8 A Cellular Neural Network based system for cell counting in culture of biological cells - Proceedings of the 1998 IEEE International Conference on Control Applications - Trieste, Italy 1-4 September 1998
A Cellular Neural Network based system for cell counting in culture of biological cells - Proceedings of the 1998 IEEE International Conference on Control Applications, Trieste (Italy) 1-4 September 1998, Vol. 1, pp. 341-345.
di L. Bertucco, G. Nunnari, C. Randieri
Abstract
Cell counting methods are important tools in molecular biology as well as clinical medicine. It is not always technically possible to measure quantitatively the events of cellular growth and fission. When it can be done, the procedures are neither so simple nor without excessive tedium as to lend themselves practically to the necessary replication of observations with large number of individual cells. In this paper, we describe a CNN based system that uses a CNN simulator for counting cells. The performances of the proposed system are illustrated by a simple cell counting experiment using a Petroff- Hauser based counter system.
Semelhante a 8 A Cellular Neural Network based system for cell counting in culture of biological cells - Proceedings of the 1998 IEEE International Conference on Control Applications - Trieste, Italy 1-4 September 1998
Automatic food bio-hazard detection system IJECEIAES
Semelhante a 8 A Cellular Neural Network based system for cell counting in culture of biological cells - Proceedings of the 1998 IEEE International Conference on Control Applications - Trieste, Italy 1-4 September 1998 (20)
VIRUSES structure and classification ppt by Dr.Prince C P
8 A Cellular Neural Network based system for cell counting in culture of biological cells - Proceedings of the 1998 IEEE International Conference on Control Applications - Trieste, Italy 1-4 September 1998
1. Proceedings of the 1998 IEEE
International Conference on Control Applications
Trieste, Italy 1-4 September 1998 WPO2
A Cellular Neural Network based system for cell counting
in culture of biological cells
Bertucco L., Nunnari G., Randieri C.,
Dipartimento Elettrico, Elettronico e Sistemistico
Universita di Catania, Viale A. Doria, 6, 95125
Tel. +95-339535, Fax +95 330793, e-mail : gnunnari@dees.unict.it
Rizza V.
Instituto di Biochimica Universitk di Catania
Viale A. Doria, 6, 95 125
Sacco A.
ABRES S.r.L. (Associated Biotechnology Research)
Abstract
Cell counting methods are important tools in molecular
biology as well as clinical medicine. It is not always
technically possible to measure quantitatively the
events of cellular growth and fission. When it can
be done, the procedures are neither so simple nor
without excessive tedium as to lend themselves
practically to the necessary replication of observations
with large number of individual cells.
In this paper, we describe a CNN based system that
uses a CNN simulator for counting cells. The
performances of the proposed system are illustrated
by a simple ceil counting experiment using a Petroff-
Hauser based counter system.
1. Introduction
Advancements in cell culture techniques often rely on
rapid and accurate methodologies for evaluating
cellular multiplication rate. Hence, in unanticipated
ways cell count is of great value to know whether the
variables under study do or do not affect growth and
multiplication in the same way. The application of
Cellular Neuronal Networks (CNNs) based
methodology for counting cells suggests a novel
approach for implementing techniques in molecular
biology. Particularly relevant to such application is the
fact that at a given instant of time, large number of
cells can be rapidly counted. Hence, this feature would
permit the comparison of representative populations of
cells at the same or different stages of individual
development. Repeated cell counts as a whole of
sufficiently large populations facilitates setting up
calibration curves correlating ceil number to increases
in some parameter of protoplasmic mass such as cell
nitrogen, nucleic acids content, cell volume etc. As a
0-7803-4104-X/98/$10.00 01998 IEEE 341
result, development of standard curves would provide a
simultaneous quantitative estimation of protoplasmic
mass or some constituent of it and cell number. In
addition it would give the investigator a greater return
in terms of effort and time for evaluating parameters
influencing the multiplication rate of biological cells.
From the consideration reported above, it appears
interesting to study and develop an on-line cell counter
having a fast image processing system for estimating
cell number. Cellular Neural Networks (CNNs),
proposed by L. 0. Chua (1988) [l-2], has been
demonstrated to be a promising tool for the
implementation of very fast image processing systems
since several semiconductor manufactures have planned
to produce CNN based chips in a few years [3].
The first part of this paper considers the cell counting
methods commonly used to quantify bacterial cells in
culture. The second part deals with the automation of
the counting methods by employing appropriate tools
familiar to Cellular Neural Networks (CNN).
2. Preliminaries about CNNs
CNNs are essentially non linear analog electric circuits,
locally interconnected for distributed computation [I].
