CVPR Daily - Thursday

Aaron Valero Puche is a Research Engineer, and JaeWoong Shin and Jeongun Ryu are Research Scientists at Lunit in South Korea. Lunit develops AI-powered solutions to detect early-stage cancer and optimize treatment. They have proposed and publicly released a dataset dedicated to cell-tissue relationships for cell detection in histopathology. They speak to us ahead of their poster this afternoon. OCELOT: Overlapped Cell on Tissue Dataset for Histopathology 16 DAILY CVPR Thursday Poster Presentation In the field of computational pathology , researchers analyze biopsy samples using large-scale digital images. However, these images pose a challenge due to their enormous size, making it difficult to input them into computational models directly. Pathologists traditionally use sophisticated visualization tools to examine these images, allowing them to observe fine-grained details such as cell information and zoom out to gain insights into larger tissue structures. The combination of the two is very informative . Existing works usually approach the tasks of cell detection and tissue segmentation separately. In this paper, the team seeks to combine these tasks to enhance their performance . “ These two tasks, if we do them simultaneously, have great potential for the identification of cells, ” Aaron tells us. “ We propose a dataset called OCELOT , where we have annotated tissue overlapped with cell patches. That means a patch that contains tissue annotations. Within that region, we select a patch at random, and within these cell patches contained in the tissue patch, we annotate the JaeWoong Shin Aaron Valero Puche Jeongun Ryu