PhD subject: Registration of multiple point clouds in a deep learning framework. Application to single molecule localization microscopy
Starting date: September/November 2020
Profile of the candidate
Application before May 8th: Send a CV and a description of your motivation, as well as the transcript of marks for the past 2 years to Denis Fortun (email@example.com) and Sylvain Faisan (firstname.lastname@example.org).
Single molecule localization microscopy (SMLM) is one of the most powerful and widely used fluorescence imaging techniques. To push further the resolution, the goal of this thesis is to develop single particle reconstruction (SPR) methods for SMLM. The principle is to reconstruct a single particle that fuses information from several of its replicates imaged at random orientations. Recently, SPR has been successfully investigated for several fluorescence imaging modalities. However the data provided by SMLM is in the form of point clouds and requires the development of a dedicated registration method. Traditional approaches for point clouds registration in computer vision have revealed critical limitations to deal with the specific features of our SMLM data, namely (i) strong anisotropic localization uncertainty associated to each point, (ii) very high number of particles to be registered jointly, and (iii) symmetry priors on the shape of the particles. To overcome this issue, the methodological approach of the thesis will be based on a deep learning framework, which has been emerging in recent years for point clouds processing and offers promising perspectives. The PhD student will have to develop innovative methods in this framework to handle the aforementioned issues of SMLM data. The project will be carried in close collaboration with biologists of the University of Wurzburg, to guarantee the access to real data and a relevant biological application.
(c) GdR 720 ISIS - CNRS - 2011-2020.