MEDFINDER

Using Semantic Web, Web 2.0 and Geolocation Methods to Develop a Decision

Support System to Locate Doctors

Alejandro Rodríguez, Jesús Fernandez, Enrique Jimenez, Myriam Mencke

Mateusz Radzimski, Juan Miguel Gómez

Department of Computer Science, Universidad Carlos III de Madrid

Avenida de la Universidad 30, 28911, Leganés, Madrid, Spain

Giner Alor-Hernandez, Rubén Posada-Gómez

Division of Research and Postgraduate Studies, Instituto Tecnológico de Orizaba, Mexico

Keywords: Differential Diagnosis, e-Science, Ontologies, Reasoning, Web 2.0, Semantic Web, Geolocation.

Abstract: Currently the introduction of new technologies for efficient medical diagnosis and treatment is having a

profound social and organizational impact, initiating the need to exploit the latent potential of novel

methods. A requirement has emerged to maximally exploit the fusion of new technologies for more efficient

patient care than is presently available. This scenario motivated the objective of the current paper,

combining an existing medical diagnosis system which uses semantic technology and probabilistic

techniques with Web 2.0 and geolocation methods to develop a system to locate the most appropriate doctor

for a patient. Results of an initial prototype implementation are promising.

1 INTRODUCTION

In today’s IT landscape, a multitude of systems exist

which permit a user to perform medical queries

based on a series of factors, such as symptoms,

laboratory tests, and medicines, among others.

Computations made based on these factors are

generally capable of providing a relatively precise

diagnosis using a formal reasoning process, and in

some cases, the diagnostic may be accompanied by

an associated probability of the diagnosis outcome.

However, until now, the possibilities which such a

system could offer have not been fully exploited. A

series of technologies could be added to enhance

these types of systems in order to provide much

more complex and complete functions. The

objective of this paper is to describe a framework for

the implementation of such a system, by combining

an existing system which performs medical

diagnosis using the methods just described with new

technologies. This forms the basis for the

construction of an architecture which relates both

technologies.

The initial application used is an expert system

which enables doctors to perform medical diagnosis

based on determined parameters, which uses

Semantic Web technologies for this purpose

(Rodriguez, 2008).

An additional component will be introduced to

the system, where once the diagnosis has been

determined and the diagnosed illness has been

classified as the most probable, and the illness has

been verified as correct by a medical expert, a

database of medical experts will be consulted. The

objective of this step is to locate a series of experts

which fulfil certain criteria in relation to the illness

diagnosed. This database will interact as a

positioning system, which among various other

functions, is able to obtain the shortest distance

between the expert and the patient, thereby

determining the optimal route for the patient to reach

the location of the expert. The system is used to

realize this process in the fastest and least disruptive

form available.

288

RodrÃ guez A., Fernandez J., Jimenez E., Mencke M., Radzimski M., GÃ¸smez J., Alor-Hernandez G. and Posada-GÃ¸smez R.

MEDFINDER - Using Semantic Web, Web 2.0 and Geolocation Methods to Develop a Decision Support System to Locate Doctors.

DOI: 10.5220/0001831402880293

In Proceedings of the Fifth International Conference on Web Information Systems and Technologies (WEBIST 2009), page

ISBN: 978-989-8111-81-4

In the motivating scenario section we will shortly

explain the main reason for this research. In the

MEDFINDER section, we will deeply talk about the

system, explaining its architecture: the ODDx

system, that is the system diagnosis provider; the

Sorter subsystem, which will categorize the diseases

into speciality groups; the Expert database, that

contains all the information about medical experts

that can deal with the speciality specified by the

Sorter; the Web 2.0 Feedback system, that allows to

change the expert punctuation; and the Geolocation

system, which will locate the nearest doctor and

design the best route to meet him. In the related

work section we will introduce some systems that

use similar techniques and, in the final chapters, we

will talk about the findings and conclusions we came

to, and the future work needed. Finally, in the

acknowledgements section we will mention the main

projects in which we are currently involved, and that

allow us to make a better job.

