Application to Quantify Fetal Lung Branching on Rat Explants

Pedro L. Rodrigues

1,2

, Sara Granja

, António Moreira

1,2

, Nuno Rodrigues

3,4

and João L. Vilaça

1,3

Algoritmi Center, School of Engineering, University of Minho, Guimarães, Portugal

ICVS/3B’s, PT Government Associate Laboratory, Braga/Guimarães, Portugal

DIGARC, Polytechnic Institute of Cávado and Ave, Barcelos, Portugal

HASLab/INESC TEC, University of Minho, Braga, Portugal

Keywords: Branching Morphogenesis, Lung Development, Image Segmentation, Region-based Algorithms.

Abstract: Recently, regulating mechanisms of branching morphogenesis of fetal lung rat explants have been an

essential tool for molecular research. The development of accurate and reliable segmentation techniques

may be essential to improve research outcomes. This work presents an image processing method to measure

the perimeter and area of lung branches on fetal rat explants. The algorithm starts by reducing the noise

corrupting the image with a pre-processing stage. The outcome is input to a watershed operation that

automatically segments the image into primitive regions. Then, an image pixel is selected within the lung

explant epithelial, allowing a region growing between neighbouring watershed regions. This growing

process is controlled by a statistical distribution of each region. When compared with manual segmentation,

the results show the same tendency for lung development. High similarities were harder to obtain in the last

two days of culture, due to the increased number of peripheral airway buds and complexity of lung

architecture. However, using semiautomatic measurements, the standard deviation was lower and the results

between independent researchers were more coherent.

1 INTRODUCTION

Branching morphogenesis is fundamental to the

growth and development of several organs such as

lung, pancreas, salivary gland, mammary gland, and

kidney and prostate. During the last decades,

analysis of lung branching morphogenesis of fetal

rat explants grown in vitro has been an essential tool

to the research of the underlying molecular and

cellular development mechanisms (Muratore et al.,

2009); (Nogueira-Silva et al., 2008). Therefore, this

methodology has been widely used in many research

centres due to its stability and versatility.

The analysis of branching morphogenesis

involves monitoring lung development in explants

culture, using images acquired at 24-hours intervals

by a stereo microscope during a 5 day period. A

morphometric analysis is usually done to study the

differentiation and growth of lung explants structure.

It quantifies several parameters, such as the inner

and outer epithelial perimeter and area and the

determination of the number of peripheral airway

buds (Nogueira-Silva et al., 2008).

Currently, this analysis is obtained by manual

delineation using generic 2D curves software, which

leads to a time-consuming, dependent on user

expertise and error-prone procedure. Consequently,

this process often results in inaccurate measurements

and forbids the comparison among different

researchers results (Muehlethaler et al., 2008).

Several image processing strategies have been

proposed in the literature to quantify, classify and

segment cellular regions from different image

sources (Yu and Tan, 2009); (Farjam et al., 2007). In

addition, several authors proposed watershed based

algorithms to microscope image processing due to

its ability to produce closed cell contours (Debeir et

al., 2008); (Fan et al., 2008); (Mouelhi et al., 2011).

To best of our knowledge the previous

techniques were never applied to this issue. This

work presents an image processing application that

allows a semiautomatic determination of inner and

outer perimeter of lung braches of rat explants

cultures.

2 METHODS

All methods described below were implemented

under C++ and VTK (Visualization ToolKit).

L. Rodrigues P., Granja S., Moreira A., Rodrigues N. and L. Vilaça J..

Application to Quantify Fetal Lung Branching on Rat Explants.

DOI: 10.5220/0004220900670070

In Proceedings of the International Conference on Computer Vision Theory and Applications (VISAPP-2013), pages 67-70

ISBN: 978-989-8565-48-8

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

2.1 Pre-processing Stage

The aims of this stage are the reduction of the noise

that corrupts the image, while maintaining and

enhancing the relevant object boundaries and

selecting a ROI for further processment.

All lung explant images were acquired with an

Olympus SZX16 stereo microscope in RGB format.

These images were firstly converted to grayscale

values by averaging and normalizing the 3 RGB

components. Then, the grayscale image was input to

an anisotropic diffusion algorithm (implemented as

in (Black et al., 1998)). This algorithm depends on

three parameters, namely the number of iterations (it

= 80), edge parameter (σ=4.5) and a diffusivity

function (Tukey’s biweight as edge-stopping

diffusivity function g(x,σ), Equation 1) (values were

experimentally calculated).

(1)

The anisotropic diffusion algorithm worked as a

Gaussian filter for noise reduction, but with

perseverance of sharper boundaries and image

contours, producing uniformity in the output image

intensity.

