AUTOMATED DETECTION OF INTERPHASE AND METAPHASE

NUCLEI IN THE FISH IMAGES

Jan Schier

, Bohumil Kov

1,2

and Eduard Koc

arek

Institute of Information Theory and Automation of the ASCR

Pod vod

arenskou ve

ı 4, Prague, Czech Republic

Faculty of Transportation Sciences, Department of Applied Mathematics, Czech Technical University in Prague

Na Florenci 25, Prague, Czech Republic

Department of Biology and Medical Genetics, 2nd Faculty of Medicine, Charles University Prague

Uvalu 84, Prague, Czech Republic

Keywords:

FISH analysis, Fluorescence microscopy, Image analysis, Image segmentation.

Abstract:

The ﬂuorescence in-situ hybridization (FISH) belongs to the most common cytogenetic methods and is widely

applied in routine clinical genetic diagnostics. We are paying attention to FISH analysis of chromosomal ane-

uploidies – the deviations from chromosomal number. Such analysis is based on evaluation of up to several

hundreds of microscopic images. Computer support for this process includes using methods of image process-

ing and data mining. In this paper, we focus on the image processing part in more detail: ﬁrst, the properties

of FISH images are reviewed, then, the processing ﬂow is outlined. Our aim is to ﬁnd the interphase and

metaphase nuclei and the hybridization signals contained in the image. A simple method using the raw and

central moments of detected objects as measures to distinguish between the two types of nuclei is proposed.

1 INTRODUCTION

Fluorescence in situ Hybridization (FISH) is a tech-

nique used to visualize the location of speciﬁc DNA

sequences within the nucleus. It belongs to the most

common of molecular cytogenetic methods and is

widely applied in routine clinical genetic diagnostics.

The FISH technique concerns application of a ﬂu-

orescently labeled DNA or RNA probe onto a biolog-

ical sample containing the target DNA or RNA from

the patient (e.g. as tissue sections, separated cells,

isolated interphase nuclei or chromosomal metaphase

spreads). In the process of hybridization, the probe

attaches to its complementary sequences on the target

DNA or RNA molecule. Under the ﬂuorescence mi-

croscope we observe the sequence with the attached

probe as a strikingly luminous hybridization signal (a

ﬂuorescent dot or an area within the biological struc-

ture). Details of the principles of the ﬂuorescence mi-

croscopy can be found e.g. in (Rittscher et al., 2008)

FISH is also a useful diagnostic tool in repro-

ductive medicine, oncology, preventive medicine, and

other medical branches – examples can be found in

(Beatty et al., 2002) or (Varella-Garcia, 2003).

We are paying attention to FISH analysis of chro-

mosomal aneuploidies – i.e., deviations from chromo-

somal number. These abnormalities are responsible

for well-known clinically signiﬁcant syndromes such

as Down’s syndrome (trisomy 21) or Turner’s syn-

drome (monosomy X), Klinefeter’s syndrome (XXY

constitution) – see (Harper, 2004), (Roizen and Pat-

terson, 2003), (Anner

en, 2007), and (Visootsak and

Graham, 2006).

To be able to conclude examination of a single pa-

tient, a set of tens or hundreds of images have to be

evaluated. An automated data acquisition and pro-

cessing is needed to achieve acceptable accuracy and

reproducibility of results; manual processing would

be time consuming and subjective.

Our paper is focused on the task of image segmen-

tation, that is, on ﬁnding and evaluation of the objects

of interest — the interphase cell nuclei, metaphase

spreads and the hybridization signals. We ﬁrst re-

view the properties of images resulting from the FISH

analysis in Section 2 and propose an image process-

ing system in Section 3. In Section 4, we discuss the

edge-based segmentation technique used for identi-

ﬁcation of cell nuclei and detection of hybridization

signals. In Section 5, we describe the experiments

and, ﬁnally, Section 6 concludes the paper.

347

Schier J., Kovár B. and Kocárek E..

AUTOMATED DETECTION OF INTERPHASE AND METAPHASE NUCLEI IN THE FISH IMAGES.

DOI: 10.5220/0003797703470350

In Proceedings of the International Conference on Bioinformatics Models, Methods and Algorithms (BIOINFORMATICS-2012), pages 347-350

ISBN: 978-989-8425-90-4

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 FISH IMAGES

Example images from the FISH analysis are shown

in Figure 1. They show the interphase nucleus (Fig-

ure 1a) and the metaphase spread of chromosomes

(Figure 1b), which correspond with two basic phases

of the so-called cell cycle.

