IMAGE SEGMENTATION TO EVALUATE

ISLETS OF LANGHERANS

C. Grimaudo, D. Tegolo, C. Valenti

Dipartimento di Matematica e Applicazioni, Universit`a di Palermo, via Archiraﬁ 34, 90123, Italy

F. Bertuzzi

Istituto Mediterraneo per i Trapianti e Terapie ad Alta Specializzazione, via Tricomi 1, 90127, Italy

Keywords:

Islets of Langherans, implantation advisability, image segmentation.

Abstract:

This contribution deals with an unsupervised system to process digital photomicrographs in order to locate

and analyze islets of Langherans in human pancreases. The experiment has been conducted on real data and,

though we are still going to complete the evaluation of the whole method, we expect to deﬁne a set of proper

features (e.g. area, perimeter, fractal dimension, shape complexity, texture and entropy) useful for a fast and

reliable counting of healthy cells. In particular, this research aims to measure the advisability of a possible

implantation in patients affected by type 1 diabetes mellitus.

1 INTRODUCTION

This paper introduces a new system for the auto-

matic analysis of high power magniﬁcation photomi-

crographs of the human islets of Langherans. The

cells that make these clusters can be divided into a

few classes which include the α cells, that secrete

glucagon, and the β cells, responsible for the pro-

duction of insulin. This research ﬁeld is of particular

interest because of the demand to evaluate the state

of these endocrine tissue for preoperative planning in

patients that suffer from severe type 1 diabetes melli-

tus, otherwise scarcely treatable by conventional ther-

apies (Ryan et al., 2005; Shapiro et al., 2006). It has

been veriﬁed that the probability of obtaining a fa-

vorable implantation increases when a large number

of viable and puriﬁed islets is transplanted in to the

patients (Bertuzzi and Ricordi, 2007). In a multivari-

ate analysis aimed to identify some in vitro parame-

ters for islet quality or function predictive of in vivo

graft function of the same islets after their transplan-

tation in diabetic patients, islet morphology (in terms

of the maintenance of their round shape proﬁle, sim-

ilar to what they showed in the native pancreas) was

demonstrated to be correlated with 1 month recipi-

ent c-peptide production (Ricordi et al., 2001); islet

morphology therefore should be considered an indi-

rect parameter of islet viability. These results call

for the identiﬁcation of some standardized strategies

to characterize islet morphology and to quantify their

degree of maintenance of their native round morphol-

ogy (Nano et al., 2005).

At present, the analysis is also performed by im-

proving the appearance through image processing

softwares or ad hoc systems (Metamorph). A grid is

laid on the slide so to ﬁx the islets and to let easily

count their different typologies (see Figure 1). This

process is done by hand to separate those cells useful

to the implantation and obviously it is slow, subjec-

tive and liable to errors; an environment to help the

expert analyst is therefore desirable both to enhance

the quality of the digital photos and to elaborate the

images in order to locate automatically the zones of

interest.

A variety of methods is already present in liter-

ature for both supervised and unsupervised segmen-

tation of photomicrographs depicting cells (Coelho

et al., 2002; Tripodo et al., 2006; Montseny et al.,

2004; Bak et al., 2004). Usually these techniques

are taken back to the elaboration of histograms, appli-

cation of mathematical morphology, texture analysis,

Fourier and wavelet transforms to extract the shapes

of the components that have been found. Often the

images have noise due to the presence of small arti-

facts, distortions and blurring introduced by the op-

tical system, inherent inaccuracies due to the lattice

Grimaudo C., Tegolo D., Valenti C. and Bertuzzi F. (2008).

IMAGE SEGMENTATION TO EVALUATE ISLETS OF LANGHERANS.

In Proceedings of the First International Conference on Bio-inspired Systems and Signal Processing, pages 72-76

DOI: 10.5220/0001056800720076

 SciTePress

(e.g. the thickness of the sample that must be ana-

lyzed), imperfections of the coloring of the contrast

agent (e.g. due to variations of exposure time and to

the quantity of the marker itself).

Figure 1: Two slides that put in evidence the presence of a

grid to easily count the number of the islets.

We have realized a system that acts through a li-

brary of tools to preprocess the data; the segmenta-

tion of the various components in the images often re-

quires the intervention of an expert user who locates

the promising clusters of cells. This approach can be

applied not only to the islets of Langherans, but also

to analyze other vital cells (e.g. hepatocytes, bone

marrow cells). Finally it should be veriﬁed the pos-

sibility to apply this strategy also in ﬁxed tissue after

different immunohystochemistry staining.

The following Section 2 describes the new en-

vironment to elaborate and classify the islets of

Langherans. Experimental results are presented in

Section 3, while remarks and possible future works

are introduced in Section 4.

