WoundArch: A Hybrid Architecture System for Segmentation and

Classification of Chronic Wounds

Carlos Diego F. da Rocha

Bruno M. Carvalho

, Vítor G. Marques

and Bruno S. Silva

Informatics and Applied Mathematics Dept., Fed. Univ. of Rio Grande do Norte, Natal, Brazil

Informatics Institute, Fed. Univ. of Rio Grande do Sul, Porto Alegre, Brazil

Digital Metropolis Institute, Fed. Univ. of Rio Grande do Norte, Natal, Brazil

Keywords: Chronic Wounds, Devices Furniture, Watershed, Image Processing.

Abstract: Every year, millions of people are affected by wounds worldwide. The wound treatment process is costly and

requires the nurse to perform activities during patient care: tissue classification and calculation of the wound

area. Thus, this work proposes to build a hybrid computer system with two configurations to support wound

care. The first configuration uses a smartphone to perform the capture, segmentation and classification of the

wound images. The other configuration has a client-server architecture, the images are captured and segmented

in the application and sent, via the internet, to the web server, which is responsible for classifying the tissue

of the wounds. The proposed methodology is the segmentation of images using the watershed algorithm and

classification of tissues in Necrosis, Granulation or Crushing. fulfilled. To evaluate the application,

experiments were performed with 20 images of wounds and the system was evaluated in two architectures:

client and client-server. The results show that the client-server reached accelerations of up to 3.2 times in

relation to the client-only architecture. The client-server architecture also saves energy and space in the client

units, increasing the uptime of smartphones, in addition to reducing the storage load of the same.

1 INTRODUCTION

A wound is an interruption in the continuity of tis- sues

by external mechanical force by physical force, but

also damage by heat, cold, chemicals, electricity and

radiation (Pounder, 1969). Wounds can be clas- sified,

according to tissue repair time, in acute and chronic

(Whitney, 2005). Acute wounds are those that heal in

a timely manner and without complications, while

chronic wounds take longer to heal due to their

extension and severity, and usually present

inflammation in the wound tissues (Clark, 2014). The

causes for their existence are usually associated with

pre- existing diseases (comorbidities), such as

diabetes and venous insufficiency, for example, and

complications such as infections that hinder the

healing process. Wound is a worldwide health

problem and affects a large number of people,

compromising their quality of life. According to Sen

https://orcid.org/0000-0003-4018-664X

https://orcid.org/0000-0002-9122-0257

https://orcid.org/0000-0000-0000-0000

https://orcid.org/0000-0002-7689-8000

et al. (Sen et al., 2009), annually, in the United States

alone, as chronic wounds affect 6.5 million patients,

generating an estimated expenditure of U$ 25 billion

annually in the treatment of chronic wounds; In the

Netherlands, annual costs for pressure ulcer

treatments range from U$ 362 mil- lion to U$ 2.8

billion, representing 1% of the Nether- lands budget

(Brem and Lyder, 2004). These figures indicate the

magnitude of the problem with significant

socioeconomic repercussions. Chronic wound

treatment incur into high costs for health services and

the patient, especially when referring to chronic

wounds due to increased prolongation of treatment

(Nussbaum et al., 2018).

The nursing process (Garcia and Nóbrega, 2009)

guides the performance of these health professionals

through steps that should be developed during patient

care, namely: collection of nursing data, nursing

diagnosis, nursing planning, implementation and

F. da Rocha, C., Carvalho, B., Marques, V. and Silva, B.

WoundArch: A Hybrid Architecture System for Segmentation and Classiﬁcation of Chronic Wounds.

