AUTOMATIC DETECTION OF PALE PATH AND OVERLAPS

IN CHROMOSOME IMAGES USING ADAPTIVE SEARCH

TECHNIQUE AND RE-THRESHOLDING

Rajeev Ranjan

, Akila Subasinghe A.

, Jagath Samarabandu

, Peter K. Rogan

and Joan H. M. Knoll

Indian Institute of Technology, Electronics and Electrical Communication Engineering, Kharagpur, India

Image Recognition & Intelligent Systems Laboratory, Electrical and Computer Engineering,

University of Western Ontario, Ontario, CA, N6A 5B9, Canada

Departments of Biochemistry, Schulich School of Medicine & Dentistry,

University of Western Ontario, Ontario, CA, N6A 5C1, Canada

Departments of Pathology, Schulich School of Medicine & Dentistry,

University of Western Ontario, Ontario, CA, N6A 5C1, Canada

Keywords: Image Analysis, Segmentation, Chromosomes, Cut-points, Pale Path, Search Window, Thresholding.

Abstract: Detection and separation of overlapping and touching chromosomes is a critical issue in image analysis

applications for cytogenetics where accurately segmented chromosomes are essential. We present a novel

method of automatic pale path detection for all types of stained chromosome images. Optimum number of

cut-points for each cluster of touching or overlapping chromosomes is obtained and analysed sequentially

for the pale paths. A self-adaptive search window searches for the minimum grayscale intensity beginning

from the vicinity of a cut-point and propagates gradually till the end of pale path. Efficient image and area

thresholding restricts the faulty detection of touch or overlap in a chromosome cluster.

1 INTRODUCTION

Images of human metaphase cells show individual

chromosomes as discrete objects, each of which can

adopt varied and inconsistent morphologies. For

cytogenetic analysis, it is important that the

chromosomes in an image be analyzed individually

in order to detect structural abnormalities

responsible for congenital diseases or cancer. During

preparation, chromosomes may overlap or touch one

another. The degree to which they coincide with one

another may significantly vary from cell to cell. The

process of segmenting images into individual

chromosomes involves defining the boundaries of

chromosomes. This procedure needs to be both

accurate and comprehensive in order to detect

chromosome abnormalities.

In order to analyse the structure of individual

chromosomes, a digital cell image has to be

segmented accurately. The segmentation process can

be challenging mainly due to intensity dispersions

and overlapping between chromosomal bodies.

Detecting and separating touching and overlapping

chromosomes in a digital cell image will improve

segmentation accuracy significantly.

Manual detection and separation of occluded

chromosomes in digital images can be very time

consuming. Early work in automating this process

included applying a variety of segmentation methods

like heuristic edge-linking, region growing, fuzzy-

logic, etc. A rule based chromosome segmentation

procedure (Liang, 1994) explores the possibility of

using a rule based classification and segmentation

parameter set for each cell. In the knowledge-based

chromosome contour searching method (Agam and

Dinstein, 1997) an edge-preserving smoothing

nonlinear filter is applied to remove random noise

while preserving the edges of chromosomes. A

classification driven partially occluded object

segmentation (CPOOS) method was proposed to

resolve partial occlusion in chromosome images

(Lerner, Guterman, and Dinstein, 1998). Minimum

entropy segmentation algorithm was also

investigated to decompose or separate groups of

chromosomes that touch and overlap each other

(Schwartzkopf, 2001). Automatic segmentation of

metaphase cells was also discussed based on global

context and variant analysis (Ritter and Gao, 2008).

Watershed algorithm also finds its utilisation in

segmentation of narrow-touching chromosomes

462

Ranjan R., Subasinghe A. A., Samarabandu J., K. Rogan P. and H. M. Knoll J..

AUTOMATIC DETECTION OF PALE PATH AND OVERLAPS IN CHROMOSOME IMAGES USING ADAPTIVE SEARCH TECHNIQUE AND

RE-THRESHOLDING.

DOI: 10.5220/0003862604620466

In Proceedings of the International Conference on Computer Vision Theory and Applications (VISAPP-2012), pages 462-466

ISBN: 978-989-8565-03-7

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

(Ming and Tian, 2010). Most of the procedures

discussed above cater only to a particular type of

banded chromosomes. They fail in analysing any

chromosome images that are stained in a different

manner than what the procedure demands.

In this paper, we propose an algorithm that

automatically separates the touching chromosomes

and detects overlapping chromosomes by optimum

thresholding and self-adaptive search window for

pale path detection. Unlike other methods, this

algorithm is applicable for all types of stained

chromosome images.

The rest of the paper has been organized as

follows. Section 2.1 provides an insight into the

classification of chromosome images into two basic

categories. Section 2.2 introduces the proposed

method of obtaining high concavity points using

moving average filter while section 2.3 presents the

proposed method of local thresholding for pale path

detection in case of touching chromosomes. A given

area constraint is also applied to prevent

segmentation of unwanted small groups of pixels

from a chromosome and is discussed in section 2.4.