Finally, the ROI was selected by removing the

existing noise around the lung explant, by cropping

its region using a labelling and thresholding

algorithm.

2.2 Image Partitioning

Since epithelial boundaries of lung explants are

often characterized by a significant change in image

intensity, an image partitioning yielding meaningful

regions is obtained by detecting all the image edges.

Watershed regions were labelled by a starting

point and follow the flow line, whose direction was

the gradient of intensity, to a local maximum (uphill

direction) or minimum (downhill direction). Both

directions were implemented, but only the downhill

direction was used due to its better performance in

the inner lung explants segmentation.

In the end, the whole image is segmented into

primitive regions where the boundaries of the

watersheds regions coincide with the ridges of the

gradient magnitude surface.

The results of the watershed algorithm are

oversegmented images as illustrated in Figure 1 (A)

and (B).

Figure 1: Result of the watershed partitioning in rat lung

explants.

2.3 Merging Procedure and Inner

Lung Explants Quantification

Although the probability that watershed region

boundaries correspond to boundaries of important

objects increases with oversegmentation, it can also

create many insignificant boundaries. This stage

describes how one dealt with this problem and the

inner lung explants, perimeter and area, were

segmented.

Briefly, this procedure consists on the

identification of regions edges and its connections

with similar intensities, ignoring all others regions

with wide mean intensities variations. It assumes

that:

1. All pixels within the same region are

homogeneous;

2. Regions within lung explant epithelial have

homogeneous intensity variations;

3. Regions within lung explant epithelial are

considerably different from other outside

neighbouring regions;

After establishing the image segmentation through

the application of the watershed algorithm, this

merging procedure performs feature identification in

each region, by identifying the: centroid; mean

intensity distribution (MID); minimum and

maximum values; region edges; edges region

neighbours and the intensity entropy of each region.

Subsequently, the following processing was

considered to produce the final merged and

segmented result:

The user selects one watershed region within the

lung explant epithelial – selected region (SR in red,

Figure 2 (A) and (C));

The boundary information of the SR was

determined in order to found common intersection

edges (IE

with i = 1,2,… n, Figure 2 - C) between

neighbourhood regions SRNH;

The MID of each new region (MID_SRNH) was

compared to the MID of the SR (MID_SR). If

MID_SRNH lays between MID_SR±10%, then the

VISAPP2013-InternationalConferenceonComputerVisionTheoryandApplications

(with i = 1,2,… n) is removed and the two

regions are merged; The 10% value was

experimentally calculated by manually evaluating

the mean intensity of the epithelial regions;

If the entropy value of the newly merged region

is not within an interval determined by a 90%

confidence level, which takes into account the pixel

intensity of the two previous regions, these regions

should not have been merged; therefore they are

unmerged and step 2 is repeated in the opposite

direction;

After the merging procedure, the SRNH with the

highest MID similarity value becomes the new SR

and step 2 and 4 are repeated (Figure 2 (B) and (D)).

This algorithm ends when no IE

between similar

MID and entropy neighbourhood is found.

Figure 2: Merging procedure overview.

2.4 Outer Lung Explants

Outcome of stage 2.1 was also input to a threshold

algorithm in order to determine the outer lung

explant area and perimeter, by simply counting the

pixels around the object and within it.

3 RESULTS

The suitability and validation of the algorithm was

conducted on stereo microscope images (Olympus

SZX16) acquired at the Life and Health Sciences

Research Institute (ICVS) of School of Health

Sciences, University of Minho -Portugal. These tests

were performed in 50 images, corresponding to

images of 10 sets of lung explants on each day of

culture. All images were previously segmented by

three experienced researchers and this manual

segmentation was used as reference for the

evaluation of the algorithm (some results in Fig. 3).

Each user manually segmented the same image two

times, and the mean and standard deviation of the

inner epithelial perimeter for all sets is shown in

Table 1.

Table 1: Inner epithelial perimeter results obtained by

three different users and by the semiautomatic algorithm.

Culture

Day

User

Interactive

Method

655

± 51

645

± 45

630

± 42

703

± 5

841

± 63

850

± 55

830

± 70

954

± 12

1565

± 196

1500

± 184

1586

± 177

1752

± 53

2202

± 313

2380

± 419

2331

± 450

2592

± 56

2482

± 298

2614

± 506

2800

± 357

2887

± 78

Table 2: Summary of the DSC values (%) for each culture

day.

Culture

Day:

1 2 2 4 5

DSC

(%)

92.85

±3.87

91.54

±6.12

90.87

±6.88

88.80

±9.96

86.72

±11.75

Figure 3: Branching morphogenesis in rat lung explants

culture system at each time point. The lines represents the

segmentation results of the inner (red) and outer (yellow)

epithelial perimeter obtained by the automatic algorithm.

