THEORETICAL

B I O LO G Y
F O RU M

107 · 1-2/2014

PISA · ROMA
FABRIZIO SERRA EDITORE
MMXIV
 Autorizzazione del Tribunale di Pisa n. 13 del 14 maggio 2012.
Già registrata presso il Tribunale di Genova: registrazione n. 22/96 del 2 maggio 1996.
Direttore responsabile: Fabrizio Serra
*
Amministrazione e abbonamenti
Fabrizio Serra editore ®
Casella postale n. 1, succursale n. 8, I 56123 Pisa
Uffici di Pisa: Via Santa Bibbiana 28, I 56127 Pisa,
tel. +39 050542332, fax +39 050574888, fse@libraweb.net
Uffici di Roma: Via Carlo Emanuele I, I 00185 Roma,
tel. +39 0670493456, fax +39 0670476605, fse.roma@libraweb.net
*
I prezzi ufficiali di abbonamento cartaceo e/o Online sono consultabili
presso il sito Internet della casa editrice www.libraweb.net.
Print and/or Online official subscription rates are available at Publisher’s website www.libraweb.net.
I pagamenti possono essere effettuati tramite versamento su c.c.p. n. 17154550 indirizzato a:
Fabrizio Serra editore ® o tramite carta di credito (American Express, Eurocard, Mastercard, Visa).
Suscriptions should be paid by cheque/international money order to
Fabrizio Serra editore, casella postale 1, succ. 8, I, 56123 Pisa, or by Bank transfer
(Monte dei Paschi di Siena, iban it 90g 0103014009000000012660, swift code pascitm 1y 71),
or by credit card (American Express, Eurocard, Mastercard, Visa).
*
Proprietà riservata · All rights reserved
© Copyright 2014 by Fabrizio Serra editore, Pisa · Roma.
Fabrizio Serra editore incorporates the Imprints Accademia editoriale,
Edizioni dell’Ateneo, Fabrizio Serra editore, Giardini editori e stampatori in Pisa,
Gruppo editoriale internazionale and Istituti editoriali e poligrafici internazionali.
www.libraweb.net
A norma del codice civile italiano, è vietata la riproduzione, totale o parziale
(compresi estratti, ecc.), di questa pubblicazione in qualsiasi forma e versione
(comprese bozze, ecc.), originale o derivata, e con qualsiasi mezzo a stampa o internet
(compresi siti web personali e istituzionali, academia.edu, ecc.), elettronico, digitale,
meccanico, per mezzo di fotocopie, pdf, microfilm, film, scanner o altro,
senza il permesso scritto della casa editrice.
Under Italian civil law this publication cannot be reproduced, wholly or in part (included offprints, etc.),
in any form (included proofs, etc.), original or derived, or by any means: print, internet
(included personal and institutional web sites, academia.edu, etc.), electronic, digital, mechanical,
including photocopy, pdf, microfilm, film, scanner or any other medium,
without permission in writing from the publisher.
ISSN 2282-2593
ISSN ELETTRONICO 1825-6538
 CONTENTS

Editorial 9

articles
Pieranna Fietta, Elvira Costa, Giovanni Delsante, Interleukins (ILS),
a Fascinating Family of Cytokines. Part I: ILS from IL-1 to IL-19 13
Katell Guizien, Lorenzo Bramanti, Modelling Ecological Complexity for
Marine Species Conservation: the Effect of Variable Connectivity on Species
Spatial Distribution and Age-structure 47
Gianluca Martelloni, Alisa Santarlasci, Franco Bagnoli, Giacomo
Santini, Modeling Ant Battles by Means of a Diffusion-Limited Gillespie Algo-
rithm 57
Giulia Menichetti, Daniel Remondini, Entropy of a Network Ensemble:
Definitions and Applications to Genomic Data 77
Marco Pellegrini, Giuseppe Pirillo, On the Dinucleotide Circular Codes of
Maximum Cardinality 89
Sergio Pennazio, Elements of Plant Physiology in Theophrastus’ Botany 97
Dušan Ristanović, Bratislav D. Stefanović, Nela Puškaš, Fractal
Analysis of Dendrite Morphology of Rotated Neuronal Pictures: The Modified
Box Counting Method 109
Alessia Rossi, Simone Panigada, Massimo Arrigoni, Margherita Za-
nardelli, Cristina Cimmino, Luigi Marangi, Piero Manfredi, Gio-
vanni Santangelo, Demography and Conservation of the Mediterranean Fin
Whale (Balaenoptera physalus): What Clues can be obtained from Photo-Identifi-
cation Data 123
Sureyya Mert Selimoglu, Conscience Dilemma: to become a Bioengineer or to
survive as a Biologist 143
Masomeh Taherian, Megerdich Toomanian, Mohammadreza Molaei,
A Dynamical Model for Influenza under Seasonal Variables 151
Armando Bazzani, Paolo Freguglia, Dynamics on Genes Network Struc-
tures. An Ago-Antagonist Approach 163