2 SEGMENTATION OF THE

ISLETS

In this paper, we aimed to describe the system which

has been developed to provide an unsupervised anal-

ysis of the human islets of Langherans (see Figure 2).

Different techniques have been implemented to en-

hance the quality of the images, to segment all com-

ponents, to distinguish among the cells and to evalu-

ate their conditions in order to quantify the advisabil-

ity of the implantation.

The photos in our database have been acquired

through a digital tool; they suffer from artifacts due

to the equipment (e.g. only the center of the image is

correctly in focus and a few impurities can be present

on the lenses). Predetermined threshold values result

in a poor separation between the components of the

images, but we have experimentally veriﬁed that the

Otsu method (Otsu, 1979) is able to compute these

optimal values in order to locate the imperfections on

the red and green channels of the RGB color space.

We have carried out a statistical examination on both

the background and foreground to determine their

starting threshold values; should the input image be

very different from the database we have considered,

then, to better calibrate the values, the user can select

some regions of interest, representative of the differ-

ent parts of the islets. Figure 3 shows the previous

input image soon after the preprocessing step.

Figure 2: A sample photomicrograph of a cluster of the hu-

man islets of Langherans.

We have successfully applied the same adap-

tive self-tuning technique that has been introduced

in (Tripodo et al., 2006) to discriminate between the

pure β cells, or the mixed β and exocrine/ductal cells

that are highlighted by the marker as red and orange

zones respectively, while the yellow parts correspond

to dead cells or impurities or simply exocrine/ductal

cells. This usually leads to a rough representation of

the cells, but a simple median ﬁlter is sufﬁcient to

remove all small objects (5×5 kernel) and pointlike

noise (3×3 kernel). The shape of the cells so far ob-

tained can be further enhanced by the use of a math-

ematical morphology opening with a structuring ele-

ment represented by discrete disk of radius 2 (Soille,

2003). In such a way the cells of the islets are bet-

ter separated and, moreover, we can safely delete all

components that are too small (the allowed number of

pixels has been pre-deﬁned according to the present

magniﬁcation power of the microscope). Figure 4

shows the ﬁnal result obtained on the reference im-

age; another example is reported in Figure 5. We have

highlighted the ﬁnal contour just to easily check the

segmentation of the relevant islets.

IMAGE SEGMENTATION TO EVALUATE ISLETS OF LANGHERANS

Figure 3: Some artifacts present in Figure 2 have been re-

moved. Due to the huge ﬁeld of view, in the following we

will propose the results relative to the superimposed box.

Figure 4: Left: the remaining artifacts and dead cells have

been automatically removed from Figure 3. Right: the ﬁnal

contour has been plotted on the input image of Figure 2.

3 EXPERIMENTAL RESULTS

Images have a size of 2088×1550 pixels and were ac-

quired at a sample dilution equal to 2500×, by a stere-

omicroscope Leica MZ12-5 with a 2× zoom magni-

ﬁcation and equipped with a digital camera, able of

a 4.34765µm/pixel picture calibration. The set of im-

ages we have studied has been obtained by isolating

the islets through the automated method from multi-

organ donors (Ricordi et al., 1989). After pancreas

digestion the islets from 3 preparations have been pu-

riﬁed by COBE processor (Vargas et al., 1996) and

placed in a culture media for additional 48 hours at

◦

C. The islets have been ﬁnally stained with dithi-

zone (a vital stain that cross-reacts with zinc) and

therefore it has been used to recognize the α, exocrine

and ductal cells (in which zinc is absent) from the β

cells (rich in zinc).

A set of parameters that describe each kind of

cluster of cells has been extracted from the segmented

images. The area, the perimeter, the compactness (i.e.

the normalized ratio between the area and the squared

perimeter) and the eccentricity of the ellipse which

approximates the shape of the islet and the measures

of convexity/concavity of its edges return a quanti-

tative esteem of its aspect. In particular, compactness

and eccentricity measure the roundness: healthy islets

should not have protrusions.

Figure 5: The edges of the islets within the box have been

marked in blue.

The amount of information directly deducible

from the luminosity of the pixels is another useful

characteristic: the more homogenous an islet is, the

smaller its local entropy is. We are still investigat-

ing on the ability of operators that return marks about

the value of local sharpness and textures (which are

closely connected to the presence of luminosity gra-

dients).

