Detection and characterization by local statistical approaches of dynamical events in image sequences : application to membrane fusion in TIRF microscopy

Abstract : In this thesis, we investigate statistical methods to detect, estimate and characterize dynamical events in image sequences. Our main focus is on fluorescence microscopy images, which represent a fundamental tool for cell biology. There are two cases : 1. Studied objects do not interact, and individual dynamics can be independently analyzed ; 2. Studied objects interact, and group dynamics must be analyzed as a whole. In the case of individual dynamics, our primary focus is on biological image sequences showing proteins evolving in a cell, and more precisely at the cell frontier named plasma membrane. Proteins transported in the cell by vesicles, are observed in total internal reflection fluorescence microscopy (TIRFM), an observation technique well adapted to plasma membrane dynamics analysis. At the end of the exocytosis process, vesicles fuse to the plasma membrane and release proteins, which then diffuse. We first propose a new spot detection method aimed at localizing fusion events. Then, we model the protein dynamics and estimate the biophysical parameters in TIRFM image sequences for further biological analysis. We also address the processing of image sequences at lower magnifications, that is, depicting groups of cells, instead of an isolated cell. We propose a method to jointly estimate quantitative and qualitative motion measurements. It is used to classify the group motion, recover principal paths followed in the scene, and detect localized anomalies. Since they are free of appearance model, the developed methods are quite general and also applied to other applications including crowd motion analysis in videos. Whether it is for spot detection, protein dynamics estimation or group motion analysis, a common approach is ubiquitous, however. First, statistical arguments are used to automatically infer the method parameters. Secondly, we rely on local approaches, which have the advantage of being computationally efficient. Local modeling handles spatially varying image statistics much more easily and more accurately than global modeling. Local approaches also allow neglecting contextual variations such as spatially varying background contrast or, in fluorescence microscopy, temporal fading known as photobleaching.
Document type :
Complete list of metadatas

Cited literature [146 references]  Display  Hide  Download
Contributor : Abes Star <>
Submitted on : Wednesday, April 20, 2016 - 12:21:43 PM
Last modification on : Tuesday, May 22, 2018 - 2:18:27 PM
Long-term archiving on : Thursday, July 21, 2016 - 12:49:00 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01304780, version 1



Antoine Basset. Detection and characterization by local statistical approaches of dynamical events in image sequences : application to membrane fusion in TIRF microscopy. Image Processing [eess.IV]. Université Rennes 1, 2015. English. ⟨NNT : 2015REN1S096⟩. ⟨tel-01304780⟩



Record views


Files downloads