2 PROPOSED ALGORITHM

2.1 Image Classification

The chromosome images can be broadly classified

into two groups: one with more of “cross-like”

chromosomes (in which sister chromosomes have

prematurely separated) and another in which the

sister chromosomes have coalesced into a single,

uninterrupted homologous pair.We apply an

automatic classification mechanism based on

counting branch-points of the binarized and thinned

image. This classification plays crucial role in

decision making of pale path (Section 2.3).

2.2 Determination of Cut-points

A window size of 200 x 200 pixels is manually

selected from nuclei free grayscale image containing

the object of our interest as shown in Figure 1.

To determine the points of concavity in the

contour, we use the exterior angle of boundary

pixels (calculated between adjacent pixels). Due to

the discreteness of the pixels, they belong to the set

of angles {π/2,3π/4, π, 5π/4 and 3π/2}

A high concavity point in the boundary layer

demands following characteristics.

• Its angle should be either 3π/4 or π/2

(a) (b)

Figure 1: (a) Binary image of the selected object. (b) The

single pixel thick contour developed from the given binary

image.

• The mean of the angles for a given set of

neighbourhood [x-N, x+N] must be less than π

(1)

• This mean obtained above should never exceed

π for all the sets of neighbourhood less than

(+N) range of that point.

(2)

To obtain the mean of the angles, a moving

average filter with a width of 3 is applied to the set

of angles.

Moving average filter is then applied with

iterating filter width initially set to 1. The iteration

stops if the mean angle value exceeds pi or when the

iterations reach N cycles. To meet all the 3

conditions specified above for a concave point, we

define two sets of boundary pixels X and Y. Set X

contains all the pixels having angles π/2 or 3π/4. Set

Y comprises of pixels with angles after application

of moving average filter being less than π. The

intersection of Set X and Set Y gives us the desired

high concavity points or cut-points.

2.3 Singularity/touch/overlap Detection

An object is classified as a single chromosome if the

number of cut-points (NCP) is less than 2 because

any overlap or touch requires at least two high

concavity points. If their number is more than 2, we

try to find a pale path originating from each cut-

points to judge whether the object contains touching

chromosomes (section 2.3.1). Once the pale path

succeeds in dividing the object, the sub-parts are

further analysed individually by finding out their

NCPs. If they exceed the value 2, we go for the

search of pale path in that sub-part until there are no

(

)

)(

∑

−

ixangle

(

)

,)(

NKallforixangle

<<+

∑

−

AUTOMATIC DETECTION OF PALE PATH AND OVERLAPS IN CHROMOSOME IMAGES USING ADAPTIVE

SEARCH TECHNIQUE AND RE-THRESHOLDING

463

more pale paths possible. Each newly sub-parts

formed are analysed in the same fashion till all the

cut-points are analysed for pale paths.

After analysing all the high concavity points, the

sub-parts are fed through morphological thinning

operation and again the X or T intersections are

checked. We already know whether an image

constitutes mainly of homologous chromosomes in

which sister-chromatid separation occurs (class A)

or does not (class B). In general, an X shaped

chromosome consists of 2 T-intersections or nodes

whereas a normal chromosome does not contain any

of them. Therefore, a sub-part with number of cut-

points greater than or equal to 2 is considered a

single chromosome, if it contains 2 or more skeleton

nodes for the image of class A or no node for the

image of class B. Any sub-part not complying with

this rule is tagged as overlap.

2.3.1 Pale Path Detection

We can categorize the cut-points into three different

types depending on their local convergent angles:

π/2, 3π/4 and π. The cut-point with convergent angle

π is resultant for close vicinity cut-points reduction.

All the cut-points are set to ‘0’ in the binary image.

For each of them, a 3x3 search is defined. Smallest

intensity within this window is then set to 0, which

is then used as the center of next iteration. The

iteration terminates only when the centre reaches the

boundary of object or the size of the search window

is reduced to 0.

An important point to note here is that the

iterative 3-pixel window search won’t always

provide the correct pale path. It will simply find a

path from one boundary end to another irrespective

of the intensity value traced. That may lead to

dissection of “cross-like” chromosome which would

result in false positive pale path. To overcome this

limitation, we classify the object further into high

intensity regions and pale regions. Only those pixels

which lie in the pale regions are searched for

minimum intensity in the 3-pixel window. If none of

the pixels of the window lie in the pale region, then

the search iteration is stopped, which indicates a

“cross-shaped” chromosome. The pale paths

detected for a group of 8 overlapping and touching

chromosomes is shown in Figure 2.

The classification into relatively high intensity

and pale regions is done by re-thresholding of the

object. The Otsu method applied to the object forms

a pale region that may contain the pale path as well

as a lot of the unwanted surrounding area. To avoid

this, we propose a novel method of classification

using the object’s image histogram.

It can be clearly seen that the pixels of grayscale

intensity less than the Otsu’s threshold plus an offset

can be subdivided into two regions: relatively high

intensity region comprising of chromosome bands

and relatively low regions comprising of pale paths.

In the envelope of the image histogram plot, we find

peaks corresponding to these two regions. Our aim is

to find the threshold around the minima to separate

the pale path region from the unwanted dark bands.