(a) Interphase cells

(b) Metaphase spread

Note: the images are not in the same scale. The metaphase cell

image has been preprocessed to suppress the background.

Figure 1: Examples of FISH images.

The hybridization signals are — in this exam-

ple — in red and green colour. Blue objects are

the nuclear structures counterstained with the DAPI

dye. Very often, both the interphase nuclei and the

metaphase spreads are contained in the same image.

The objects of interests are characterized by the

following features: the nuclei are of roughly round

shape; the spread of chromosomes has ideally a

roughly circular envelope. The chromosomes selves

are observed as X-shaped or V-shaped bent stripes,

frequently overlapping one another in the image. The

hybridization signals must be located either within the

interphase nuclei, or on a chromosome. The signals

within a single nucleus should be clearly separated.

For detection of the chromosomal aneuploidies

(cf. the Introduction section), it is important to evalu-

ate the quality of the nuclei and hybridization signals.

The rules for evaluation should be rather stringent and

conservative, to eliminate dubious objects and objects

Figure 2: Interphase nucleus of an irregular shape as an ex-

ample of object to be rejected from processing.

in ambiguous arrangement.

Examples of objects to be rejected from process-

ing include (see Figure 2): interphase nuclei of irreg-

ular shape, nuclei touching or overlapping each other,

metaphase spreads of signiﬁcantly irregular shape,

signals located in the background, etc.

For each dye, separate greyscale image is taken

using ﬁlter set matching the excitation wavelength

of the illuminating light and the wavelength of the

light emitted by this that dye (details can be found

in (Rittscher et al., 2008), Chapter 2). The resulting

image is then computed from these greyscale images.

3 IMAGE PROCESSING FLOW

The processing ﬂow of the image processing system is

outlined in Figure 3, for blue-stained cells and green

(or green and red) signals.

While the operations used to process the cell com-

ponent and the signal components of the input image

are very similar, it is advantageous to process each

image, corresponding with a single ﬁlter set, sepa-

rately. The reason is that the cells and the signals of

each colour are in separate layers. The ﬁrst operation

is segmentation, used to separate the objects of inter-

est from the background. It is followed by computa-

tion of object parameters (size, roundness, brightness,

etc.), which, in turn, are used to sort out objects that

satisfy the quality criteria. The results of cell process-

ing are used to eliminate any signals located in the

background.

4 IMAGE SEGMENTATION

4.1 Detection of Cell Nuclei

4.1.1 Otsu Thresholding Algorithm

For image binarization, we have chosen the Otsu algo-

BIOINFORMATICS 2012 - International Conference on Bioinformatics Models, Methods and Algorithms

348

Blue channel

Processing of nuclei

Green (red) channel

Processing of signals

Object

parameters

Segment.

Object

Classiﬁcation

Signal

parameters

Image Segment.

Signal

Classiﬁcation

Masking

Interphase

cells

Metaphase

spread

Output

Figure 3: Processing ﬂow of the image processing subsystem.

Figure 4: Results of the Otsu-based image binarization.

rithm, introduced in 1979 by Nobuyuki Otsu (Otsu,

1979). Using discriminant analysis, Otsu deﬁned the

between-class variance of the thresholded image as

(t) = ω

(t)σ

(t) + ω

(t)σ

(t), (1)

where ω

are the probabilities of the two classes sep-

arated by a threshold t and σ

are variances of these

classes. For bi-level thresholding, Otsu veriﬁed that

the optimal threshold t

∗

is chosen so that the between-

class variance σ

(t) is maximized; that is,

∗

= argmax(σ

(t)), 1 ≤ t ≤ L, (2)

where L is the highest gray-level intensity value in the

image.

4.1.2 Parameters of Cell Nuclei

As the next step, the eccentricity e of each object is

computed, using its contour C. It is deﬁned as the

relative difference in magnitude of the eigenvalues λ

and λ

(4) of the covariance matrix of the object con-

tour C.

e =

1 −

, (3)

1,2

+ µ

4µ

+ (µ

− µ

)

. (4)

In this equation µ

are second order central moments

deﬁned as

= µ

/µ

where µ

is the central moment of order pq deﬁned

∑

(x − ¯x)

(y − ¯y)

C(x, y). (5)

Only the nuclei with circular or slightly elliptical

shape (e > 0.6) are accepted for further processing.