For each islet I

we compute the product g

be-

tween its average luminosity ℓ

and its entropy e

If we indicate with µ

and σ

respectively the mean

and the standard deviation of all g = ℓ×e, then the

islets with a score |g

−µ

| < 2σ

can be considered

as promising candidate. A further important charac-

teristic is given by the compactness κ (Rangayyan,

2005): with an analogous approach, the islets till now

accepted with a compactness value κ

< µ

+σ

are

deﬁnitely classiﬁed as reasonably good. For the sake

of clarity, an islet is classiﬁed as good if it passes the

test on g and then on κ. The ﬁnal evaluation of the

whole input photo of Figure 5 is summarized in Ta-

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

Table 1: The features of each islet (33 in this example) have been represented by two columns (top: g and bottom: κ). The

threshold values are represented by dashed lines and both tests have to be passed: κ reduces the number of candidates already

obtained by g (good islets have been marked with a H).

1 2

5 6 7 8 9 10

11 12

15 16

18 19 20

21 22

25 26

28 29 30 31 32 33

+2σ

−2σ

27.2

32.5

33.1

49.2

64.5

67.1

70.4

70.5

70.8

72.0

73.7

74.6

74.7

76.9

79.6

80.6

81.5

81.8

82.7

82.8

84.0

85.3

86.9

87.0

87.5

89.8

90.9

91.6

93.0

97.6

99.2

120.8

H H H H H H H H H H H H H H H H H H H H H H H H H H

+σ

0.86

0.74

0.58

0.37

0.20

0.09

0.55

0.25

0.05

0.52

0.05

0.36

0.25

0.28

0.35

0.71

0.23

0.19

0.12

0.39

0.44

0.18

0.07

0.13

0.21

0.13

0.15

0.19

0.20

0.48

ble 1. Figure 6 shows how the system highlights a

single islet and proposes its features.

The percentage of the area of the yellow zones

(more precisely, the ratio between red and yellow)

indicates the puriﬁcation of islet preparation and the

eventual presence of embedded islets, that means

islets surrounded by exocrine tissue (Ricordi et al.,

1995). The ﬁnal ratio between the area of good islets

and the area of all islets summarizes the goodness of

the inspected photomicrograph. Several parameters

have been therefore available now by an automated

method of analysis for the characterization of an islet

preparation in terms of:

• islet number (the number of red clusters);

• islet dimension (the red area);

• islet puriﬁcation (the ratio between yellow and red

areas in the whole preparation);

• percentage of embedded islets (the ratio between

red and yellow areas within an islet);

• islet morphology.

4 REMARKS AND FURTHER

WORKS

We have introduced an unsupervised system to locate

the human islets of Langherans in photomicrographs.

These clusters of cells have been characterized in or-

der to deﬁne some parameters representative of their

number and morphology. The predictive role of these

features towards their in vivo graft function should be

matter of further studies.

Figure 6: A screenshot of the graphical interface of the

system. Small green boxes automatically delimit bubbles

(present as artifacts in the photo). A selected islet is pointed

out by a white arrow and the values of the relevant features

are presented to the user.

From a computer science point of view, the efﬁ-

ciency of the proposed method is still at the testing

stage (Altman, 1999) and our system should be con-

sidered as a tool to help the experts in obtaining a

quantitative esteem of the reliability of the islets in fa-

vorable implantation. The ﬁnal results have been val-

idated by biologists involved in implantations to treat

patients affected by severe forms of type 1 diabetes

mellitus. It is interesting to note that the methodolo-

gies we have applied to segment the components of

the photos are quite standard and general enough and

that the extracted features can be extended to differ-

entiate between the α and β cells which compose the

islets; this is to correlate their peculiarities with infor-

mation of the state of the patients. Moreover, though

preliminary results are encouraging, we are improv-

ing the segmentation procedure by including further

algorithms based on mathematical morphology and

IMAGE SEGMENTATION TO EVALUATE ISLETS OF LANGHERANS

watershed/level sets.

To the best of our knowledge, our environment

is the ﬁrst attempt to automatically analyze islets of

Langherans for implantations. Previous works rely

on manual segmentation of their photomicrographsor

are too general, thus to require to be adapted in or-

der to process images containing these kind of cells.

Therefore, a comparison of the results obtained by our

system is still desirable.

Additional projects should be the in vitro char-

acterization of the human islet preparations after the

staining with vital probes (i.e. propidium iodide, ﬂu-

orescein diacetate (Barnett et al., 2004; Miyamoto

et al., 2000) and probes for apoptosis (Ichii et al.,

2005). This should allow the direct quantiﬁcation of

vital, apoptotic and necrotic islets. Finally the auto-

mated system for imaging analysis should be applied

in ﬁxed tissues after immunostaining for insulin and

glucagon thus allowing a complete characterization of

islet cell composition (Ichii et al., 2005; Street et al.,

2004).

ACKNOWLEDGEMENTS

The authors wish to thank Doctor Domenico Bosco

of the Hˆopitaux Universitaires de Gen`eve for useful

discussions and his kind contribution in providing the

input images.

REFERENCES

D.Altman (1999). Practical statistics for medical research.

Chapman & Hall/CRC.

