Ph.D. thesis by Tiago José Arieira Esteves
Author: Tiago José Arieira Esteves
Date: June 17, 2016
Aurélio Campilho, Full Professor, Faculty of Engineering of the University of Porto (President)
Armando José Formoso de Pinho, Associate Professor, Electronics and Tecommunications Department, University of Aveiro;
João Miguel Raposo Sanches, Assistant Professor, Bioengineering Department, Instituto Superior Técnico, University of Lisbon;
Ana Maria Mendonça, Associate Professor, Faculty of Engineering of the University of Porto;
Jaime dos Santos Cardoso, Associate Professor, Faculty of Engineering of the University of Porto;
Pedro Silva Quelhas, Researcher, I3S-Instituto de Investigação e Inovação em Saúde, University of Porto (Supervisor);
Maria Cardoso Oliveira, I3S-Instituto de Investigação e Inovação em Saúde, University of Porto (Co-supervisor).
The study of cancer cell mobility under different conditions is fundamental to dissect the associated molecular mechanisms and to validate possible therapies for its regulation. To evaluate cancer cell mobility, biologists established in vitro assays, recording time-lapse microscopy videos. In such procedure only through the use of quantitative automatic analysis tools is it possible to gather evidence to firmly support biological findings. However, cell mobility is still mostly analyzed by visual inspection alone, which is time consuming and prone to induce subjective bias. This makes automatic cell mobility analysis essential for large scale objective studies.
To perform the automatic analysis of cancer cell mobility assays a method to estimate the cell position through time is required. Cell’s positions are normally estimated using object tracking techniques: tracking by detection-association, state modeling tracking through the use of techniques such as the Kalman filter or through the use of Particle filters. In most cases the referred techniques are based on the identification of cells prior to tracking using image segmentation techniques such as automatic thresholding, watershed segmentation, level sets and graph cuts. More recently the use of local interest point detectors for this task has also been proposed with good results.
In the scope of the application of mobility analysis of cancer cells in brightfield microscopy we applied several computer vision automatic object detection and tracking techniques with the aim of testing and comparing the viability of such methods for laboratory use.
In order to improve the performance of the automatic tracking of cells, we propose the incorporation of cell shape information during the tracking process, with the aim of exploring the influence of cell shape on the directionality of cell motion. With such aim, we developed a Particle Filter based joint tracker for cell shape and mobility, which uses the relationship between the two properties of cells. This proposed approach is explained in detail in this thesis and compared with the current techniques described in the state of the art.
In order to allow biology researchers to use our work in their experimental analysis tasks, we developed an easy to use software to perform cell detection and tracking called Mobility-Analyser. This software enables easy manual or fully automatic cell tracking and performs cell mobility analysis through the computation of several measures important to characterize the cell mobility in each specific in vitro assay. Another software called BacteriaMobilityQuant was developed for the tracking of bacteria in order to categorize their behaviour under different light conditions. This is also a user-friendly application that automatically quantifies bacteria mobility in recorded time lapse videos. We have also developed a software for background patterned surfaces removal from cell brightfield images. This tool automatically detects the existing background pattern allowing for its removal from the original image which enables the tracking and mobility analysis of cells without background interference of the background.
Finally we have participated in the 3rd edition of the Cell tracking challenge organized in the scope of the ISBI’15 conference, where we had the opportunity to test some of the proposed cell detection and tracking approaches in multiple microscopy image types. Among several participants we ranked 4th and we were the only team to submit results for all the different datasets (different cell types and image modalities).
Keywords: Cell segmentation, shape estimation, local image filtering, local convergence filters, cell tracking, object state modeling