CNNs consist of several identical computation units
called “cells”, all directly or indirectly connected, and
possess some of the salient characteristics of neural
networks, such as the capability of processing data in
parallel. Analytically, a two-dimension MxN CNN can
be considered as a matrix whose elements are
computation cells. CNNs are analog in nature and
intrinsically operate in the parallel mode, thus allowing
a real time image processing and overcoming the two
drawbacks mentioned above. CNN are also easily
programmable by appropriately choosing the
BACK TO PROGRAM SESSION
2. coefftcients of three matrices called “templates” [2]
Unfortunately, at present cellular neural networks only
exist on a prototype level; in addition, the dimensions
of such prototypes are no longer than 32x32 [3] or
64x64 (L. Chua, Personal Communication). In the
future CNN architectures with larger dimensions will
be implemented. It will also be possible to interface
CNN with cameras by using appropriate optic sensors.
This will allow image to be processed directly in analog
form, thus avoiding previous A/D image conversion.
Image pixels will also be have processed
simultaneously (as the CNN will be the same size as the
processed image).
At the present, to overcome the drawback mentioned
above, a CNN simulator has been implemented referred
ad SimulCN2 written in “C” language [4].
Simultaneous constructions of the dynamic evolution of
CNN ( bi- or multidimensional) may be developed by
integration of equations of state of the cells in the
networks according to the method of Runge-Kutta and
Euler0 [5].
The routines have been optimized for guaranteeing a
greater resources while minimizing the computation
load. These characteristics are also suitable for
application on less powerful instruments including
personal computers.
3. Traditional Cell Counting Methods
While estimates can be made of the stage of growth of a
culture from its appearance under the microscope,
standardization of culture conditions and proper
quantitative experiments are difhcult unless the cells
are counted before and after, and preferably during,
each experiment [6].
The multiplication of a bacterial cells can be measured
in time by counting the increase in the number of cells,
and it is o&n referred to the specific growth rate of the
culture. There are a number of methods which can be
employed for measuring cell count; the method of
choice often depends on the microbial cell size and the
specific features one wishes to investigate [7]. In this
study, we will treat a method which counts cells
directly using an appropriate counting chamber and
visualizes the cells by optical microscopy. Other
methods for counting cells are available commercially
and make use of electronic particle counting.
Although a number of different automatic methods
have been developed for the counting of cells in
suspension, the system devised originally by Coulter
Electronics is the one most widely used. Briefly, cells
drawn through a fine orifice change the current flow
through the orifice, producing a series of pulses that
are sorted and counted. by use conductometric methods.
Sample preparation for use in these methods are often
laborious in that the cell suspension needs to be
repeatedly centrifuged to eliminate interference from
particles and cell debris suspended in solution.
4. Petroff-Hauser or Helber Bacteria Counters or
Hemocytometers (Improved Neubauer)
The principle of their use is the determination of
bacteria (Pen-off-Hauser or Helber Bacteria Counters)
or red blood cells (Hemocytometers) in the fluid volume
above calibrated ruled areas etched into glass slides.
Since the depth of the fluid above the ruled area is
known, the average number of organisms per unit
volume may be calculated. In the Petroff-Hauser
counter the depth of fluid is 0.02 mm and the ruled
areas are l/400 square mm2, Each bacterium in the
volume of fluid above one square thus represents
20,000,OOO organisms per ml of culture. Obviously the
advantages of this counting chambers is that they can
be used for estimating large numbers of organisms.
Moreover, if appropriately sealed, they can be used to
estimate bacterial growth rates under a specific set of
conditions.
In the hemocytometer, the concentration of a cell
suspension may be determined by placing the cells in
an optically flat chamber under an optical microscope
(Fig. 1). The cell number within a defined area of
known depth is counted and the cell concentration
derived from the count.
3 Co)
Fig. 1: Hemocytometer chamber: (a) Adding cell
suspension to the assembled slide; (b) Longitudinal
section of slide showing position of cell sample in
0.lmm deep chamber and.top view of slide with
magnified view of total area of grid.
5. Analysis
The number of cells/ml in a sample is calculated as
the average of the two counts as follows:
where:
C=n/v
342
BACK TO PROGRAM SESSION
3. c = cells concentration (cells/ml),
n = number of cells counted,
v = volume counted (ml).
For the Improved Neubauer slide, the depth of the
chamber is 0.1 mm , and assuming only the central I
mm’ is used, v is 0. I mm ’ or 10e4ml.
hence:
C=n.104
Hemocytometer counting is cost effective and gives a
direct observation of the cells counted. In addition, if
the cells are mixed previously with an equal volume of
a viability stain (see below and fig. 2), a viability
determination may be performed at the same time.
simulator, can furnish output images having the same
basic shape. The operational sequences describing the
basic features of the algorithm for counting will be
illustrated in the following paragraph.
Consider a random assortment of shapes and forms that
conceivably might depict a biological cell. A
representative input image “Cells.bmp” 70~70x256 is
shown in Fig. 4(a) and illustrates 14 different computer
generated shapes contained in a unit cell or
compartment.
Subsequently, the shapes in the image are converted in
black and white (2 level of gray) as shown in Fig. 4(b).