Instructions for Authors 171

FRACTAL ANALYSIS OF DENDRITE MORPHOLOGY
OF ROTATED NEURONAL PICTURES:
THE MODIFIED BOX COUNTING METHOD
Dušan Ristanović1 · Br atislav D. Stefanović2
Nela Pušk aš2
1. Department of Biophysics, Medical Faculty, University of Belgrade, Serbia
2. Department of Histology. Medical Faculty, University of Belgrade, Serbia
Corresponding author: Dušan Ristanović
E-mail: dusan@ristanovic.com

Contents: 1. Introduction. 2. Material and Methods. 2. 1. Experimental. 2. 2. Statistics.

2. 2. 1. Testing small samples for normality. 2. 2. 2. The paired experiment 2. 2. 3. Fisher’s
F test. 3. Results. 3. 1. Elementary geometry of screen’s pixels. 3. 2. Fractal dimensions and
space-filling capacities. 3. 3. Testing the samples for normality. 3. 4. Vertical versus oblique neu-
ronal pictures. 3. 5. The modified box-counting method. 4. Discussion. 5. Conclusion.
Keywords: Box dimension, Fisher’s test, Non-stellate neurons, Space-filling capacity,
Statistical analysis.

Abstract: The fractal dimension of a non-stel-
late neuron changes continuously with rotation
of the neuronal picture. For a stellate neuron
such changes cannot be noticed. During prepro-
cessing for the box counting, non-stellate neurons
should be arranged so that the major diameters
of their dendrite fields are parallel. It was shown
that a non-stellate neuronal picture had the
smallest box dimension when the angle between
the horizontal or vertical axis and its major di-
ameter was about 45°.The box counting method
which uses ImageJ software does not consider the
position of a picture on the computer’s screen. A
dispersion of the box dimension values of a sam-
ple is generally rather large so that their mean
value is with larger standard deviation. Modi-
fied box counting method partly diminishes these
findings. To improve a dependence on neuronal
rotation for the box counting dimension of non-
stellate neurons, prior to applying the box count-
ing, the non-stellate neurons should be arranged
so that the major diameters of their dendrite
fields are parallel.

