Deep learning framework applied to optical diffraction tomography (ODT)

William Pierré (1) , Lionel Hervé (1) , Cédric Allier (1) , Sophie Morales (1) , Sergei Grudinin (2, 2) , Shwetadwip Chowdhury (3) , Laura Waller (3) , Christophe Arnoult (4) , Pierre Ray (4) , Magali Dhellemmes (4)

1 CEA - Commissariat à l'énergie atomique et aux énergies alternatives
2 DAO - Données, Apprentissage et Optimisation
3 UC Berkeley - University of California [Berkeley]
4 IAB - Institute for Advanced Biosciences / Institut pour l'Avancée des Biosciences (Grenoble)

Optical diffraction tomography allows retrieving the 3D refractive index in a non-invasive and label-free manner. A sample is illuminated from various angles and the intensity of the diffracted light is recorded. The light wave can be calculated layer after layer and the inverse problem is usually solved using a gradient descent based algorithm. Here we propose a solution to solve the inverse problem using a neural network where the weights of each layer are the unknown refractive index of the object. Importantly, the matrix product between each layers is replaced by the physics of light propagation.

Domaines: Optique [physics.optics]; Optique / photonique; Apprentissage [cs.LG]

William Pierré, Lionel Hervé, Cédric Allier, Sophie Morales, Sergei Grudinin, et al.. Deep learning framework applied to optical diffraction tomography (ODT). Three-Dimensional and Multidimensional Microscopy: Image Acquisition and Processing XXVIII, SPIE, Mar 2021, Online Only, France. pp.1, ⟨10.1117/12.2582361⟩. ⟨hal-03454249⟩