ICube - MIV, Université de Strasbourg, Illkirch
Date(s) : 26/11/2021 iCal
14 h 30 min - 15 h 30 min
Recent advances in super-resolution techniques have greatly improved the resolution of fluorescence microscopy. However, this progress is still hampered by resolution anisotropy and partiallabelling issues. In this talk we will address these limitations in the single particle reconstruction paradigm. The idea is to perform multi-view reconstruction of a given biological particle from 3D images containing hundreds of copies this particle withunknown poses. We will first give an overview of the current challenges in SPR at each step of the reconstruction pipeline.
We will then focus on the pose estimation problem in single molecule localization microscopy (SMLM), which provides data in the form of point clouds corrupted with high anisotropic localizationnoise. Our approach follows the framework of reconstruction of a Gaussian mixture model (GMM) with an expectation-maximization (EM) algorithm. Contrary to existing methods that implicitly assume isotropic Gaussian noise, we introduce an explicit localizationnoise model that decouples shape modeling with the the GMM from noise handling. We design a stochastic EM algorithm that considers noise-free data as a latent variable, with closed-form solutions at each EM step. The first advantage of our approach is to handlespace-variant and anisotropic Gaussian noise with arbitrary covariances. The second advantage is to leverage the explicit noise model to impose prior knowledge about the noise available in SMLM.