1. Introduction

F ractal analysis is a contemporary nontraditional mathematical method de-

rived from Mandelbrot’s Fractal Geometry of Nature [1], Euclidean geometry and
calculus. This can be used to determine the fractal dimension of a geometrical or
natural object [2, 3]. It is also an ideal method for quantification of branching pat-
terns of dendrite trees [2, 4]. The fractal dimension is considered to be a morpho-
metric quantitative parameter characterizing the complexity of an object [5-7]. It
can be geometrical and computer-based. A computer-based fractal dimension is
higher if the object is more complex, meaning that its border contains more inflec-
110 dušan ristanović · bratislav d. stefanović · nela puškaš
tive departures from smoothness via scaled reiterations of discontinuities [3, 7]. For
dendritic morphology, such fractal dimensionality may represent a strategy for
more thoroughly filling interstitial spaces in which the object is embedded. Many
computer-based methods for measuring the fractal dimension exist [2, 5, 8]. One of
the most popular methods is the box counting method [3, 6, 7, 9-14]. Most scientists
use the box counting methods since there is many simple and available computer-
based programs (such as ImageJ) to measure the fractal (box) dimension.
Scientists use the box counting method to exploit its efficiency, suitability for
large data sets, and ready computability (e.g., program ImageJ [15]). To do the box
counting, the object of analysis is overlaid with a rectangular coordinate grid [8] and
the squares (‘boxes’) intersected by the pattern are counted (N). Each set of boxes
is characterized by the square side r. This procedure is repeated with grids of dif-
ferent box sizes. The fractal (box) dimension D of an image is determined as the ab-
solute value of the slope of log-log relation between N and r. We have reported that
the box dimension measured using ImageJ software depends on the rotation posi-
tion of the image indicating that the use of the box counting method requires care-
ful usage [14].
Three main aims of the present study are as follow:
(1) It is well-known that a picture exhibits rotational symmetry if rotation by a spe-
cific angle about some central axis point can return the picture to its original con-
figuration. Squares and hexagonal snowflakes are common examples: rotation by
90° of a square or by 60° of a hexagonal snowflake about their centers returns the
shapes to their original configurations. It is reasonable to presume that the box di-
mensions of such picture in these positions have the same value. On the other hand,
neurons with dendrites homogeneously filling a spherical volume are termed stel-
late [16]. Dendrites of such neurons radiate in all directions from the cell body. A
projection of a stellate neuron on a plane gives rise to a 2D picture in which pro-
jections of dendrites also radiate in all directions from the cell body projection with-
in a circle. It is clear that the box dimension remains the same at any rotational po-
sition of such a neuron. In contrast, neurons lacking strong radial symmetry can be
termed non-stellate (such as pyramidal, Purkinje, those with fusiform bodies, etc.).
One of the aims of the present paper is to explore how the fractal dimension
changes in the course of continuous rotation of non-stellate neurons between such
extreme positions and offer a possible explanation for these findings.
(2) The fractal dimension has also been used to quantify the extent to which an
arborization fills its spatial domain [2, 5, 9, 16]. However, in this assertion there is a
non-resolved issue that needs to be addressed. In mathematical analysis, a space-fill-
ing (Peano’s) curve is a close packing curve whose definition range contains the en-
tire two-dimensional unit square. It is a self-intersecting curve that passes through
every point of the unit square. By analogy, most of the neuroscientists use the com-
puter-based fractal dimension to measure the space-filling capacity of a pattern [9]
expressing how densely the pattern occupies the portion of the metric space in
which it is embedded [5]. In the computational analysis the space-filling capacity of
a binary image can be thought of as the number of its black pixels. We have shown
that the box dimension of a given object increased with its space-filling capacity and
 fractal analysis of dendride morphology 111

Fig. 1. Sixteen representative examples of pyramidal neurons. Neurons are selected within fron-
toparietal cortex being dorsally and dorso-laterally positioned in the coronal sections of the rat
forebrain and of cerebral cortex hemispheres. All apical primary dendrites are oriented vertically.
The cortical layers are denoted by Roman numerals. Scale bar: 50 Ìm.

that these two parameters of an apical dendrite of a pyramidal neuron decreased if

this dendrite from its vertical position rotated for 45° [14]. It is one of our purpos-
es to prove out and statistically analyzed this finding for neuronal samples.
(3) Every engineer, physicist or scientist who investigates using exact measure-
ment knows that the result of a measurement does not worth much without the cor-
responding experimental error. For that purpose such a scientist chooses a method
of measurement which enables the smallest experimental error. For example, it
seems to be easy to average the fractal dimensions of disorganized pictures using
box counting method, as did Panico and Sterling [9], but we showed that our modi-
fied box counting method offers the way how to obtain more exact mean fractal di-
mension with smaller standard deviation. One of the aims of the present study is to
present, using Fisher’s F test, a modification to box counting methods to address ro-
tational variation in non-stellate neurons. This method diminishes possible errors
related to rotation and spurious effects on the box counting dimension that are at-
tributable to image orientation and can obscure measures of a pattern’s complexity.