2 MOTIVATING SCENARIO

The choice of a medical professional by an expert

system to cure or treat a specific disease within a

particular field is not a trivial subject. Generally, a

person in urgent need of medical treatment arrives at

emergency services where a qualified medical

professional is assigned to attend to the case,

however, on certain occasions, the patient is able to

select the doctor due to circumstances such as

having access to private healthcare. One of the

objectives of the current system is to construct a link

between the patient and doctor in such a form that

the patient can access the professional aptitudes of

diverse doctors as a function of his/her necessities,

taking into account particular variables, such as the

distance which separates both.

A different scenario with rather interesting

application is the case of the contracting of

specialists by clinics, private or public. In this case,

the system would be a modification of the system

previously described, where the component which

makes the diagnostic would essentially be dispensed

with, given that in large part it is no longer

necessary. In this scenario, the person responsible

for contracting specialists already knows the branch

of medicine in which each specialist is qualified,

therefore the diagnostic component is no longer

fundamental. Once the specialization is known, the

rest of the process is carried out similarly to that

detailed before, given that the system is responsible

for searching for the specialist or expert which best

fits the data provided.

Additionally, this technology could be

incorporated into many further scenarios unrelated

to medicine, but where the entire system could be

implemented with some modifications. For example,

implementing the system for the detection of

mechanical failures in a garage, where the software

process is responsible for detecting the fault or

breakdown, and the system locates the most

appropriate expert for repairing the said defect.

Additionally, in the case of replacement of parts, the

closest part sellers with the best quality and price

could be located by means of the system.

3 MEDFINDER

Given the proposed scenarios and the problems to

resolve, the subsequent step is to describe the

approximation for the solution of the problems. This

solution is based on previous work described and

developed in another existing solution, in particular,

(Rodriguez, 2007), which is an existing medical

differential diagnosis system based on an ontology,

logical inference and probabilistic techniques.

Additionally, in what follows, the remainder of the

subsystems and parts involved in the elaboration of

the framework will be explained.

3.1 System Architecture

The current structure of MedFinder is based on the

construction of various components which

interchange certain information among each other.

The conjunction of these systems permits that the

entire set of components functions correctly, and the

necessary data can be obtained to achieve the

desired outcome.

A diagram is shown below which displays the

general structure of the system and the relationships

between its elements. In the subsequent section, the

internal functioning of each of the elements will be

described, as well as their behavior with regard to

the information interchanged with the rest of the

components in the system. As additional detail for

the understanding of the diagram, consider the

following elements:

[I]: Represents the information which one

subsystem sends to another.

[A]: Represents the action which one component

performs on another component. This may include

the transfer of information or not.

MEDFINDER - Using Semantic Web, Web 2.0 and Geolocation Methods to Develop a Decision Support System to Locate

Doctors

289

Figure 1: System Architecture.

3.1.1 ODDX

The first component of the system with which the

user interacts is ODDX, as can be observed in the

diagram. Viewed globally, it is in fact the initial and

only part which involves user interaction. It is also

this component which is the front end of the expert

system that realizes the diagnostic for the user. The

fundamental difference of this system compared

with the previous system is that the architecture is

extended considerably when compared with the

existing architecture.

ODDX is an expert system which was capable of

inferring medical diagnoses based on a given

number of parameters, without any further functions.

However, in the new architecture, this component is

no more than the beginning of an entire process

which not only diagnoses the illness, but also

provides the user with much more information

related to the diagnosis.

Therefore, the architecture of ODDX has not

changed, however, a number of variations have been

made.

The basic components of ODDX are the

following:

• Probabilistic. The probabilistic system is the

system that is responsible for managing

and/or calculating the probabilities of every

diagnostic inferred. Every disease that is

diagnosed (one or more) from a group of

indications has its own probability of

happening.

• Data Loading. This is the engine which

performs data loading from the ontology. It

employs the Jena API to read the ontology

file, which is an open source Java

programming environment for Semantic Web

applications, and supports the use of various

languages, such as RDF, RDFS, OWL and

SPARQL.