DOI: 10.5220/0010341406510658

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 5: HEALTHINF, pages 651-658

ISBN: 978-989-758-490-9

651

evaluation. In the context of wound treatment

(Kordestani, 2019), this process can be specialized as

follows: the professional begins analyzing the

patient’s health history, wound history and previously

performed treatments; then it analyzes the current

state of patient health conditions and wound

conditions (tissue types, size, location, etc.) to make a

diagnosis. Subsequently, the professional performs

wound and scar treatment, possibly with debridement,

cleaning, med- ication (or equivalent intervention) and

maintenance of moist tissue. Finally, the nurse

performs the procedure of the coverage necessary for

the wound and records the information about the

treatment (Figure 1).

Figure 1: Steps of the nursing process.

In the context of wound treatment, one of the

fundamental steps to arrive at a diagnosis is the

identification of characteristics of the lesions, such as:

the causal agent, the depth, the shape, the size, the

amount of exudate, the location, the appearance and

treatment environment (Dealey, 2008). Also

according to the author, the objective of the

evaluation is to extract information about the stage of

the wound and, consequently, its follow-up based on

the organism’s response to the treatment being

performed. The correct identification of these

variables is highly dependent on the professional’s

expertise. In this way, it is based on the individual

ability to visually assess how injuries have been

shown to be susceptible to errors.

In this context, in order to reduce human

limitations, computer systems have received great

attention as a work tool in the health area, mainly for

the power of information processing, for the ability to

work with a large volume of information and for the

convergence between media and devices. This type of

sup- port provides greater precision and agility at

work, helps, personalizes and expands the activities to

be performed by health professionals (Tibes et al.,

2015). In the literature, it is possible to find some

studies that present proposals for informational

systems that sup- port wound care.

Different scientific studies have been developed

focusing on the use of smartphone-web server plat-

forms in nursing diagnoses in patients with wounds.

Wallis et al. (Wallis et al., 2016) shows the use of a

mobile application that captures images of wounds

due to burns and sends them to the server so that

doctors can evaluate and make the diagnosis. Sir-

azitdinova and Deserno (Sirazitdinova and Deserno,

2017) describe a wound assessment system that uses

a smartphone device to capture images and a server

that processes, stores and reconstructs wounds in 3D

models.

In this work, we propose the use of a client-server

application to perform a necessary need for chronic

wounds, a classification of their tissues in splinter-

ing, granulation and necrosis. The construction app

will allow not only to support wound assessment, but

also offer a tool that will help prevent complications

that can result in early amputations. In addition to

facilitating the work of nurses in the classification of

wound tissues and presenting the data regarding the

injury in an appropriate way. Nursing professionals

increasingly have mobile devices that allow them to

capture images of wounds. Thus, knowing and im-

proving the efficiency of processing wound images on

mobile devices is essential for scientific knowl- edge

to reach the market and make sense in the reality of

nursing professionals. The literature still lacks re-

search that evaluates the execution time and battery

consumption of wound segmentation and classifica-

tion systems on mobile devices.

2 RELATED WORK

During the development of this project, several

systems that have similar objectives to those proposed

in the work were studied. At this stage, some of the

systems found in the literature will be described,

pointing their practical motivations and their respec-

tive limitations. In particular, we consider the works

of (Cohen and Bard, 2015), (DigitalMedLab, 2019),

(Ciancio et al., 2016) and (Friesen et al., 2016). We

also searched the literature for studies that evalu- ated

the efficiency of computer systems in supporting

wound treatment. Three nursing studies were found

that assessed efficiency and other aspects of quality of

systems aimed at end users: (Sperandio, 2008), (Tibes

et al., 2015) and (Oliveira and Peres, 2015).

Cohen and Bard (Cohen and Bard, 2015) describe

a mobile application developed at the Worcester Poly-

technic Institute, called Sugar, to help people with

diabetes manage their blood sugar level and the state

of chronic foot ulcers. The app is available only for

Android smartphones and communicates via wire- less

internet to the patient’s glucose meter in order to track

blood sugar levels and weight. The app uses the

device’s camera to capture and analyze images of

HEALTHINF 2021 - 14th International Conference on Health Informatics

652

chronic foot injuries. It tracks the wound area and

healing state and then reports the information in an

easy format for patients. The app can be used only to

analyze injuries in the foot region.