2. Material and Methods

2. 1. Experimental
Male Wistar rats were used in this study. A detailed description of the histological
procedure used and data processing can be found in Ristanović et al. [14]. We con-
sidered cortical pyramidal neurons and focused our attention on a sample of 16 soli-
tary pyramidal cells with completely impregnated apical dendrite arborization (Fig.
1). It should be noticed that an apical dendrite is a dendrite that emerges from the
apex of a pyramidal cell body. It is a single, long, thick and rather straight dendrite
that usually branches obliquely several times. Drawings of the apical dendrite ar-
borization were further analyzed after removing all other parts of these pictures
(cell bodies, spines and basal dendrites). We used tools available in the public
domain software ImageJ (http://rsbweb.nih.gov/ij) for image reconstruction and
determining the box dimensions.
112 dušan ristanović · bratislav d. stefanović · nela puškaš

2. 2. Statistics
2. 2. 1. Testing small samples for normality
In order to use a Chi-square distribution to test a population for normality, a fairy
large sample from the population is necessary. In the case of small samples, where
the number of variates is 30 or less [19], we have adopted the criteria for normality
of data distribution offered by Smirnov and Barkovskiy [20]. The method is related
to estimating the third (m3) and fourth (m4) empirical moment about the mean. The
moment of the order k is
n

∑ (X − X
i =1
i av )k
(1)
mk =
n
where Xi is the variate i of the variable X, Xav is the sample mean and n is the num-
ber of variates.
One of the measures describing the asymmetry of a distribution about its maxi-
mum is called the skewness (A), and a measure of the degree of the kurtosis (m4) of
the distribution is called the excess (E). They are given, respectively, as
m3 m
A= 3
E = 44 − 3 (2)
s s
where s is the empirical standard deviation (known as the best estimate of the stan-
dard deviation of the population) which is given by
n

∑ (X − X
i =1
i av )2
(3)
s=
n −1
The value of the skewness for a perfectly symmetrical distribution of data is zero
and the excess for the normal curve has also the value equals zero. In the case of an
experimental normal distribution these parameters should be very small (close to
zero). The degree of how much they should be small to enable the distribution nor-
mality can be estimated by dividing the corresponding mean standard deviations of
asymmetry and excess
6( n − 1) 24n( n − 2)( n − 3)
σ3 = , σ4 = (4)
( n + 1)( n + 3) ( n − 1) 2 ( n + 3)( n + 5)
by the values given by Eq. (2), where n is the number of variates in the sample. If
σ3 σ4
≤ 2.0 ≤ 2.0 (5)
A E
there are grounds for strong believing that the data are normally distributed [20].
 fractal analysis of dendride morphology 113

2. 2. 2. The paired experiment

The paired Student’s T test is a hypothesis test that is used to compare the means
of two populations when each variate of one population is related to a correspon-
ding variate from the other population [21]. To analyze such a paired experiment,
we formed for each pair the difference between the corresponding variates Xi and
Yi and denote the difference by di, i.e., di = Xi – Yi. We look for the estimated stan-
dard error SE of the mean of d, which is denoted by dav
n

∑ (d − d )
2
i av
i =1 (6)
SE =
n( n − 1)
where n is the number of pairs.
The resulting t statistic is determined by
dav
. t= (7)
SE
If ‰ is the difference between the means of the two populations, the hypotheses are:
H0: ‰ = 0 (there is no difference between the populations); H1: ‰ ≠ 0 (there is a dif-
ference between the populations - treatments). We reject the null hypothesis at a sig-
nificance level p if t > ttab, where ttab is the critical value from the Student’s table,
with n – 1 degrees of freedom, where n is the number of pairs. If t < ttab the hy-
pothesis is accepted.

2. 2. 3. Fisher’s F test
For a sample of n variates, X1, X2, X3, …, Xn, whose mean is Xav, the best estimate of
the standard deviation of the population, denoted by s, is given by Eq. (3). The
square of the standard deviation [Eq. (3)] is called the variance. The variance is wide-
ly used as a measure for dispersion of a set of variates about their mean. The vari-
ances of two samples are said to be homogeneous if the populations from which the
two samples are selected, have identical variances (or identical standard deviations).
If X and Y are normally distributed populations with identical standard deviations,
than the ratio of the squares of the best estimates of the standard deviations (the
ratio of the variances), that is,
s x2
F= (8)
s y2
Where
nx ny