This conversion is obtained through the use of
“threshold templates” [S] and the resulting image is
subjected to further elaboration which include:
G-4 (b)
Fig. 2:
(a)Low-power (I OX objective) microphotograph
showing 20 of the 25 smaller squares of a slide loaded
with cells pretreated with naphthalene black (amid0
black). Viable cells are unstained and clear with a
rtzfractile ring around them; nonviable cells are dark
and have no r<fractile ring.
(b)High-power (40X objective) microphotograph of one
?f the smaller squares, bounded by three parallel lines
and containing 16 of the smallest squares. The distance
between each set Qf triple lines is 200 pm.
These counting procedures are, however, rather slow
and prone to error both in the method of sampling and
the size of samples and requires a minimum of 10”
cells/ml.
Most of the errors occur by incorrect sampling and
transfer of cells to the chamber. In addition, other
sampling errors encountered are improper mixing, cell
adherence to the sides of glass test tube, sedimentation
on standing or cell clumping.
6. Cell counting by CNN
In this section, we will describe briefly the CNN based
algorithm proposed for counting cells. A block diagram
of the algorithm is illustrated in Fig. 3. The results
obtained with the developed image processing software
will be illustrated with a test run.
The implemented software makes use of input images
distinguishable by 256 gray levels and, with the aid of a
Bitmap ty 256 gray)
.
And Template
I
Fig 3: Block diagram representing the algorithm,for
counting.
Fig. 4: (a) ‘%ells. bmp “‘before threshold, (6)
“Cells. bmp ”qjier threshold
l “Brimming or filling in”. This operation would
identify interesting objects: bacterial cells which might
343
BACK TO PROGRAM SESSION
5. l The numerical difference obtained between the black
pixels in image “Al” (Fig. 5b) and “A2” (Fig. 8b)
corresponds to the number of objects originally
present in the starting image, i.e. 15-I = 14.
It should be mentioned that the algorithm, in its
present stage of development, cannot distinguish two
superimposed objects and hence would consider them
as one object having the same gray level. Thus, such
shortcomings in the model can be corrected through
statistical averages obtained after repeated counts of
samples containing a large number of cells.
Other features which perhaps can cause disturbances in
the counting model might be due noise levels present in
the original frame (Fig. 9a). In order to prevent such
inconsistencies, it is necessary to work under optimum
conditions which assure the maintenance of clean
frames before proceeding with the subsequent steps in
the process. Alternatively, a CNN based model utilizing
appropriate template libraries, i. e. “Figextr” can be
used for “cleansing” the frame.
Fig.9a to 9c illustrates the “cleansing” steps for
“filtering” an image from particles which do not match
or conform to shapes familiar to biological cells. These
steps include a threshold frame represented in Fig. 9b
and the frame 9c processed by using the “Figextr”
templates [S]. The latter can be used subsequently for
counting the cells present in the representative image.
a CNN based chip. The latter when interfaced with
video sensors and software for direct analog image
processing can reduce the scanning speed to the order
of microseconds.
8. References
[I] L. 0. Chua and L. Yang, Cellular Neural Networks:
Theory, in IEEE Transactions on Circuits and Systems,
~01.35, no, IO, October 1988.
[2] K. R. Crounse and L. 0. Chua, Method.s,for Image
Processing and Pattern Formation in Cellular Neural
Networks: a Tutorial, in IEEE Transactions on Circuits
and Systems: Fundamental Theory and Applications,
~01.42, October 1995.
[3] A. Rodriguez-Castro, S.Espejo, R. Dominguez-
Castro, R. Carmona, A CNN Universal Chip in CMOS
technology, Third IEEE Int. Workshop on Cellular
Neural Networks and their Applications, Rome, Italy,
Dec. 1994, pp. 9 I-96.
[4] G. Nunnari, L. Occhipinti, L. Bertucco, A cellular
neural network simulator,fbr on-line imuge processing.,
An application to monitoring uctive volcanoes,
accepted to 15th IMACS World Congress, Vol. 4, pp.
215-220, Berlin, 1997.
[5] J. R. Rice, Numerical Methods, Sojtware und
Analysis, 1992.
[6] T. D. Brock el al., Biology qfMicroorgunism, 1994
Fig. 9: (a) noise level in imuge, (b) Threshold,
(c) Figextr
7. Conclusions
The innovative aspect of the application presented in
this paper is the use of CNN for cells counting
operation and in particular the speed at which the
operation can be performed which is independent of the
complexity of the image or the number of objects to be
counted. The aspects mentioned above furnish greater
overall versatility for CNN based counting
methodologies as compared to conventional techniques.
In a simulated process such as the one we have
described, the time required for explicating a count is
in the order of few minutes. This time estimate is
purely indicative and is based on processing images
70x70 and utilizing a Pentium Processor 200. However,
further considerations on the merits of this
methodology depend on the commercial availability of
[7] R. Ian Freshney, Culture qfanimal cells, 1994
[S] Computer and Automation institute Hungarian
Academy of Sciences, CNN analog (dual) sqftware
library, in Dual and Neural Computing System
Laboratory, Budapest, December 1992.
345
BACK TO PROGRAM SESSION