DigitalMediaLab developed WoundDesk

(DigitalMedLab, 2019), a mobile solution for wound

management, which, according to them, saves 60% of

documentation time, medical errors and increased

wound healing improvement. The application has

wound evaluation features, automated analysis, semi-

automatic and sterile wound measurement, and

documents patient information. The app is available

for both Android and IOs for free. The calculation of

the area is semi-automatic, as you need to place a ruler

next to the wound so that the processing algorithm

can calculate a coarse estimation of the PU area. This

can lead to risks of wound contamination.

In Friesen (Friesen et al., 2016), the development

of the SmartWoundCare app is described, designed to

document and evaluate chronic wounds through An-

droid and iOS smartphones and tablets. The

application uses the Braden scale to predict the risks

of wounds, evaluates and stores all patient

information on the device in a standalone mode,

without the patient’s registration function. In

addition, the application does not analyze the area of

the lesion and does not generate a report of the data

collected from patients.

Ciancio et al. (Ciancio et al., 2016) describes the

MOWA - Mobile Wound Analyzer app. The applica-

tion performs the identification of the existing tissues

in a wound from the capture of an image of the site.

With this, it provides a list of care suggestions. Like

WoundDesk, MOWA calculates the area of the lesion

with the aid of a ruler, making the procedure

vulnerable to contamination of the region and it can

produce bad estimations. Because it is an app without

new up- dates its interface is to be desired and is

probably no longer compatible with some versions of

Android and IOs systems.

In Oliveira (Oliveira and Peres, 2015), the study

aims to evaluate the functional performance and tech-

nical quality of the Electronic Documentation System

of the Nursing Process at the University Hospital of

the University of São Paulo, called PROCEnf-USP.

The evaluation was based on the Quality Model of the

standard 25010 and the Evaluation Process de- fined

in standard 25040 and used the following parameters:

functional suitability, reliability, usability,

performance efficiency, compatibility, security,

maintainability and portability.

In Sperandio (Sperandio, 2008), the research

sought to assess the functional performance and

technical quality of the prototype software developed

for nursing assistants. The software developed for

mobile devices with integrated wireless network

interface, al- lows nurses to access information and

document data about patients at the bedside. Said

author used the Evaluation Process Model provided in

the ISO / 9126

standard, which deals with the external and

internal quality of the software. This model uses six

quality evaluation parameters, namely: functionality,

reliability, usability, efficiency, maintainability and

portability. However, in his study, the evaluation of

the quality attributes of the prototype software was

based on usability and efficiency.

In Tibes (Tibes et al., 2015), the UpCare applica-

tion was evaluated by a committee of computer spe-

cialists and nursing specialists. The evaluation was

based on aspects of functionality, reliability, usabil-

ity, efficiency, maintainability and portability. For

this, two questionnaires were developed, based on the

study by (SPERANDIO, 2008), which evaluate these

aspects.

Although it is possible to find studies that have

evaluated the efficiency of in supporting the treatment

of wounds, there is a lack in the literature regarding

regarding works that address battery consumption and

run time of image segmentation and classification al-

gorithms in classification systems of wounds. The pa-

pers did not address wound segmentation and classi-

fication algorithms, and did not analyze consumption

battery life of the proposed systems.

3 WoundArch ARCHITECTURE

The present system follows the client-server ar-

chitecture shown in Figure 2, where the client uses its

mobile device to capture and segment the image of the

wound, while the server is responsible for the

classification of the tissues and the computation of the

tissues’ areas.

In this work, we built on the work proposed by

Godeiro et al. (Godeiro et al., 2018), which per- forms

the segmentation and classification of chronic

wounds. Figure 3 describes the steps performed for

the execution of the segmentation and classification

steps, which we will describe shortly.

