Talk 25 Jan 2019, 11am, INRIA Grenoble (Montbonnot), F107
Deep learning technologies for living cells microscopy
Dr. Cédric ALLIER (CEA-Leti Grenoble)
In the present talk, I will discuss the recent development of deep
learning technologies applied to microscopy techniques of living cells
(2D and 3D cell culture, organoids, tissue etc.). A large number of
publications appear recently discussing the advantages of CNN applied to
microscopy, to e.g. improve the image quality, reduce the acquisition
time, reduce the number of frame (super-resolution, 3D tomographic
reconstruction), to perform cell segmentation and classification, etc.
In particular I will discuss the combination of deep learning and
lens-free microscopy, a technique that we are developing since 2009 at
CEA-Leti. This technique relies already heavily on computation, since in
this setup the optics are replaced by holographic reconstruction
algorithms. In comparison with existing techniques, this microscope
combines several advantages: it is very small, cheap, yet it provides an
extremely large field of view (30 mm2) showing at a glance N>20.000
cells. Recently, we showed, for the first time, 3D+time acquisitions of
3D cell culture with a lens-free microscope. We obtained time-lapse
views of 3D culture of prostate epithelial cells showing complex
cellular self-organizations that occurred through migration of single
cells, displacement of large clusters, and merging and interconnection
over long distances (>1 mm). Today, our team is pushing further the
computational aspect of holographic microscopy by integrating
deep-learning technologies. Our aim is to develop algorithms that will
help the biologists and clinicians to perform automatic acquisitions and
analysis of cultured cells, cultured tumors, and tissue.