AN AGENT-BASED ARCHITECTURE FOR CANCER STAGING

Miguel Miranda, Ant

onio Abelha, Manuel Santos, Jos

e Machado and Jos

e Neves

Uiversidade do Minho, Departamento de Inform

atica, Campus de Gualtar, Braga, Portugal

Keywords:

Group Decision Support Systems, Artiﬁcial Intelligence and Medicine.

Abstract:

Cancer staging is the process by which physicians evaluate the spread of cancer. This is important, once in

a good cancer staging system, the stage of disease helps to determine prognosis and assists in selecting ther-

apies. A combination of physical examination, blood tests, and medical imaging is used to determine the

clinical stage; if tissue is obtained via biopsy or surgery, examination of the tissue under a microscope can

provide pathologic staging. On the other hand, good patient education may help to reduce health service costs

and improve the quality of life of people with chronic or terminal conditions. In this paper it is endorsed a

theoretical based model to support the provision of computer based information on cancer patients, and the

computational techniques used to implement it. One’s goal is to develop an interactive agent based computa-

tional system which may provide physicians with the right information, on time, that is adapted to the situation

and process-based aspects of the patients’s illness and treatment.

1 INTRODUCTION

A healthcare unit may be approached as a distributed

computational environment, where different services

(here conceived as agents or multi-agent systems),

and people, need to communicate, exchanging data

and knowledge (e.g., cirurgical experiencies). Indeed,

whenever one is faced with a problem, the use of a

multidisciplinary team presents in itself as a solution

for the extension and complexity of the intersecting

medical speciﬁes. Such a team, acting as a group

decision supported one, built on the base of human

beings and software agents, will intertwine all health-

care related areas. Its role will be crucial on the em-

ulation of expert behavior and to ease of the intricacy

of the diagnose process.

The explained architecture depends on the devel-

opment of key cornerstones which make the system

possible and with meaning, once a scheduled system

has no utility in a distributed environment. First of all,

in an healthcare unit one cannot allow the heterogene-

ity of the methodologies for problem solving and pro-

cedurals inherent to each medical speciality act to dis-

turb the ﬂow of data and information among the dif-

ferent partners. Therefore, a platform for integration,

archiving and diffusing of information is required,

making all resources accessible to everyone and from

elsewhere in the environment, while not requiring any

considerable alteration of any of the medical area

speciﬁcities. The next phase on the development of

the architecture referred to above is concerned with

the use of web based interfaces as means towards ho-

mogeneity and platform independency when storing,

accessing, communicating and displaying medical in-

formation. The Electronic Health Record (EHR) is

such an application which presents the information

of the patient according to their particular problems

and manages the front-end interaction and inputs with

the speciﬁc end-user (e.g., physician, nurse, auxil-

iary personnel). The modularity inherent to the men-

tioned EHR interface is essential in order to prop-

erly integrate and establish a Group Decision Sup-

port System (GDSS), enabling workﬂows, scheduling

and structured patient information, leading towards

a better quality of service. The integration platform

as well as the GDSS architecture considered in this

work are designed based on an agent or multi-agent

based approach to problem solving, taking into con-

sideration not only the intricacy and heterogeneity of

both healthcare software solutions and human interac-

tion, but also the scalability and integration of differ-

ent technologies and methodologies for problem solv-

ing.

Due to the arising need for such an environment

313

Miranda M., Abelha A., Santos M., Machado J. and Neves J. (2009).

AN AGENT-BASED ARCHITECTURE FOR CANCER STAGING.

In Proceedings of the 11th International Conference on Enterprise Information Systems - Artiﬁcial Intelligence and Decision Support Systems, pages

313-316

DOI: 10.5220/0001863503130316

 SciTePress

in cancer staging, the intricacy of the staging itself,

and the fact that it is usually performed by none ex-

perts in the area cancer, staging was the chosen ﬁeld

to start implementing such an architecture. The pro-

posed architecture is in production at an healthcare

unit and stands as a stepping-stone inﬂuence process

towards a platform which will allow distinct and tech-

nology independent decision support systems to be

fully adapted to the healthcare environment through

the use of EHR technologies.