3.1 Image Acquisition

At this stage, a specific protocol is defined for captur-

ing the images by using the mobile device, in order to

avoid the presence of noise as well as the presence of

other parts of the human body that may interfere with

the image preprocessing process. In addition, the

WoundArch: A Hybrid Architecture System for Segmentation and Classiﬁcation of Chronic Wounds

653

Figure 2: Client-server architecture.

Figure 3: System flow chart.

following protocol aims to ensure equal lighting

conditions for all photos taken. Godeiro (Godeiro,

2018) defined the following acquisition protocol:

Take the photo at a distance between 30cm and

40cm from the wound;

Take the photo with a white or blue background

to avoid adding unwanted objects in the

background (e.g. parts of the human body or

other object other than the wound);

Take the photo without flash to avoid adding

extra brightness to the image;

When taking the photo, place an object with a

known size in the image, at the same depth of the

wound, to retrieve the scale of the pixels;

When taking the photo, use white lighting to

avoid variations in image colors;

When taking the photo, make sure that the entire

wound is equally illuminated to avoid shadows

in the image.

3.2 Preprocessing

With the photo of the wound acquired, the applica-

tion performs a color conversion of the RGB image

to the HSV color system. The HSV model is used

when it is necessary to identify colors very similar to

other colors (Cardani, 2001). Unlike the RGB color

system, whose color determinants are all parameters

(red, green and blue), in the HSV model the color

determinant is the hue. In our tool, the H and S

components are used to detect skin color and remove

the parts of the image that do not contain skin or

wound. Then, the result obtained is converted from

the HSV to the RGB color system.

3.3 Segmentation

At this stage, the objective is to identify the region

of interest of the image so that the algorithm can

eliminate the regency where there is no presence of

lessions on the skin. For this, the Watershed

segmentation method (Meyer, 1992) is used in the

application. This region-based segmentation method

delivers results with closed and well-defined

contours, which is great importance to the image

segmentation process.

This is a semi-automatic process, that receives

user markings on the smartphone screen as an in- put

to the segmentation method. This process is also

used so the user can isolate a wound for processing,

in the cases where more than one wound is present

in the same picture.

3.4 Spatial Color Reduction

In the last step on the client side, a color reduction of

the image is performed. First, each RGB color com-

ponent of each pixel is quantized using 6 bits by divid-

ing its value by 4. With this reduction, the histogram

size changes from 256

possible colors to 64

possible

colors. After this, the image in the RGB format is

converted to the CIELab color space because it was

designed to approximate the human vision, as a

perceptually uniform model, i.e., the distance between

colors in this space are approximately proportional to

the color distances perceived by humans. Then, the

HEALTHINF 2021 - 14th International Conference on Health Informatics

654

frequency of colors in the histogram is computed and

those with at least 0.05% instances are stored in a list

of representative colors. The colors of the image are

analyzed and those that were not saved in the list, i.e.,

not labeled as representatives, will be mapped into the

closest color based on the Euclidean distance in the

CIELab space.

After this color space reduction, the image is ready

for the classification step, that can be performed

locally or remotely by a server through HTTP re-

quests that manage and forward the image data so that

it can be processed. The choice of the model used for

processing, being processed completely on the client

side, or divided between the client and server depends

on smartphone capabilities such as processing power

and available storage, as well as network bandwidth

and also security issues related to the available wire-

less networks.

3.5 Classification

Upon receiving the image, the server is responsible for

classifying the tissues present in the wound as

Necrosis, Granulation, or Slough. For this, a

clustering is carried out by computing the Earth

Mover’s Distance (EMD) (Rubner et al., 2000). This

distance measures the difference between signatures

that are compact distribution representations by

comparing two histograms and verifying how

different these histograms are.

