Computer Vision News - March 2022

33 Naronet: Discovery of Tumor Microenvironment the low number of available labeled tissue images. A data-efficient contrastive learning loss pre-processing step was proposed and seem to be efficient. A visualization of the patch contrastive learning is shown on the following image with a step-by-step illustration of the patch contrastive strategy using random crops from the images and CNN which are fed as image patches in a series of CNNs. NaroNet can learn relevant TMEs local phenotypes, cell- interaction neighborhoods, and neighborhood interaction areas - even when their presence in the tissue is rare. NaroNet can achieve more accurate predictions while providing inherent interpretability of the reason behind those predictions. Overall, the network comprises an ensemble of networks that unsupervised identifies and annotates relevant TMEs that drive patient outcomes with clinical predictions which are directly based on the annotations of TMEs. Let’s catch up Let’s meet next month with a surprise article! See you soon 

RkJQdWJsaXNoZXIy NTc3NzU=