∑ (X − X
i =1
i av ) 2
∑ (Y − Y
i =1
i av )2
(9)
s x2 = s y2 =
nx − 1 ny − 1
114 dušan ristanović · bratislav d. stefanović · nela puškaš

satisfy an F-distribution with nx – 1 and

ny – 1 degrees of freedom. To test two
samples for homogeneity of variances,
we set up the hypothesis that the vari-
ances of the two populations are identi-
cal, that is, the value F is expected to be
near 1. We reject the hypothesis if the
value of F is larger than its critical value
(from Table ‘F distribution’), that is,
whether sx and sy differ significantly; oth-
erwise we accept it. This makes sense if
the larger standard deviation is placed in
the numerator of Eq. (8) [19].
Fig. 2. Geometry of screen’s pixels. Number of
pixels does not determine the geometric length 3. Results
on a computer screen. The length r of the series
of horizontally and vertically arranged pixels (10 3. 1. Elementary geometry
squares) is approximately equal to the length r of screen’s pixels
of oblique pixels (7 squares). O is the center of a
circle (shown by broken line) of the radius r.
In digital imaging a pixel is a physical
point in a raster image, being the small-
est controllable element of a picture on
the screen. In computational methods, an image composed of many pixels is known
as a bitmapped image. For convenience, pixels are normally arranged on a com-
puter’s screen in a regular two-dimensional grid. A binary (black-on-white) image
is a digital image that has only two possible values for each pixel. Typically the two
colors used for a binary image are black and white. The color used for the object in
the image is the foreground color (usually black) while the rest of the images is the
background color (usually white). A special kind of binary pictures is a skeletonized
picture. This picture consists of one-pixel-wide lines. It is a custom to call binary
skeletonized picture as skeletonized, and binary non-skeletonized pictures as binary.
In a skeletonized line on a screen two adjacent pixels can be in touch either by
their sides or by their vertices [14]. In the first case the pixels can form horizontal or
vertical lines, while in the second case they form an oblique straight line the angle
from the horizontal and oblique line being 45° (Fig. 2). The example in Fig. 2 shows
that the same geometric length (r) can be achieved either with 10 horizontally or
vertically arranged squares (pixels) or with approximately 7 squares placed as an
oblique straight line. Since it was shown that the fractal dimension measures the
space filling [9], it is acceptable that the fractal dimension of an asymmetrical
oblique object (7 squares, Fig. 2) is smaller than that of a horizontal or vertical
object (10 squares). To illustrate this finding we present three simple examples:
(1) Consider a skeletonized square (side length 60 mm) drawn on the computer
screen at a resolution of 300 ppi. From its starting position (two sides of the square
are horizontal and other two are vertical) denoted by 0, the square rotates for a
series of angles 7.5°, 15°, 22.5°, 30° etc. to 90°. From Fig. 3A it is seen that the box
 fractal analysis of dendride morphology 115

Fig. 3. Relations between the fractal (box) dimensions and the rotational angle (A) for the
skeletonized square (the data of each branch are fitted by a general second-degree function, the co-
efficient of correlation being R1 = 0.988), (B) and for the binary square (the data of each branches
are fitted by a general third-degree polynomial, the coefficient of correlation being R2 = 0.959).

Fig. 4. Relations between the fractal (box) dimensions and rotational angle (A) of a skeletonized
neuron 5 selected at random from Fig. 1, (B) and of a binary picture of the same neuron. In both
cases the minima are about 40°. Data points are connected using the spline-B function.

dimension of the square decreases when the angle of the square changes from 0°
to 45°. After that the box dimension increases. Each branch of these data is fitted
using the second-degree polynomial. In Fig. 3B where the square is shown as a bi-
nary image (the width of the lines is 1 pt) a similar result is obtained. The branches
are fitted using the third-degree polynomial.
(2) From Fig. 1 we selected, at random, a neuron (say, neuron 5), chose its api-
cal dendrite and performed with it the same procedure as with the square in the
preceding example. In Fig. 4A it is shown that the fractal dimension of the skele-
tonized neuron also decreases to the value around 40° and that of the binary pic-
ture of the neuron decreases to the value around 45° (Fig. 4B). Small deviations
from 45° are the consequence of some branches of the primary dendrite which
emanate from the apical stem in diverse directions.
116 dušan ristanović · bratislav d. stefanović · nela puškaš
Fig. 5. Relation between the fractal (box)
dimension and angle of rotation of the binary
neuron shown in Fig. 2. Data points are
connected using the spline-B function.
(3) We applied the ImageJ on a non-
stellate binary picture in Fig. 10 [3],
which represents the arbor of a non-
stellate neuron used from a 34-week-old
human fetus (adopted from Schoenen
[17] - with permission). The picture is
continuously rotated from 0° to 90° and
the relation between the fractal dimen-
sion and angle of rotation is presented
in  Fig. 5. The graph is similar to that
shown in Fig. 4. From Fig. 5 it is seen
that the fractal dimension of the picture
whose major diameter (see later) is un-
der the angle of about 45° is again the
smallest one.