In order to do that, a set of 30 wound images was

used to serve as a training base. During the training

phase, we used patches 11 × 11, a size determined

empirically (Godeiro et al., 2018), and used the EMD

to retain only patches which were at a reasonable

distance from the other retained patches. We also use

an heuristic to reduce the number of patches per class

that are used for comparisons in the classification, by

keeping only the patches which have larger sum of

distances from the other patches of the class. By

doing this, we select only the more external elements

of the cluster representing a class, i.e., the patches the

define the external borders of the clusters.

Each patch of the analyzed image is compared by

EMD with the entire set of patches retained in the

training step. The tissue label associated with a pixel

is defined by the smaller distance, computed using the

EMD, between the patch centered at it and the

training set patches.

4 EXPERIMENTS

All the experiments performed here were developed

on a notebook equipped with an Intel Core i5

Processor (6th Generation) Model 6200U with 2

cores running at 2.3GHz, with 3MB of cache, 12GB

DDR3L 1600 MHz RAM and running Linux Mint

19.3.

For the development of the server, the PHP

language Version 7.2 was used, along with the web

text editor PHPStorm 2019.3.4. The Laravel

Framework for Web Artisans 6.0 was used in

conjunction with the dependency manager Composer

1.10.1. For the computer vision stage, the OpenCV

4.1 library with C++ was used. For the development

and emulation of the client, we used the Kotlin

Language 1.3. The client used to perform the tests was

the Redmi Note 8 Smartphone equipped with a

Qualcomm SDM665 Snapdragon 665, an Octa-core

(4 ×2.0 GHz Kryo 260 Gold & 4x 1.8 GHz Kryo 260

Silver) CPU, an Adreno 610 GPU, and 64GB of

storage and 4GB RAM. This smartphone has a

Quadruple Camera: 48 MP, f/1.8, (wide), 1/2”, 0.8µm,

PDAF + 8 MP, f/2.2, 13mm (ultrawide), 1/4”, 1.12µm

+ 2 MP, f/2.4, 1/5”, 1.75µm (dedicated macro camera)

+ 2 MP, f/2.4, 1/5”, 1.75µm, depth sensor; 6.3 inches;

with a maximum resolution of 1080 × 2340 pixels.

In this section, the prototype of the application is

described (see Figure 4). The application interface is

very lean and objective so that the nurse can ob- tain

the wound information with a few touches on the

screen. When opening the app, the nurse attaches the

image of the wound, either by a photo captured at that

moment or by an image from the gallery of the smart-

phone. The patient’s name is then filled in and then

the image is uploaded to the server or processed

locally.

5 RESULTS AND DISCUSSION

For the purpose of validation of the client-server

system, a base with 20 images of wounds was used.

Table 1 presents the results obtained by the system

divided into two scenarios: segmentation and

classification in the client architecture; and

segmentation on the client with classification of the

wound on the server. In order to verify the efficiency

between client and client-server, the execution times

measured in sec- onds were compared. Each image

was executed five times for each scenario. The

Segmentation, Classification, and Total columns

show the average times and standard deviation

obtained. Even though the images are ordered by their

total number of pixels, we can see that the second

smallest image, 650 × 375, has the third highest

execution average, 9.718 seconds, a circumstance that

is related to the size of the wound area being analyzed.

WoundArch: A Hybrid Architecture System for Segmentation and Classiﬁcation of Chronic Wounds

655

Figure 4: Example of the mobile application screen: processing seen on the client’s App.

Figure 5: Execution times required for the segmentation (blue) and classification (red) tasks. The bars labelled C correspond

to the execution times when both tasks were performed on the client side, while the label CS indicates that the classification

task was performed on the server side.

In the client-only application, when analyzing the

results of segmentation and classification, we can see

that the classification time is roughly twice as long as

the segmentation time. In classification, the vast ma-

jority of the results were below 10 seconds, only im-

ages 19 and 20 were classified on 21.584 and 17.990

seconds, respectively.

