Rethinking deep learning in bioimaging through a data centric lens

Abstract

Deep learning has become essential in bioimaging for tasks. By examining data-centric strategies in general AI and revisiting existing deep learning methods in bioimaging, we describe a prototypical “BioData-Centric AI” framework. For AI users in bioimaging, this framework promotes a more practical approach beyond simply annotating large datasets or relying on a universal model. For method developers, it highlights key research directions to enhance AI toolboxes for the bioimaging community.