3. 2. Fractal dimensions and space-filling capacities

In Fig. 6A it is shown the graph of the relation between the fractal dimension and
space-filling capacity for the skeletonized square (shown in Subsection 3.1), and in
Fig. 6B the graph for the binary square. In both cases the fractal dimension in-
creases linearly with the space-filling capacity. This is in accordance with the find-
ing in Fig. 4 that a figure rotated for 45° needed less numbers of pixels than the
same pictures in vertical position.

3. 3. testing the samples for normality

To apply parametric statistics to experimental data, it is necessary to show that pop-
ulations involved are normally distributed [19]. For that purpose, we selected two
samples from these populations: one sample was a group of 16 neurons which are
vertically oriented (Fig. 1), and the other sample was the same group of neurons
but rotated for 45°. Each sample was also considered to be binary and skeletonized
pictures. Using the method shown in Subsection 2.2.1 we show that all four samples
are normally distributed since the absolute values of Û3/A and Û4/E (Eq. 5) are not
larger than 2.0 (Tab. 1).

Picture Û3 Û4 Û3/A Û4/E

0° 45° 20°-320° 0° 45° 20°-320°
Binary 0.5279 0.8823 0.6 0.8 0.7 0.8 0.4 0.4
Skeletonized 0.5 0.4 2.0 0.8 1.1 1.1

Tab. 1. Results of testing for normality of the box dimensions of binary

and skeletonized pyramidal cells. The neurons were in vertical position 0°,
rotated for 45° and rotated between 20° and 320° (see text). The parameters Û3 and Û4
are the mean standard deviations of asymmetry (A) and excess (E).
 fractal analysis of dendride morphology 117

Fig. 6. Relations between the fractal (box) dimensions and the space-filling capacity
(A) of the skeletonized square (the data are fitted by a general first-degree function,
the coefficient of correlation R1 = 0.988), (B) and of the binary square
(the data are fitted by a general first-degree polynomial, R2 = 0.963).

3. 4. Vertical versus oblique neuronal pictures

The paired Student’s T test is generally used to compare the means of two popula-
tions when each variate of one population is related to a corresponding variate of
the other population [21]. To apply such a paired experiment to our data, we
formed, as noticed above, four samples of these populations. These samples are pre-
sented in Fig. 7 (binary neurons) and Fig. 8 (skeletonized neurons) as bar graphs.
In both cases the box dimensions of the vertical images (black bars) are larger than
those of the rotated images (gray bars). We did not apply statistical analysis to con-
firm significant differences between the means of fractal dimensions of binary and
corresponding skeletonized as well as binary pictures since these differences are
well-known. We compared the means of the binary vertical and rotated neurons
(Fig. 7), as well as the means of skeletonized vertical and rotated neurons (Fig. 8).
It was shown that these differences are statistically significant (Tab. 2). For exam-
ple, if we choose a very high significance level p = 0.001, the value of ttab is 4.073 for
n – 1 = 15 degrees of freedom and it is seen that the mean fractal dimension of the
skeletonized vertical neurons is highly significantly larger (being 5.821) than that of
skeletonized oblique neurons.

Picture Dav ± SE t F
(0°) (45°) (20° - 320°) 0° - 45° 0° - (20° - 320°)
Binary 1.357 ± 0.005 1.325±0.006 1.338 ± 0.008 4.301** 3.684* 2.63
Skeletonized 1.139 ± 0.006 1.110 ± 0.004 1.133 ± 0.009 5.821** 3.227* 2.49
* p < 0.01. ** p < 0.001
Tab. 2. Results of statistical analysis of the distribution of box-counting fractal dimensions
for binary and skeletonized pictures of pyramidal neurons. The values of mean fractal
dimensions are given as the means (Dav) and standard errors (SE) when the neurons were
in vertical position 0° (Fig. 1), rotated for 45° and rotated between 20° and 320° (see text).
Parameter t is the Student’s statistic and F is the Fisher’s statistic.
118 dušan ristanović · bratislav d. stefanović · nela puškaš

Fig. 7. The bar graph. The graph represents the box dimensions of 16 binary neurons
in vertical position, as in Fig. 1 (the first bars, in black) and those of the same neurons
rotated for 45° (the second bars in groups of two, in gray).