In the client-server application, due to the

segmentation continuing to be performed on the

client, the values showed little variations in relation to

the exe- cution time of the client application, that can

be ex- plained by different computational loads being

exe- cuted on the smartphone. In the classification

column, Images 19 and 20 achieved the largest time

reduction compared to the client application, from

14.69 sec- onds to 0.70 and 12.18 to 0.55, respectively.

Actually, all classifications in the server were

performed in un- der a second, including the

transmission time of the files in both directions.

The graph in Figure 5 shows the execution times

for the two options of the proposed architecture,

where the blue bars represent the segmentation time

and the red bars represent the classification times. An-

alyzing the columns, it can be noted that the longest

execution time was the one for Image 19 performed in

the client application, with more than 21 seconds,

while it took 8.01 seconds for the client-server

architecture, followed by Image 20, with 17.99

seconds for the client application, and 5.57 seconds

for the client- server.

A similar pattern can be observed for Images 2,

11, 12 and 17. The speedups obtained by the usage of

the client-server architecture in relation to the client-

only varied from 1.35 to 3.23 times. Therefore, it can

be easily observed on Table 1 and Figure 5 that the

Client-Server architecture proposed is advantageous

when compared to a standalone, or as it is called here,

client-only architecture for solving this problem.

HEALTHINF 2021 - 14th International Conference on Health Informatics

656

Table 1: Execution times (in seconds) for the segmentation and classification. The Classification* times reported for the

Client-Server system includes the transmission times of the segmented image to the server and the classification map back to

the client.