It was under this assumption that an Agency for

the Integration, Archive and Diffusion of Medical In-

formation (AIDA) was developed, using new com-

putational paradigms and methodologies for problem

solving, which have been based based on the concept

of agent (Machado et al., 2008).

2 THE AJCC STAGING NORM

In information sharing, an understanding of the con-

cepts and vocabulary involved is paramount. Anal-

ogously, in a healthcare environment, a ﬁrm knowl-

edge on specialized areas and their uniform lexicon

is required, from physicians and other staff, in order

to normalize information communication and the pro-

cess of their dissemination among healthcare units,

in performing a group decision meeting, or even on

a peers exchange on routine practices, where the ex-

istence of different thesaurus associated to particular

backgrounds may result in dubious or mistaken as-

sumptions and culminate in misjudgments with com-

plex complications. For this reasons, several the-

saurus have been created, trying to establish an uni-

form syntactic and semantic indexation of medical in-

formation, such as SNOMED CT or even more spe-

ciﬁc normalization, like the American Joint Commis-

sion on cancer staging norms (Andrews et al., 2008).

This norm consists of a structure of uniform staging

variables within every area where there may be can-

cer, keeping however many specialized variables on

speciﬁc cancer areas (Greene et al., 1998). The speci-

ﬁcity of the staging of each particular area increases

the difﬁculty of following this normalization without

any kind of help or, at least, any mean of consulta-

tion. Furthermore, case speciﬁc information or more

detailed information not presented on the form itself,

but presented in other sources, may be of great impor-

tance for a better understanding and compliancy of the

norm deﬁned by the AJCC (Greene et al., 2003).

3 AN IMPLEMENTATION OF A

HEALTHCARE GROUP

DECISION SUPPORT SYSTEM

The system was developed with prospects of future

implementation in a particular healthcare environ-

ment. This fact greatly affected the design and soft-

ware used, resulting in a custom tool, for this spe-

ciﬁc healthcare environment. Moreover, its adaptabil-

ity towards the necessities of this particular institu-

tion and its already well designed information system,

greatly enriched the tool, giving it a broader applica-

bility when integrated with services, while at the same

time enabling a seamless area adapted service. In the

design of this system, a multi-layer architecture was

used in order to increase its reliability and scalability,

as well as to improve the results in providing the ﬁ-

nal service as a whole, i.e., enabling a fully working

group decision platform.

In a more speciﬁc description, the GDSS follows

a client-server hierarchy and can be decomposed into

four individual layers (Figure 1). The database layer,

the cornerstone of this implementation, is the one re-

sponsible for the messaging system, which allow the

communication between human agents and software

based ones. As well as this role, it is responsible

for containing structured information for the agents

to extract and analyze. At the present time, it stores

information concerning the AJCC cancer staging sys-

tem, being this informations structured in areas and

staging variables. The application server and the user

layer are responsible for the web interface available

to human agents. The users, through web browsers,

and mediated by software agents, converse and ne-

gotiate with other agents (human or software ones)

that are able to help the user in a particular area of in-

terest or related with the problem. This communica-

tion is achieved through conveyed messages between

users and agents, stored and managed by relational

databases.

Holding the software agents, the agent layer in-

cludes several types of services that go beyond system

management, being able to interact with both human

and software agents, as well as integrating the GDSS

with the already established information system of the

healthcare institution. The human agents can com-

municate among themselves or even with other agent

systems as long as the other layers are still functional.

Within the agent system platform, each agent respon-

sible for interaction with human agents has an in-

dividual role and can interact with the user through

the messaging system, answering questions and es-

tablishing a dialogue aimed towards the problem so-

lution.