• Combination System. The combination

system computes all of the diagnostic

combinations possible which may be the

result of the interaction of drugs. Basically it

is a method which allows, given a patient

with a group of indications and associated

drugs, the calculation of the possible

interactions which may be caused by drugs.

• Inference. The inference engine is the main

engine of application, because it is a principal

constituent of the system which really

enables the diagnoses to be made. This

engine requires access to the knowledge base

with the diseases, symptoms, etc., and at the

same time needs access to the knowledge

base that contains inference rules.

• Search. This component takes the form of a

search engine, which realizes SPARQL

queries to the ontology to consult all of the

data stored it. This permits fast access to all

of the data stored within the ontology.

3.1.2 Sorter

The second component involved in the entire

process has been entitled Sorter. This is essentially a

system which is capable of generalizing or

specifying. It receives an illness as input, which is

assigned an ICD code (International Classification

of Diseases by the World Health Organization), and

by means of this code, the system is capable of

classifying this illness within one of the multiple

specialisation of the ICD’s own classifications.

The implementation of this part of the system

may be viewed as both a client and server at the

same time. On the one side it acts as a server to

receive information (the ICD code of the illness),

which is provided by the client, and at the same time

behaves as a client at the moment it sends its output

(the specialization) to the subsequent subsystem.

Whether the use of this subsystem is dependent

upon the final implementation is optional. This is

because ODDX in its current state does not

implement this classification component, which

would be easily implementable and in fact a more

efficient solution, given that this service could be

discarded, thus speeding up the final solution.

Regarding the technologies used to implement

this facet, the principal constituent proposed is an

ontology, which allows the classification of the

illness and its categorization within a speciality. For

the present framework, it is proposed that the system

WEBIST 2009 - 5th International Conference on Web Information Systems and Technologies

290

uses the same ontology which the ODDX system

uses, given that this ontology classifies the illnesses

in distinct superclasses. In a specific level of these

superclasses, they represent precisely the

specialization. Due to the design of the ontology, the

classification of the illness is thus a rather simple

process.

3.1.3 Expert Database

The objective of the database of experts is precisely

to provide a database which contains the most

relevant data relating to an expert. The database

contains a series of data, which are listed below:

• Latitude Coordinate. Coordinate to locate

the expert, principal location.

• Longitude Coordinate: Coordinate to locate

the expert, principal location

• Expert Specialization Code. Code(s)

representing the distinct specializations of the

expert.

• Possible Alternative Coordinates. Re-

ference to possible alternative codes to locate

the expert. These coordinates could also be

linked to a particular specialization.

• Patient Scores. Scores which the patients can

assign to the expert once having received the

treatment. Further details will be provided

below.

• Expert Scores. Scores of other experts in the

same matter.

• Other Data. Other series of data such as

name, usual address, telephone, email,

treatment times, and observations, among

others.

A diagram of the Entity-Relations in the database

is provided below, followed by a description of the

relations.

Figure 2: Entity-Relation Schematic Representation.

The ER Diagram description is provided above.

Entities:

Doctor. The main entity containing information

about medicine doctors, such as name, address,

contact and other data.

Patient Points. Patient Point entity represent each

mark given by patient to score a doctor.

Expert Points. Export Point entity, similar to

Patient Points, is a representation of scores given by

doctors/experts to other doctors.

Location. Exact location of the doctor’s/expert’s

office. Contains the exact geolocation data

(longitude and latitude), address and hours of

consultation.

Specialization. Represents all available experts’

specializations, which are described by unified

specialization code.

Doctor_Specialization. Additional entity which

represent each doctor’s/expert’s specialization. As

one expert may perform more than one

specialization, each in different location. This entity

uses foreign keys from Doctor and Specialization

entities.

Relations:

Specializes_in (related entities: Doctor, Doctor_

Specialization) – an one-to-many relation describing

doctor’s specializations. Since some doctors/experts

may perform more than one specialization, each one

is represented by a distinct entry in

Doctor_Specialization entity.