This is fulfilled by applying Intermeans Method on

the pixels less than the sum value of object’s Otsu’s

Threshold and the offset (set to 10) so that the search

window pixels are formed of grayscale intensity less

than the new threshold.

(a) (b)

Figure 2: (a) Binary image of a cluster of overlapping and

touching chromosomes. (b) The pale paths detected for

touching chromosomes.

For the case of class A type chromosomes, the

Intermeans method becomes inefficient because of

the large number pixels in the pale regions. To

compensate for this, Intermeans method is applied

only between the maxima of the two peaks of image

histogram. Hence, the reduced range of intensity

values increases the accuracy for this method.

The new threshold divides the object in a more

constricted pale region. A pixel in a search window

is applicable for search only if it lies in the pale

region obtained from the final threshold. When none

of the window pixels are a part of that region, the

search process stops.

2.4 Optimum Area Criterion

Sometimes, a small-sized extension of the

chromosome may split itself if it lies in the pale

region generated. To avoid such kind of anomalies,

we restrict the ‘pale-path splitting’ by an area

threshold. Only those sub-parts with a size more

than the area threshold can split to form a separate

chromosome. The area threshold is decided by

computing the mean and standard deviation (SD) of

the area of all the objects of binary image. Objects

with very large area are excluded, as they have a

VISAPP 2012 - International Conference on Computer Vision Theory and Applications

464

high chance of containing an overlap. The minimum

area of an object is also calculated and stored (A

min

We define an area threshold which is given by:

= Mean – (i/2)*SD, i ϵ I (3)

We increment ‘i’ starting from 1 and compute

successively. At each iteration, A

is compared

against A

min

. The iteration stops when A

becomes

less than this minimum area. The final threshold area

is chosen to be A

i-1

that is, the area threshold just

greater than A

min

(a)

(b)

(c)

(d)

Figure 3: Pale path detection in chromosomes, left:

grayscale image, centre: binary image, right: binary image

with pale paths. (a) Class A, non-banded. (b) Class A,

banded. (c) Class B, non-banded. (d) Class B, banded.

Hence, A

i-1

is taken as the minimum area

threshold. This area thresholding ensures prevention

of unnecessary splitting of very small pixel groups

in the pale region of object.

3 RESULTS AND DISCUSSION

The method discussed above was tested for

successful separations of touching chromosomes

using a variety of chromosome images. It was tested

with ‘cross-like’ chromosomes as well as smooth

curvature chromosome images in which it succeeded

in providing reliable results.

Pale path, touch and

overlap detection in a typical chromosome cluster of

‘Class A’ and ‘Class B’ images for both banded and

non-banded chromosomes is depicted in figure 3.

The proposed does not require any prior set of

knowledge, hence minimising the complexity. It also

requires minimum pre-processing and post-

processing techniques, hence further reducing

computational burden. Optimum area and intensity

thresholds restrict any plausible flaw that may arise

during pale path detection.

4 CONCLUSIONS

This paper introduces a simple and effective method

for pale path detection to separate touching

chromosomes in images of metaphase cells. A low

order finite impulse response moving average filter

is used to evaluate potential cut- points. Efficient re-

thresholding helps in discarding the unwanted search

regions, especially those which produce false paths

due to chromosome banding.

An analogous approach can be used for the

segmentation of overlapping chromosomes so that

they can also be analysed for valuable information.

This would involve the length ratio of the arms of

overlapping ‘class A’ chromosomes. Multiple

overlaps could also be separated by first computing

the number of overlaps and satisfying the arm length

ratios accordingly.

REFERENCES

Otsu, N., 1978. A threshold selection method from gray-

level histograms, IEEE Trans. Sys. Man. Cyber.,

SMC8 1277-1294.

Liang, J., 1994. “Fully automatic chromosome

segmentation,” Cytometry, vol. 17, pp. 196–208.

Agam, G., Dinstein, I., 1997. Geometric separation of

partially overlapping nonrigid objects applied to

automatic chromosome classification IEEE Trans.

Pattern Anal. Mach. Intell. 19 1212–22.

Lerner, B., Guterman, H., Dinstein, I., 1998. A

classification-driven partially occluded object

segmentation (CPOOS) method with application to

chromosome analysis. IEEE Transactions on Signal

AUTOMATIC DETECTION OF PALE PATH AND OVERLAPS IN CHROMOSOME IMAGES USING ADAPTIVE

SEARCH TECHNIQUE AND RE-THRESHOLDING

465

Processing, 46:2841-2847.

Schwartzkopf, W., Evans, B. L., Bovik, A. C., 2001

Minimum entropy segmentation applied to multi-

spectral chromosome images, International

Conference on Image Processing vol 2, pp 865–8.

Ritter, G., Gao, L., 2008. Automatic segmentation of

metaphase cells based on global context and variant

analysis. Pattern Recognition, 41(1):38-55.

Ming, D., Tian, J., 2010. Automatic Pattern Extraction and

Classification for Chromosome Images. Journal of

Infrared, Millimeter and Terahertz Waves, Volume 31,

866-877.

VISAPP 2012 - International Conference on Computer Vision Theory and Applications

466