ICEIS 2009 - International Conference on Enterprise Information Systems

314

Figure 1: The GDSS Architecture.

3.1 The User Interface

The user interface is a web application that is both

based on Asynchronous Javascript and XML (AJAX)

and Hypertext Preprocessor (PHP). With the use of

AJAX, the human agent is faced with a much more

interactive environment similar to usual desktop ap-

plications, where events and other variables can be

updated on request or need without full page refresh.

The use of area speciﬁc or case speciﬁc chat

rooms is extremely important in this implementation,

as it has a vital role in enabling a specialized group de-

cision platform. In these rooms, physicians and other

healthcare personal can discuss matters, being medi-

ated or not, depending of its speciﬁc objective. The

GDSS has a great potential when integrated within a

healthcare facility, not only just with the possibility of

intercommunication within this institution on simple

work related subjects, just like any intern messaging

system, but as well as a mean of storing and easy shar-

ing of information provided by both human and soft-

ware agents. In this speciﬁc case users have a fully

functional and user friendly platform with which they

can actively interact with the information system it-

self through software agents.

The GDSS system is as well integrated with the

web application used by the same institution to ac-

cess the Electronic Health Record (EHR). In this way

agents, both human and software can generate events

when intending to interact with an user when he/she is

in any of the interface systems. As a result the system

can widely interact with users and be fully integrated

with the institutions information infrastructure. Other

functions played by other agent systems can be added

into this architecture fulﬁlling the continuous need to

scalability within this system.

Figure 2: GDSS Integration Platform.

3.2 Agent System

The problem addressed in this agent based system

is concerned with the AJCC’s cancer staging forms,

more concretely with the intricacy it presents to non

oncologist when staging a cancer. The need for a spe-

cialized agent may guide a physician in solving a par-

ticular staging problem, supporting the decision mak-

ing process and presenting the ﬁnal staging score in a

normalized form, and then integrated it with the EHR

of the patient, make the difference and set our goals.

Indeed, the agents share decision support procedures

and resulting intents, presenting themselves as the so-

lution for the complex proceedings that may occur at

the institutional level.

In the GDSS, agent system interaction with hu-

man agents is based mainly on messages sent by

means of the database platform; there exists, how-

ever, communication agents listening at speciﬁc ports

using socket connections that can receive informa-

tion directly from an auxiliary interface. From a

brief conversation with the user, a best ﬁtted agent is

called in order to successfully help in solving the pre-

sented problem. The data respective to cancer stag-

ing was extracted from the forms, and the supplemen-

tary guidance presented by other AJCC releases ex-

plaining the norm(Greene et al., 1998) (Greene et al.,

2003). There exists 45 forms out of the 48 chapters

of the AJJC norm, being the data in each form repre-

sented in different relations of the relational database,

distinguishing each staging variable and the relevant

information involved in the staging procedure.

3.3 The Technology

Among the possible tools to the analysis and devel-

opment of our system, the JADE framework (Java

Agent DEvelopment Framework) stands as a promis-

ingly one for GDSS. Once it was developed on top

AN AGENT-BASED ARCHITECTURE FOR CANCER STAGING

315

of the cross-platform technology Java, allows for a

strong connection among two similar computational

paradigms, such as The Agent and The Object Ori-

ented Programming ones, as well as the possibility to

easily integrate the rich Java libraries into the agent’s

behaviours. However, developed under an object ori-

ented paradigm, the difference between objects and

agents is fully secured in JADE, so that agents can

have the autonomy to choose their own actions and re-

act according to their inherent behaviour(Bellifemine

et al., 2007).

