Recurrent Neural Networks for Lagrangian Tracking of Bacteria

Sammanfattning

Research in microbiology is crucial for development of antibiotics, vaccines and othermedicines that cure diseases and prevent spread of viral infections. One methodof studying microorganisms is Lagrangian tracking, where the movement of singlemicroorganisms, such as bacteria, is tracked over long periods of time, which isimportant when studying for example chemotaxis. Lagrangian tracking has previouslybeen implemented using deep learning, showing promising results. However,the model, a convolutional neural network (CNN), struggled to handle overlappingbacteria, which resulted in failure of entire experiments when the model switchedwhich bacterium was currently being tracked. This thesis aimed to create a modelfor Lagrangian tracking that could accurately track over long periods of time as wellas handle overlapping bacteria. The method included simulation of fluorescence microscopicdata as well as design, training and evaluation of recurrent neural networks(RNNs) using the simulated data. The results showed that the RNNs gave lowererror distributions and were able to handle overlapping bacteria better compared tothe CNNs implemented for benchmarking. An analysis of the importance of featuresof the bacteria for tracking indicated that the tracking was harder when surroundingbacteria were close to the focal plane or had higher intensity compared to the bacteriumthat was currently tracked. Although testing the RNNs in an experimentalsetup remains, the results suggest that replacing a CNN with an RNN can improvethe accuracy of the Lagrangian tracking and to greater extent avoid losing the bacteriumduring an overlap. In turn, improving the accuracy of Lagrangian trackingcontributes to the possibility of tracking single microorganisms over long periods oftime and gain more knowledge about for example chemotaxis.