3. 5. The modified box-counting method

A dendrite field is a plane enclosed by the convex polygon that is formed by
connecting the tips of the longest dendrites [7, 22, 23]. The straight-line segment
representing the largest distance between two points on the border of the dendrite
field can be thought of as the major diameter.
In order to make a disorder of the neuronal pictures in Fig. 1, we rotated each
neuronal picture so that the neuron 1 was rotated for 20° from the starting (verti-
cal) position, neuron 2 for 40°, neuron 3 for 60° etc. and neuron 16 for 320°. First
of all, we showed that the fractal dimensions of such neurons were normally
distributed (Tab. 1) and their means were significantly different from the means of
the fractal dimensions of vertical neurons (T-test, Tab. 2). We applied the F-test and
showed that the variances of disordered neurons were significantly larger than
those of the ordered (vertical) neurons in Fig. 1 (Tab. 2). Indeed, the calculated F-
values (Tab. 2) exceed the critical value F.95(15,15) = 2.40 for n -1 = 15 degrees of
fredom. Assuming 5% level of significance we conclude that the two variances are
not homogeneous. The modified box-counting method is a recommendation to
arrange pictures of neurons of a given sample with non-stellate dendrite fields so
that the major diameters of the fields are parallel (see Fig. 1), and after that
preprocessing apply the box-counting method.

4. Discussion
Panico and Sterling [9] noticed that for large boxes the box-counting method is sen-
sitive to the position of the figure upon the grid, so they averaged as many as 200
 fractal analysis of dendride morphology 119

Fig. 8. The bar graph. The graph represents the box dimensions of 16 skeletonized neurons
in vertical position, as in Fig. 1 (the first bars, in black) and those of the same neurons
rotated for 45° (the second bars, in gray).

trials for each grid size, randomizing the grid position for each trial. We have
reported that if a picture is translated into CorelDraw software and such a new
picture is read into ImageJ software, its fractal dimension remained the same [14].
In the present study we reported that the box dimension measured using ImageJ
depends on the rotating position of the picture.
Panico and Sterling [9] stated that the fractal dimension reflects an average of the
pattern’s interior (whose fractal dimension is 2) and its boundary (whose fractal di-
mension approximates 1). Although these assertions seem attractive, no explicit
proof for them has been offered yet.
Neale et al. [7] claimed that total dendrite length could be obtained by counting
the number of black pixels in the skeletonized image. This can be correct only if
the dendrites of an arbor are horizontal and/or vertical with no axial inflections
which are seldom obtained in experimental data.
Stellate neurons comprise a rather broad class of cells. Stellate cells are found
throughout the brain and spinal cord, and predominate in non-laminated subcorti-
cal nuclei such as the inferior olive, pontine nuclei, striatum, thalamus, etc. [16]. The
box-counting method can be directly applied to these neurons. For the other types
of neurons, the modified box-counting method proposed here is advisable.
The rotation of neurons has been previously used and discussed by some authors
[11, 23, 24]. For example, Montaque and Friedlander [23] were randomly varied the
position and orientation of the overlay drawing relative to the grid. At each partic-
ular box size, this procedure was done 10 times. The number of boxes intersected
120 dušan ristanović · bratislav d. stefanović · nela puškaš
was then averaged. Thus, they calculated the mean fractal dimension for a neuron,
while we calculated this mean but for a sample of neurons.

5. Conclusion
We propose a modification to box-counting methods to address rotational variation
in non-stellate neurons. We recommend arranging pictures of neurons of a given
sample with non-stellate dendritic fields so that, when possible, the major axes of
the fields are parallel (see Fig. 1), prior to applying box-counting methods. This is
likely to reduce spurious effects on the box counting dimension that are attributa-
ble to image orientation and can obscure measures of a pattern’s complexity.

