Machine Learning for Detection and Localization of Motion-Blurred Nano-Emitters in Single-Particle Tracking

Abstract

In this thesis, simulation software was optimized to accurately mimic real fluorescent imaging data, by performing a sequence of experiments. Subsequently, simulations were performed containing motion-blurred emitters (fast moving emitters during frame acquisition). Thereafter, convolutional neural networks were trained using labelled data, to be able to detect these low signal-to-noise emitters. A comparison was made to detection methods from literature, which our neural network outperformed. Furthermore, the average position of the motion-blurred particles in the images were also determined and compared to literature. However, in the case of localization the methods from literature outperformed the neural network.