JADE is according to FIPA ( Foundation for In-

telligent Physical Agents) speciﬁcations, allowing for

a standard communication procedure, which can be

used to interact with other FIPA compliant agent

systems. With JADE, different performatives asso-

ciated with the intent on the interaction are easily

differentiated, allowing for a better comprehension

of the communication contents and objective by the

agent(Bellifemine and Rimassa, 2001). In the de-

velopment of a MAS, in order to be able to com-

municate in a rich manner, presents a potential nor-

malized infrastructure of openness, autonomy, ro-

bustness, scaleability and ﬂexibility(Charlton et al.,

2000). The coordination among agents was success-

fully implemented following this norm, though when

integrating with the institutions proprietary agent sys-

tems, the HL7 norm was used in order to ensure that

both platforms were compliant with the same stan-

dard(Machado et al., 2008).

4 CONCLUSIONS

Modern science utilizes some basic approaches to the

study of how nature works, namely Observational

Science, Experimental Science, Theoretical Science,

and Computational Science. However, Computa-

tional Science is the newest, made possible by the

tremendous improvements in both computer hard-

ware and software over the past thirty years. Com-

putational Science, sometimes known as Modeling

and Simulation or Scientiﬁc Computing, is used in

Medicine, and beneﬁts from knowledge in a large

amount or scale. On the other hand, staging, a key

word in our work, refers to a clinical process that

is based on knowledge and on the way cancer de-

velops, with some staging systems covering different

types of cancer, when others focus on a particular type

of cancer (e.g., distinctive staging systems are used

for countless cancers of the blood or bone marrow

such as lymphoma). Indeed, doctors gather dissim-

ilar types of information about cancer to determine

its stage. The various tests used for staging depend

on the type of cancer, and may include physical ex-

ams, imaging tests (e.g., XR, CT or MRI), laboratory

tests (e.g., blood, urine, uids or tissues), or pathol-

ogy and surgical reports, i.e., in terms of the algo-

rithm, in Computational Science, the scientiﬁc prob-

lem must be expressed mathematically, known as the

Algorithm. In terms of the architecture, once a suit-

able algorithm has been determined, that algorithm is

translated into one or more computer programs and

implemented on one or more types of hardware. In

our work, the combination of software and hardware

is referred to as the Computational Architecture, the

AIDA agency referred to above.

REFERENCES

Andrews, J. E., Patrick, T. B., Richesson, R. L., Brown, H.,

and Krischer, J. P. (2008). Comparing heterogeneous

snomed ct coding of clinical research concepts by ex-

amining normalized expressions. J. of Biomedical In-

formatics, 41(6):1062–1069.

Bellifemine, F. and Rimassa, G. (2001). Developing multi-

agent systems with a ﬁpa-compliant agent framework.

Softw. Pract. Exper., 31(2):103–128.

Bellifemine, F. L., Caire, G., and Greenwood, D. (2007).

Developing Multi-Agent Systems with JADE (Wiley

Series in Agent Technology). John Wiley & Sons.

Charlton, P., Mamdani, E., Cattoni, R., and Potrich, A.

(2000). Evaluating the ﬁpa standards and its role

in achieving cooperation in multi-agent systems. In

HICSS ’00: Proceedings of the 33rd Hawaii Interna-

tional Conference on System Sciences-Volume 8, page

8034, Washington, DC, USA. IEEE Computer Soci-

ety.

Greene, F., Page, D., Fleming, I., Fritz, A., Balch, C.,

Haller, D., and Morrow, M. E. (1998). AJCC Can-

cer Staging Forms for Microsoft Word). Lippincott

Williams & Wilkins.

Greene, F. L., Page, D. L., (Editor), I. D. F., Fritz, A., and

Balch, C. M. (2003). AJCC Cancer Staging Handbook

- From the AJCC Cancer Staging Manual. Springer-

Verlag, sixth edition edition.

Machado, J., Abelha, A., Novais, P., ao Neves, J., and

Neves, J. (2008). Improving patient assistance and

medical practices through intelligent agents. In Pro-

ceedings of the Workshop in Agents Applied in Health-

care, AAMAS 2008.

ICEIS 2009 - International Conference on Enterprise Information Systems

316