis is largely manual, a major goal is to digitize and automate the diagnostic pathology in order to increase efficiency and reliability. Stain normalization is an essential requirement for automatic image analysis in histological images due to strong variance (e.g. color, resolution) from image acquisition, tissue processing, staining, etc.. Deep learning methods have recently become particularly important to solve such problems. We investigate the stain normalization of histological images using Generative Adversarial Networks (GANs), which show great advantages over conventional methods and improve the performance of downstream tasks such as image classification or segmentation [1]. Our second focus is on so-called few-shot or semi-supervised learning methods in order to reduce the amount of annotated data required for the detection and classification of specific cells (e.g. tumor) in histological images.
