Jay Newby (UNC-CH), Applied Mathematics Colloquium

Tea at 3:30 in 330 Phillips Hall

Title: An artificial neural network approach to automated particle tracking analysis of 2D and 3D microscopy videos

Abstract: Tracking of microscopic species is one of the most utilized experimental technologies in materials science, biophysics, tissue engineering and nanomedicine. The goal is to draw inferences (e.g., viscous and elastic moduli, mesh spacings, passage times) by statistical analysis of particle traces. This in turn allows for selection of a candidate underlying transport mechanism, or to translate mechanistic understanding to computational models with predictive power. Routinely tracked “particles” include pathogens such as viruses and bacteria, bacteriophages, passive and active microbeads, and nano-sized drug carriers. The relevant biological fluids are typically complex, such as viscoelastic and heterogeneous mucus secretions or extracelluar matrices, which often creates imaging environments with exceedingly poor signal-to-noise ratios (SNR). Worse, due to significant time and technical constraints in extracting accurate time-variant positional data from recorded movies, virtually all particle tracking experiments are performed with 2D movies that only captured the dynamics of the species of interest for very limited durations before the species swim or diffuse out of the focal plane. To overcome these shortcomings, we have developed a new approach for particle identification and tracking, based on machine learning and convolutional neural networks (CNN), a type of feed-forward artificial neural network designed to process information in a layered network of connections that mimics the organization of real neural networks in the mammalian retina and visual cortex. Our neural network tracker is capable of exceedingly accurate 3D particle tracking (<0.5% false positive rate, <10% false negative rates) even down to SNR < 1. This technical advance consequently enables high resolution 3D particle tracking, dramatically increasing the temporal duration and richness of microscopy data. The experimental-mathematical advances will dramatically reduce uncertainty in mathematical model selection and in statistical inferences of all currently tracked species, while generating novel insights into many critical physiological processes.