Image Client Client‐Server

 Size Segmentation Classification Total Segmentation Classification* Total



500× 435 0.45± 0.84 0.84± 0.21 1.29± 0.20 0.51± 0.04 0.45± 0.44 0.96± 0.46



65× 375 3.10± 0. 22 6.62± 0.03 9.72± 0.21 3.07± 0.19 0.62± 0.30 3.69± 0.35



800× 500 0.84± 1.17 1.17± 0.09 2.01± 0.19 0.88± 0.06 0.60± 0.52 1.48± 0.58



900× 700 1.95± 0.34 2.58± 0.32 4.53± 0.66 2.10± 0.26 0.57± 0.44 2.67± 0.44



1000× 667 0.82± 0.09 1.58± 0.07 2.40± 0.15 0.91± 0.12 0.40± 0.35 1.31± 0.40



1000× 669 1.50± 0.18 3.05± 0.04 4.54± 0.21 1.47± 0.04 0.69± 0.40 2.16± 0.43



1024× 703 0.78± 0.04 1.50± 0.06 2.28± 0.09 0.80± 0.05 0.29± 0.09 1.10± 0.14



1024× 768 0.89± 0.03 1.52± 0.05 2.41± 0.06 1.06± 0.10 0.24± 0.06 1.30± 0.14



1102× 575 1.54± 0.18 3.27± 0.40 4.81± 0.57 1.61± 0.02 0.45± 0.27 2.05± 0.28

10

1111× 640 1.03± 0.10 3.38± 0.23 4.41± 0.30 1.10± 0.07 0.54± 0.39 1.64± 0.39

11

999× 1228 2.00± 0.23 5.15± 0.11 7.16± 0.33 1.94± 0.07 0.80± 0.55 2.74± 0.54

12

1280× 960 2.15± 0.28 4.54± 0.21 6.69± 0.34 2.12± 0.10 0.40± 0.07 2.52± 0.07

13

1333× 885 1.49± 0.05 3.24± 0.05 4.73± 0.09 1.43± 0.09 0.69± 0.46 2.13± 0.49

14

1400× 931 1.42± 0.03 3.13± 0.03 4.55± 0.09 1.47± 0.08 0.32± 0.11 1.79± 0.15

15

1460× 980 1.75± 0.13 2.59± 0.02 4.34± 0.14 1.64± 0.05 0.40± 0.23 2.04± 0.26

16

1600× 800 1.35± 0.08 2.73± 0.01 4.08± 0.08 1.35± 0.15 0.65± 0.40 2.00± 0.48

17

1637× 1255 2.26± 0. 13 4.23± 0.03 6.50± 0.16 2.55± 0.14 0.53± 0.40 3.07± 0.40

18

1850× 870 1.46± 0.02 2.69± 0.04 4.15± 0.05 1.48± 0.06 0.44± 0.42 1.92± 0.44

19

1505× 1852 6.90± 0. 50 14.69± 0.29 21.58± 0.23 7.31± 0.28 0.70± 0.13 8.01± 0.22

20

2048× 1536 5.81± 0. 45 12.18± 0.37 17.99± 0.66 5.01± 0.09 0.55± 0.15 5.57± 0.21

Moreover, the usage of the proposed client-server

architecture allows the system to keep synchronized

by saving the processed images and generating and

storing reports about the wound treatment evolution.

One can also point out to the fact that the computer

used as a server is far from being a high end PC, and

so, the usage of a more powerful computer can offer

much higher throughputs.

6 CONCLUSIONS

Mobile apps have proven to be important tools with

great potential for benefits for health activities. In this

work, we propose a mobile application for fol- lowing

the treatment of chronic wounds. This task is achieved

by performed the segmentation and classification of

chronic wound tissues and keeping track of the

treatment evolution. The main objective is to pro- vide

health professionals with a tool capable of accu-rately

classify the areas of these lesions occupied by slough,

granulation and necrosis tissues.

In order to validate the work, we performed exper-

iments with the segmentation and classification tasks

organized into a client-only and a client-server archi-

tectures. The results show the advantage of the client-

server architecture over the client-only, with speedups

of up to 3.2 times being achieved. The client-server

architecture also saves power and space on client

units, thus, increasing the up time of the smartphones

used as clients as well as decreasing the storage

burden on them. This allows for the usage of lower

end smartphone as clients.

The work developed here is part of a proposed

mobile application for wound care. Future work in-

clude incorporating a three-dimensional re-

construction module in order to make the area

measurements more accurate and provide information

about the wounds’ depths. We also plan to develop a

follow-up module that will store the treatment

progress and a diagnostic module that will suggest

potential treatments based on historic data. Finally, it

is necessary to improve the graphical interface to

improve its usability.

ACKNOWLEDGMENTS

The authors would like to thank the financial support

provided by the Coordenação de Aperfeiçoamento de

Pessoal de N´ıvel Superior Brasil (CAPES) Finance

Code 001, during the development of this work. BMC

would also like to thank the CNPq INCT-MACC for

its support.

WoundArch: A Hybrid Architecture System for Segmentation and Classiﬁcation of Chronic Wounds

657

REFERENCES

Brem, H. and Lyder, C. (2004). Protocol for the successful

treatment of pressure ulcers. The American Journal of

Surgery, 188(1):9–17.

Cardani, D. (2001). Adventures in HSV space. Labora-

torio de Robo´tica, Instituto Tecnolo´gico Auto´nomo

de Mexico.

Ciancio, F., Portincasa, A., Parsi, D., and Innocenti, A.

(2016). Mowa®: A simple and economic way of mon-

itoring chronic wounds outcome with your mobile de-

vices. Annali Italiani di Chirurgia, 88(1):88–94.

Clark, R. A. (2014). Chapter 76 - Wound repair: Basic bi-

ology to tissue engineering. In Lanza, R., Langer, R.,

and Vacanti, J., editors, Principles of Tissue Engineer-

ing (Fourth Edition), pages 1595 – 1617. Academic

Press, Boston, fourth edition.

Cohen, M. and Bard, M. (2015). Pilot clinical study at

UMMS to test ‘sugar’ diabetes app. Available at

https://www.umassmed.edu/news/news-

archives/2015/04/pilot-clinical-study-at-umms-to- test-

sugar-diabetes-app/.

Dealey, C. (2008). The Care of Wounds: A Guide for

Nurses. Wiley-Blackwell.

