Machine Vision and Applications (2012) 23:603–605
DOI 10.1007/s00138-012-0436-2
EDITORIAL
Special issue on microscopy image analysis for biomedical
applications
Stephen J. McKenna · Derek Magee · Nasir M. Rajpoot
Published online: 8 June 2012
© Springer-Verlag 2012
Demand for tools that extract quantitative information from
microscopy images of biological samples continues to grow.
This is providing interesting challenges for computer vision
researchers. Typical tasks are detection, segmentation, clas-
sification, motion analysis, and tracking of cells and subcel-
lular compartments. Image data are acquired using various
imaging technologies, each with its own characteristics. In
the six papers presented in this special issue these include
brightfield [1], phase contrast [7], electron [5], and fluores-
cence [2, 3, 6] microscopy. There is also considerable vari-
ation in the biological samples to be analysed. Here these
include plant roots, cervical cancer cells and neuronal den-
drites, for example.
These papers were selected and revised from submis-
sions received in response to an open call for papers. Three
of them [1, 5, 6] were developed from presentations at the
symposium on microscopy image analysis for biomedical
applications that we chaired in April 2010. The meeting
was organised by the British Machine Vision Association
(BMVA) and held at the British Computer Society in Lon-
don. A further paper based on a talk given at that meeting,
on nuclei detection using support vector machines, has also
been published recently in this journal [4].
S. J. McKenna (B )
School of Computing, University of Dundee,
Dundee DD1 4HN, UK
e-mail: stephen@computing.dundee.ac.uk
D. Magee
School of Computing, University of Leeds, Leeds LS2 9JT, UK
e-mail: d.r.magee@leeds.ac.uk
N. M. Rajpoot
Department of Computer Science, University of Warwick,
Coventry CV4 7AL, UK
e-mail: n.m.rajpoot@warwick.ac.uk
Automatic analysis of adherent mammalian cells in bright-
field images is a challenging task, not least because of their
low contrast. It is a problem that has arguably received rel-
atively little attention in the literature. Ali et al. [1] present
methods for detection and segmentation of adherent cells in
brightfield images using phase and texture features. As a side
benefit, their framework allows registration of cells in bright-
field and fluorescence images. They evaluate their methods
by comparison to manual and fluorescence-based segmen-
tation. Their software sephaCe is open-source and has been
made freely available.
Esteves et al. [3] also address detection and segmentation
but in a rather different setting: analysis of cell nuclei in con-
focal fluorescence images of Arabidopsis thaliana root tips.
They demonstrate that an improved gradient convergence fil-
ter method allows shape priors to be imposed on this largely
bottom-up process. This improves nucleus detection in sce-
narios where cells appear overlapping or in close proximity.
Sethuraman et al. [6] also analyse confocal fluorescence
images of A. thaliana but are concerned with tracking cell
membranes rather than segmenting nuclei. These membranes
appear in a given two-dimensional slice as a network struc-
ture. Their contribution is to propose a Markov chain Monte
Carlo sampling scheme for tracking this structure using net-
work snakes.
Theriault et al. [7] segment, classify and track mouse fibro-
blast cells imaged using phase-contrast microscopy. They
employ a battery of standard shape and appearance-based
features to characterise their morphological state. They use
AdaBoost to classify each cell instance into one of eight clas-
ses based on spread, polarization and orientedness attributes.
They also show that the performance of their simple tracking
algorithm increases as they use the classification algorithm to
filter out substrate clutter. Expert-labeled datasets for time-
lapse phase-contrast microscopy images of cells are scarce,
123