We demonstrate the possibility to realize supervised machine learning for a cell detection task without having to manually annotate images through the sole use of synthetic images in the training and testing steps of the learning process. This is successfully illustrated on 3D cellular aggregates observed under light sheet fluorescence microscopy with a shallow and deep learning detection approach. A performance of more than 90% of good detection is obtained on real images.

Date Published: 24 May 2018
PDF: 6 pages
Proc. SPIE 10677, Unconventional Optical Imaging, 1067728 (24 May 2018); doi: 10.1117/12.2303706

Published in SPIE Proceedings Vol. 10677:
Unconventional Optical Imaging
Corinne Fournier; Marc P. Georges; Gabriel Popescu, Editor(s)