Authors:
Pedro Canelas
1
;
Leonardo Martins
1
;
André Mora
1
;
Andre S. Ribeiro
2
and
José Fonseca
1
Affiliations:
1
Faculdade de Ciências e Tecnologia and Universidade Nova de Lisboa, Portugal
;
2
Tampere University of Technology, Finland
Keyword(s):
Microscopy, Synthetic Time-lapse Image Simulation, Cell Tracking, Cluster Tracking.
Related
Ontology
Subjects/Areas/Topics:
Computer Simulation Techniques
;
Formal Methods
;
Simulation and Modeling
;
Simulation Tools and Platforms
;
Stochastic Modeling and Simulation
Abstract:
Several major advances in Cell and Molecular Biology have been made possible by recent advances in live-cell microscopy imaging. To support these efforts, automated image analysis methods such as cell segmentation and tracking during a time-series analysis are needed. To this aim, one important step is the validation of such image processing methods. Ideally, the “ground truth” should be known, which is possible only by manually labelling images or in artificially produced images. To simulate artificial images, we have developed a platform for simulating biologically inspired objects, which generates bodies with various morphologies and kinetics and, that can aggregate to form clusters. Using this platform, we tested and compared four tracking algorithms: Simple Nearest-Neighbour (NN), NN with Morphology and two DBSCAN-based methods. We show that Simple NN works well for small object velocities, while the others perform better on higher velocities and when clustering occurs. Our new
platform for generating new benchmark images to test image analysis algorithms is openly available at (http://griduni.uninova.pt/Clustergen/ClusterGen_v1.0.zip).
(More)