The aim of the awarded project is to generate and release the first annotated data set of digitized frozen H&E-stained histological sections from various organs enabling to train and validate algorithms for nuclei instance segmentation in different human organs. At least two state-of-the-art Deep learning (DL)-based solutions comprising U-Net with weighted boundary loss and Mask-RCNN will be implemented and applied on this data set. The dataset alongside with the related codes will be made publicly available on the Kaggle website. Introducing the first rich annotated data set of digitized frozen H&E-stained histological sections, providing detailed descriptions of images and the process of ground truth generation as well as benchmarking the results from at least two advanced DL-based solutions will hopefully motivate the research community to further develop methods for computational analysis of H&E-stained frozen section. At the long run, this can help to generate computational systems supporting pathologists in the intra-operative diagnostic process.
<- Back to: IPA