TIA Centre: News

A recent study introduces a novel deep learning method for segmenting blood vessels (BVs) in H&E tissue slides, addressing the challenge of BV heterogeneity. The method uses guiding maps created through colour transformations to highlight potential BVs, which are then combined with RGB images and fed into U-Nets for segmentation. This approach significantly improves BV segmentation both quantitatively and qualitatively. Future work will focus on collecting a large dataset of H&E images registered with IHC images to assess the method's generalisability and exploring the relationship between BV characteristics and cancer prognosis and treatment response. This work has been accepted for publication in ISBI 2025. An arxiv version of the paper is available at https://arxiv.org/html/2501.12323v2Link opens in a new window