3D time-lapse imaging of a mouse embryo using intensity diffraction tomography embedded inside a deep learning framework

• Published in: Applied optics. Optical technology and biomedical optics Vol. 61; no. 12; p. 3337
• Main Authors: Pierré, William, Hervé, Lionel, Paviolo, Chiara, Mandula, Ondrej, Remondiere, Vincent, Morales, Sophie, Grudinin, Sergei, Ray, Pierre F, Dhellemmes, Magali, Arnoult, Christophe, Allier, Cédric
• Format: Journal Article
• Language: English
• Published: United States 20-04-2022
• Subjects: Animals; Deep Learning; Mice; Refractometry; Time-Lapse Imaging; Tomography; Tomography, X-Ray Computed
• ISSN: 2155-3165
• DOI: 10.1364/AO.453910

We present a compact 3D diffractive microscope that can be inserted directly in a cell incubator for long-term observation of developing organisms. Our setup is particularly simple and robust, since it does not include any moving parts and is compatible with commercial cell culture containers. It has been designed to image large specimens (>100×100×100µ
) with subcellular resolution. The sample's optical properties [refractive index (RI) and absorption] are reconstructed in 3D from intensity-only images recorded with different illumination angles produced by an LED array. The reconstruction is performed using the beam propagation method embedded inside a deep-learning network where the layers encode the optical properties of the object. This deep neural network is trained for a given multiangle intensity acquisition. After training, the weights of the neural network deliver the 3D distribution of the optical properties of the sample. The effect of spherical aberrations due to the sample holder/air interfaces are taken into account in the forward model. Using this approach, we performed time-lapse 3D imaging of preimplantation mouse embryos over six days. Images of embryos from a single cell (low-scattering regime) to the blastocyst stage (highly scattering regime) were successfully reconstructed. Due to its subcellular resolution, our system can provide quantitative information on the embryos' development and viability. Hence, this technology opens what we believe to be novel opportunities for 3D label-free live-cell imaging of whole embryos or organoids over long observation times.