Skip to main content Skip to navigation
BioImage Informatics Conference 2017

The ABCs of Rigorous Bioimage Analysis: Algorithms, Benchmarks, Challenges

Michal Kozubek, Centre for Biomedical Image Analysis, Faculty of Informatics, Masaryk University, Brno, Czech Republic

Similar to the medical imaging community, the bioimaging community has recently realized the need to benchmark various image analysis methods to compare their performance and assess their suitability for specific applications [1]. In other words, there has appeared the need to sort out the big pile of bioimage analysis algorithms developed so far and move from “any algorithm” to “the best algorithm” thinking. The common way of checking the performance of image analysis algorithms is to use benchmark image datasets of known properties and evaluate various analysis methods on the same data. Unfortunately, in biological imaging community there has been a lack of publicly available reference images for a long time, both simulated with precisely known ground truth as well as real ones annotated by experts.

This presentation summarizes the recent progress in this respect and reviews available benchmarks as well as challenges that foster the algorithm development and benchmarking by engaging the community in the form of competition, collaboration or both. Attention will be paid to the associated problems and solutions, including proper selection of reference image datasets for training and testing the methods, dealing with real data annotation problems, choosing suitable evaluation metrics and employing simulated image data in algorithm validation [2]. Finally, examples of biological applications will be shown for which benchmarking of relevant image analysis algorithms have already been performed.

References:

Kozubek M. Challenges and benchmarks in bioimage analysis. In: Focus on bio-image informatics. Advances Anatomy Embryology and Cell Biology, vol. 219, Chapter 9, pp. 231-262, Springer, 2016. doi:10.1007/978-3-319-28549-8_9.

Ulman V, Svoboda D, Nykter M, Kozubek M, Ruusuvuori P. Virtual cell imaging: A review on simulation methods employed in image cytometry. Cytometry Part A, vol. 89(12), pp. 1057-1072, 2016. doi:10.1002/cyto.a.23031.

 

Michal Kozubek is a professor of computer science at the Faculty of Informatics, Masaryk University. He is a head of the Center for Biomedical Image Analysis, and former Dean (2011-2015). His research focuses on image processing, especially as applied in microscopy for spatiotemporal cell studies. He has stayed at and collaborated with, for instance, University of Oxford (on the development of the first structured illumination microscope in 1995-1996), University of Heidelberg (on the development of computer-controlled cell tomography in 2000-2002) or University of Navarra (on computerized cancer cell imaging since 2006). Presently he mainly attends to the

development of benchmarks for the algorithms of cell segmentation and tracking including generation of synthetic cell image datasets. To this end, he has co-organized Cell Tracking Challenge as a platform for comparing cell tracking algorithms. He also participates in the integration efforts in the frame of pan-European research infrastructure for biomedical imaging Euro-BioImaging, its Czech branch Czech-BioImaging as well as COST NEUBIAS (Network of European Bioimage Analysts). He is a member of Czechoslovak Microscopy Society, European microscopy Society and IEEE Signal Processing Society.