E.Bak, K.Najarian, J.P.Brockway (2004). Efﬁcient segmen-

tation framework of cell images in noise environ-

ments. Proceedings of the 26th Annual International

Conference of the Engineering in Medicine and Biol-

ogy Society, 3:1802–1805.

M.J.Barnett, D.McGhee-Wilson, A.M.Shapiro, J.R.Lakey

(2004). Variation in human islet viability based on

different membrane integrity stains. Cell Transplant,

13:481–488.

F.Bertuzzi, C.Ricordi(2007). Prediction of clinical outcome

in islet allotransplantation. Diabetes Care, 30:410–

417.

R.C.Coelho, V.Ges`u, G.Bosco, J.S.Tanaka, C.Valenti

(2002). Shape-based features for cat ganglion retinal

cells classiﬁcation. Real-Time Imaging, Special Issue

on Imaging in Bioinformatics, 8:213–226.

H.Ichii, L.Inverardi, A.Pileggi, R.D.Molano, O.Cabrera,

A.Caicedo, S.Messinger, Y.Kuroda, P.O.Berggren,

C.Ricordi (2005). A novel method for the assess-

ment of cellular composition and beta-cell viability in

human islet preparations. Am J Transplant, 5:1635–

1645.

M.Miyamoto, Y.Morimoto, Y.Nozawa, A.N.Balamurugan,

B.Xu, (2000). Establishment of ﬂuorescein diacetate

and ethidium bromide (fdaeb) assay for quality assess-

ment of isolated islets. Cell Transplant, 9:681–686.

E.Montseny, P.Sobrevilla, S.Romani (2004). A fuzzy ap-

proach to white blood cells segmentation in color bone

marrow images. Proceedings of the IEEE Interna-

tional Conference on Fuzzy Systems, 1:173–178.

R.Nano, B.Clissi, R.Melzi, G.Calori, P.Mafﬁ,

B.Antonioli, S.Marzorati, L.Aldrighetti, M.Freschi,

T.Grochowiecki, C.Socci, A.Secchi, V.Carlo,

E.Bonifacio, F.Bertuzzi (2005). Islet isolation for

allotransplantation: variables associated with suc-

cessful islet yield and graft function. Diabetologia,

48:906–912.

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

scale histogram. IEEE Trans. on System, Man, and

Cybernetics, 9:62–66.

R.M.Rangayyan (2005). Biomedical Image Analysis. CRC

Press.

C.Ricordi, P.E.Lacy, D.W.Scharp (1989). Automated islet

isolation from human pancreas diabetes. Diabetes,

38(1):140–142.

C.Ricordi, R.Alejandro, H.H.Rilo, P.B.Carroll, A.G.Tzakis,

T.E.Starzl, D.H.Mintz (1995). Long-term in vivo

function of human mantled islets obtained by incom-

plete pancreatic dissociation and puriﬁcation. Trans-

plant Proc., 27(6):3382.

C.Ricordi, J.R.Lakey, B.J.Hering (2001). Challenges to-

ward standardization of islet isolation technology.

Transplant Proc., 33(1–2):1709.

E.A.Ryan, B.W.Paty, P.A.Senior, D.Bigam, E.Alfadhli,

N.M.Kneteman, J.R.Lakey, A.M.Shapiro (2005).

Five-year follow-up after clinical islet transplantation.

Diabetes, 54(7):2060–2069.

A.M.Shapiro, C.Ricordi, B.J.Hering (2006). International

trial of the edmonton protocol for islet transplan-

tation. The New England Journal of Medicine,

355(13):1318–1330.

P.Soille (2003). Morphological Image Analysis. Springer-

Verlag, New York, 2nd edition.

C.N.Street, J.R.Lakey, A.M.Shapiro, S.Imes, R.V.Rajotte,

E.A.Ryan, J.G.Lyon, T.Kin, J.Avila, T.Tsujimura,

G.S.Korbutt (2004). Islet graft assessment in the ed-

monton protocol: implications for predicting long-

term clinical outcome. Diabetes, 53:3107–3114.

The metamorph system,

 molecular devices and

universal imaging corporation, WWW.UNIVERSAL-

IMAGING.COM.

C.Tripodo, C.Valenti, B.Ballar`o, Z.Rudzki, D.Tegolo,

V.Ges`u, A.M.Florena, V.Franco (2006). Megakary-

ocytic features useful for the diagnosis of myeloprolif-

erative disorders can be obtained by a novel unsuper-

vised software analysis. Histology and Histopathol-

ogy - Cellular and Molecular Biology, 21:813–821.

F.Vargas, M.Vives-Pi, N.Somoza (1996). Advantages of us-

ing a cell separator and metrizamide gradients for

human islet puriﬁcation. Transplantation, 61:1562–

1566.

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing