3D cell segmentation

To understand how organs develop during pregnancies and to obtain insights into conditions related to premature birth and into the possibility of therapeutic regeneration, biomedical researchers at the University of Cambridge use 3D organoids as a system that can replicate human organs at a scale that is amenable to pharmacological and gene-targeting experiments. 

The function of biomedical tissues is often linked to their morphology, and thus researchers strive to detect the outline of all the individual cells that make up these organoids (a process called instance segmentation in machine learning) to evaluate the efficacy of different treatments. This poses several challenges, as instance segmentation of three-dimensional biomedical data as well as the interpretation of the effect of geometry of the cells on their function are still open problems. 

To solve this, DeepMirror combined two approaches: i) extending the 2D AI cell segmentation tool CytoButler to work with 3D data by training a personalised AI model, and ii) implementing custom analytics for our clients to analyse the effect of different treatments on 3D geometrical parameters and thus, their function. 

Our clients are now using our 3D CytoButler AI solution to detect the morphological parameters of each of the hundreds of cells that make up the organoids, obtaining meaningful insights that will help them study the development of organs and finding new therapeutics for harmful conditions.