In addition, for each cell nucleus {i}, we evaluate

further characteristics, such as the area, centroid, av-

erage intensity, standard deviation of intensity, round-

ness, length of contour, etc.

4.2 Detection of Hybridization Signals

Hybridization signals are detected by the same algo-

rithm as the nuclei, but used on other layers of the in-

put image. The fact that it is not necessary to change

the segmentation algorithm to process the signals can

be considered as an advantage of this approach. Illus-

trative results are shown in the Figure 5.

AUTOMATED DETECTION OF INTERPHASE AND METAPHASE NUCLEI IN THE FISH IMAGES

349

Figure 5: Detected cell nuclei with hybridization signals.

Source image is depicted in Figure 1a. Note that some of

the detected signals do not correspond to the real signals in

the sample.

5 DESCRIPTION OF TEST DATA

AND EXPERIMENTS

To test the efﬁciency of the proposed approach, ﬁfty

microscopic images acquired at a resolution of 720 ×

576 pixels, containing both the interphase nuclei and

the metaphase spreads of chromosomes, were used.

The number of nuclei in each image ranges between

1 and 10. Most images contain metaphase spread

of chromosomes and at least one interphase nucleus.

This data set was divided into two equal parts, each

of them with 25 images. First data set was used

for nearest neighbor classiﬁer training while the sec-

ond data set for the experiments. The training set

for the classiﬁer consisted of 73 nuclei used for fea-

ture reduction. All nuclei were labeled by the fea-

tures that are described in the Section 4. The fea-

tures that perform well to separate both classes (inter-

phase and metaphase), were selected manually. We

have achieved the best results if contour eccentricity

and length and object area were used. The classiﬁer

correctly identiﬁed nearly all (99%) interphase cells.

Incomplete cells located at the image border were au-

tomatically rejected from further evaluation. Unfor-

tunately, only 68% nuclei in the metaphase state were

identiﬁed correctly.

The algorithm is sufﬁciently robust to detect com-

pact spreads, such as shown in Figure 1b, but the per-

formance decreases when the spread is not compact

enough.

6 CONCLUSIONS

The paper treats automated segmentation and eval-

uation of images from the ﬂuorescence in-situ hy-

bridization (FISH) microscopic analysis.

Two problems were discussed: segmentation of

the input image to cell nuclei, hybridization signals

and background, and recognition of interphase nuclei

and metaphase spreads. To distinguish between the

two types of nuclei, eccentricity, area, and perime-

ter are used as the main criteria. While this approach

is adequate for identiﬁcation of interphase nuclei, it

does not provide acceptable results for the metaphase

spreads. This problem will be addressed in further

research.

ACKNOWLEDGEMENTS

The paper has received support from the TA01010931

project System for support of the FISH method evalu-

ation of the Technology Agency of the Czech Repub-

lic.

REFERENCES

Anner

en, G. (2007). Turner syndrome. The American Jour-

nal of Human Genetics, 68(3):2–9. John Wiley &

Sons, Ltd.

Beatty, B., Mai, S., and Squire, J. (2002). FISH. Oxford

University Press, Oxford.

Harper, P. S. (2004). Practical genetic counselling. Arnold,

London, 6 edition.

Otsu, N. (1979). A threshold selection method from gray-

level histograms. IEEE Transactions on Systems,

Man, and Cybernetics, 9:62–66.

Rittscher, J., Machiraju, R., and Wong, S. T. C., editors

(2008). Microscopic Image Analysis for Life Science

Applications. Artech House, Inc.

Roizen, N. and Patterson, D. (2003). Down’s syndrome.

The Lancet, 361(9365):1281–1289.

Varella-Garcia, M. (2003). Molecular cytogenetics in solid

tumors: Laboratorial tool for diagnosis, prognosis,

and therapy. The Oncologist, 8(3):45–48.

Visootsak, J. and Graham, J. (2006). Klinefelter syndrome

and other sex chromosomal aneuploidies. Orphanet

Journal of Rare Diseases, 1(42). Biomed Central.

BIOINFORMATICS 2012 - International Conference on Bioinformatics Models, Methods and Algorithms

350