The interactive results were also overlapped with the

manual results and its quality and performance were

evaluated using the Dice Similarity Coefficient

(DSC). The mean DSC, of the interactive method

with all the users, from the different days is shown

in Table 2, indicating a successful segmentation

(DSC > 0.85).

4 DISCUSSION

An application for image segmentation was

developed, providing assistance to the researcher

and enabling fast morphometric analysis of lung

explants. The total number of decisions to quantify

ApplicationtoQuantifyFetalLungBranchingonRatExplants

morphometric analysis was drastically reduced,

since the user only has to select one single watershed

region.

Generally, best results were obtained in the first

two days of culture with lesser standard deviations.

High similarities between manual and semiautomatic

procedure were harder to get in the last two days of

culture, due to the increased number of peripheral

airway buds and complexity of lung architecture.

The standard deviation of the interactive method was

null after all researchers selected the same start

region to begin the merging process.

The segmentation rate depended on the number

of regions needed to be merged to select the entire

region lung epithelial. However, in all cases the

interactive segmentation time was less than the

manual one (34±7% of the manual time).

The merging procedure was essential to achieve

a good segmentation, since a lot of regions were

firstly created by a watershed algorithm.

Sometimes, the presented method produced

undermerged regions due to ambiguity and lack of

definition of the inner lung explants contours. These

cases increase the probability of merging dissimilar

regions and incoherence between boundaries of

some watershed regions and boundaries of lung

explants inner contours.

5 CONCLUSIONS

Regulating mechanisms of branching morphogenesis

of fetal lung rat explants have been an essential tool

for molecular research. The application of this work

provides a technique for lung rat explants

segmentation and analysis by selecting only one

watershed region belonging to the inner lung

epithelial. The total number of decisions, time-

consumption and user dependence were significantly

decreased.

Further work is needed regarding the merging

procedure and the development of image

enhancement techniques to improve inner lung

epithelial contrast, mainly in the last days of culture,

in order to decrease the standard deviation of results

and increase its reliability. Moreover, a new

algorithm must be developed for counting the

number of peripheral airway buds of lung explants.

ACKNOWLEDGEMENTS

The authors acknowledge to Foundation for Science

and Technology (FCT) - Portugal for the fellowships

with the references: SFRH/BD/74276/2010,

SFRH/BD/68270/2010, SFRH/BPD/46851/2008 and

SFRH/BD/51062/2010.

REFERENCES

Black, M. J., Sapiro, G., Marimont, D. H., Heeger, D.,

1998. Robust anisotropic diffusion. Ieee T Image

Process 7, 421-432.

Debeir, O., Adanja, I., Warzee, N., Van Ham, P.,

Decaestecker, C., 2008. Phase contrast image

segmentation by weak watershed transform assembly.

I S Biomed Imaging, 724-727.

Fan, D., Cao, M. Y., Lv, C. Z., 2008. Wall-adherent cells

segmentation based on cross-entropy and watershed

transform. 2008 International Conference on Audio,

Language and Image Processing, Vols 1 and 2,

Proceedings, 1703-1707.

Farjam, R., Soltanian-Zadeh, H., Jafari-Khouzani, K.,

Zoroofi, R.A., 2007. An image analysis approach for

automatic malignancy determination of prostate

pathological images. Cytometry Part B: Clinical

Cytometry 72B, 227-240.

Mouelhi, A., Sayadi, M., Fnaiech, F., 2011. Automatic

segmentation of clustered breast cancer cells using

watershed and concave vertex graph,

Communications, Computing and Control

Applications (CCCA), 2011 International Conference

on, pp. 1-6.

Muehlethaler V, K. A., Seedorf G, Balasubramaniam V,

Abman SH., 2008. Impaired VEGF and nitric oxide

signaling after nitrofen exposure in rat fetal lung

explants. Am J Physiol Lung Cell Mol Physiol. 294,

110-130.

Muratore CS, L. F., Zhou Y, Harty M, Reichner J, Tracy

TF., 2009. Endotoxin alters early fetal lung

morphogenesis. J Surg Res 155, 225.

Nogueira-Silva C, M. R., Esteves N, Gonzaga S, Correia-

Pinto J., 2008. Intrinsic catch-up growth of

hypoplastic fetal lungs is mediated by interleukin-6.

Pediatr Pulmonol 43, 680-689.

Yu, J., Tan, J., 2009. Object density-based image

segmentation and its applications in biomedical image

analysis. Computer Methods and Programs in

Biomedicine 96, 193-204.

VISAPP2013-InternationalConferenceonComputerVisionTheoryandApplications