References
1. Mandelbrot BB. The Fractal Geometry of nature. 20th ed. New York: NW Freeman and Co;
2004.
2. Caserta F, Eldred WD, Fernández E, Hausman RE, Stanford LR, Buldyrev SV, et alii.
Determination of fractal dimension of physiologically characterized neurons in two and three
dimensions. J Neurosci Meth. 1995; 56: 133-144.
3. Ristanović D, Milošević NT. Fractal analysis: methodologies for biomedical research. Theor.
Biol Forum. 2012; 105(2): 99-108.
4. Losa GA. The fractal geometry of life. Riv Biol. 2009; 102(1): 29-59.
5. Morigiwa K, Tauchi M, Fukuda Y. Fractal analysis of ganglion cell dendritic branching
patterns of the rat and cat retinae. Neurosci Res. 1989; 10: S131-S140.
6. Smith TG, Marks WB, Lange DG, Sheriff WH, Neale EA. A fractal analysis of cell images.
J Neurosci Meth. 1989; 27: 173-180.
7. Neale EA, Bowers LM, Smith TG. Early dendrite development in spinal cord cell cultures: a
quantitative study. J Neurosci Res. 1993; 34: 54-66.
8. Smith TG, Lange DG, Marks WB. Fractal methods and results in cellular morphology -
dimensions, lacunarity and multifractals. J Neurosci Meth. 1996; 69: 123-136.
9. Panico J, Sterling P. Retinal neurons and vessels are not fractal but space-filling. J. Comp
Neurol. 1995; 36: 479-490.
10. Voss RF. Random fractal forgeries. In: Earhsghow RA, editor. Fundamental Algorithms for
Computer Graphics. NATO ASI Berlin: Springer; Series, Vol. F17, 1985: 805-835.
11. Fernández E, Jelinek HF. Use of fractal theory in neuroscience: methods, advantages, and
potential problems. Methods. 2001; 24: 309-321.
12. Jelinek HF, Ristanović D, Milošević NT. The morphology and classification of alpha ganglion
cells in the rat retinae: a fractal analysis study. J Neurosci. Meth. 2011; 201: 281-287.
13. Ristanović D, Krstonošić B, Milošević NT, Gudović R. Mathematical modelling of
transformations of asymmetrically distributed biological data: an application to a quantitative
classification of spiny neurons of the human putamen. J Theor Biol. 2012; 302: 81-88.
14. Ristanović D, Stefanović BD, Puškaš N. Fractal analysis of dendrites morphology using
modified Richardson’s and box-counting method. Theor. Biol. Forum. 2013; 1-2: 157-168.
15. Schneider CA, Rasband WS, Eliceir KW. NIH Image to ImageJ: 25 years of image analysis,
Nature Methods. 2012; 9: 671-685.
16. Fiala JC, Harris KM. Dendrite structure. In: Stuar G, Spruston N, Häustorn M, editors.
Dendrites. Oxford: Oxford Univ Press; 2002; 1-34.
17. Schoenen J. The dendritic organization of the human spinal cord: the dorsal horn. Neuro-
science 1982; 7: 2057-2087.
 fractal analysis of dendride morphology 121
18. Milošević NT, Ristanović D, Stanković JB. Fractal analysis of the laminar organization of
spinal cord neurons. J Neurosci Meth. 2005; 146: 198-204.
19. Alder HL, Roessler EB. Introduction to Probability and Statistics. San Francisco: W.H. Free-
man and Co; 1972.
20. Kurickiy BY, Mathematical Methods in Physiology (in Russian). Leningrad: Nauka; 1969.
21. Dodge Y. The Concise Encyclopedia of Statistics. Berlin: Springer; 2008.
22. Milošević NT, Ristanović D, Jeline HF, Rajković K. Quantitative analysis of dendritic
morphology of the alpha and delta retinal ganglion cells in the rat: a cell classification study. J
Threor Biol. 2009; 259: 142-150.
23. Montague PR, Friedlander MJ. Expression of an intrinsic growth strategy by mammalian
retinal neurons. Proc Natl Acad Sci USA. 1989; 86: 7223-7227.
24. Montague PR, Friedlander MJ, Morphogenesis and territorial coverage by isolated
mammalian retinal ganglion cells. J Neurosci. 1991; 11(5): 1440-1457.
 set in dante monotype by
fabrizio serr a editore, pisa · roma.
printed and bound by
tipo gr afia di agnano, agnano pisano (pisa).

*
December 2014
(cz 2 · fg 21)