부산대학교 도서관

Comprehensible and high-quality automated cell nucleus segmentation and classification are required to assist pathol-ogists in their decision making. Commonly, cell nucleus segmentation and classification are either treated as separate tasks or split into different branches of a convolutional neural network. In this contribution, we present our joint cell nu-cleus instance segmentation and classification convolutional neural network ciscNet. In contrast to the state-of-the-art convolutional neural network HoVer-Net that uses multi-ple branches, ciscNet uses a single branch to segment and classify cell nuclei. Our single-branch approach outper-forms HoVer-Net on the histopathological Lizard dataset. In addition, we show that training our approach with the Ranger optimizer yields better results than using the Adam optimizer. Furthermore, we participate as team ciscNet in the Colon Nuclei Identification and Counting Challenge 2022 (CoNIC Challenge 2022). Our code is available at https://git.scc.kit.edu/ciscnet/ciscnet-conic-2022.