DigitalMedLab (2019). Wounddesk: The mobile solution

for wound care. Available at https://wounddesk.com/.

Friesen, M. R., Gigliotti, B., and Poon, T. W. K. (2016).

An mHealth technology for chronic wound

management. Mobile Health Technologies: Theories

and Applica- tions, page 115.

Garcia, T. R. and No´brega, M. M. L. d. (2009). Proceso de

enfermer´ıa: De la teor´ıa a la pra´ctica asistencial y de

investigacio´n. Escola Anna Nery, 13(1):816–818.

Godeiro, V. (2018). Ulcer segmentation and tissue classi-

fication using color texture clustering. B.S. Thesis,

Universidade Federal do Rio Grande do Norte.

Godeiro, V., Neto, J. S., Carvalho, B., Santana, B., Ferraz,

J., and Gama, R. (2018). Chronic wound tissue classi-

fication using convolutional networks and color space

reduction. In 2018 IEEE 28th International Workshop

on Machine Learning for Signal Processing (MLSP),

pages 1–6.

Kordestani, S. S. (2019). Atlas of Wound Healing E-Book:

A Tissue Regeneration Approach. Elsevier.

Meyer, F. (1992). Color image segmentation. In 1992 In-

ternational Conference on Image Processing and its

Applications, pages 303–306. IET.

Nussbaum, S. R., Carter, M. J., Fife, C. E., DaVanzo, J.,

Haught, R., Nusgart, M., and Cartwright, D. (2018). An

economic evaluation of the impact, cost, and medicare

policy implications of chronic nonhealing wounds.

Value in Health, 21(1):27–32.

Oliveira, N. B. d. and Peres, H.H.C. (2015). Avaliação do

desempenho funcional e qualidade te´cnica de um

sistema de documentação eletronica do processo de

enfermagem. Revista Latino-Americana de Enfer-

magem, 23(2):242–249.

Pounder, D. (1969). Lecture notes on forensic medicine.

Annals of Internal Medicine, 71(4):886.

Rubner, Y., Tomasi, C., and Guibas, L. J. (2000). The Earth

mover’s distance as a metric for image retrieval. Inter-

national Journal of Computer Vision, 40(2):99–121.

Sen, C. K., Gordillo, G. M., Roy, S., Kirsner, R., Lambert,

L., Hunt, T. K., Gottrup, F., Gurtner, G. C., and Lon- gaker,

M. T. (2009). Human skin wounds: A major and

snowballing threat to public health and the econ- omy.

Wound Repair and Regeneration, 17(6):763– 771.

Sirazitdinova, E. and Deserno, T. M. (2017). System de-

sign for 3D wound imaging using low-cost mobile de-

vices. In Medical Imaging 2017: Imaging Informatics

for Healthcare, Research, and Applications, volume

10138, page 1013810. International Society for Op- tics

and Photonics.

Sperandio, D. J. (2008). A tecnologia computacional

mo´vel na sistematizac¸a˜o da assisteˆncia de

enfermagem: avaliac¸a˜o de um software-proto´tipo.

PhD thesis, Uni- versidade de Sa˜o Paulo.

Tibes, C. M. S., Cherman, E., Souza, V., Westin, U., Zem-

Mascarenhas, S., and Evora, Y. D. M. (2015).

Avaliac¸a˜o de um aplicativo para apoio a` decisa˜o

no cuidado de u´lceras por pressa˜o. In Anais do

Congresso Internacional de Informa´tica Educativa

(TISE), pages 191–199.

Wallis, L. A., Fleming, J., Hasselberg, M., Laflamme, L.,

and Lundin, J. (2016). A smartphone app and cloud-

based consultation system for burn injury emergency

care. PLoS One, 11(2):e0147253.

Whitney, J. D. (2005). Overview: Acute and chronic

wounds. Nursing Clinics, 40(2):191–205.

HEALTHINF 2021 - 14th International Conference on Health Informatics

658