Performs_specialization_at (related entities:

Location, Doctor_Specialization) – an one-to-many

relation defining the locations (office, clinic) where

certain doctor’s specialization is performed.

Performs (related entities: Specialization,

Doctor_Specialization) – an one-to-many relation

describing the specializations that are performed by

doctors/experts. It allows to map exact

specializations to actual doctors’ specializations.

Is_graded_by (related entities: Patient_Points,

Doctor) – an one-to-many relation between doctor’s

scores and certain doctor. Each doctor is given many

scores from patients. Relation between

Expert_Points and Doctor is analogical.

Additionally, the system described above

receives input from another subsystem, namely Web

2.0. Further details of this system will be provided

below.

MEDFINDER - Using Semantic Web, Web 2.0 and Geolocation Methods to Develop a Decision Support System to Locate

Doctors

291

3.1.4 Web 2.0 Feedback System

The aim of the Web 2.0 system is to insert or update

data relating to experts within the system. This

system also permits the modification of the scores

assigned to distinct experts, thereby facilitating the

users of the system in the decision to choose one

expert or another. This scoring may be divided into

two distinct parts:

Scores of Previous Patients. These scores are based

on the scores which patients that have been

previously treated by a particular expert may assign

to this expert.

Scoring of other Experts. This system is based on

permitting other experts to assign scores to their

colleagues, thereby improving the global vision of

these by structuring it such that their knowledge is

considered important and other professionals in the

sector value their work.

3.1.5 Geolocation System

The geolocation system is the final constituent of the

system. It is fundamentally a system which

communicates with the database of experts to obtain

certain data required to locate the expert. The

essential data are the localization, longitude and

latitude coordinates, however, other data may be

required. The system should calculate a real route

(realizable by car or on foot, not the distance

between two points), which exists between the

patient and expert(s).

There are already a number of tools in existence

which achieve this aim, however, in the current

framework, particular importance was given to the

aspects high potency, ease of management and

reliability. These features were selected having in

mind that the system not only had to establish the

physical location of the specialist, but also had to be

capable of establishing a route between the patient’s

current position and the expert.

The API which permits these characteristics is

proprietary of Google, Google Maps. This

framework provides a library which allows the

creation of maps of a determined location,

establishing distinct tags, controls, and determining

routes between two points. Another feature in favour

of selecting this platform is that it allows the

drawing up of a physical, political and hybrid view

between both places (Bühler, 2006). This implies

that the user has more visual references for the place

at which he wants to arrive, and additionally, for

intermediate points which he must pass to reach his

destination (Muller, 2004).

However, the key strength of the framework is

that it allows the calculation of the shortest route

between two points with the simple process of

introducing the coordinates of the user and the

expert, which in this case can be obtained from

ExpertDB. Additionally, a detailed description of the

route is provided indicating specifically the route

which the user should take, which turns he should

take, and the public transport available in each of the

streets on the route (Grabler, 2008). It also offers

further information about the total length of the

journey, and the approximate time it takes to

complete the entire journey.

Another very interesting aspect is that this API

offers the possibility to select how the journey

should be realized, on foot or by car, displaying

distinct alternatives according to the option chosen.

For example, if the user wants to undertake the

journey on foot, the system does not process or take

into account restrictions or prohibited ways which

would be encountered by a car.

4 RELATED WORK

In the domain of medical diagnosis systems, a

myriad of approaches exist, which are comprised of

various algorithmic techniques for automatic

diagnosis that have been tested in research, as well

as actual systems currently available for use.

Approaches in research which apply the use of

combined techniques such as the current one include

neuro-fuzzy methods (Noy, 2005), the application of

genetic algorithms (GAs) for rule selection

(Ishibuchi, 1999), or the unification of genetic

algorithms with fuzzy clustering techniques (Roubos

and Setnes, 2000).

Apart from the systems described above, more

specialized systems are available, for example, those

in which clustering techniques are used in the

detection of epidemics (Cardoso, 1999), decision

and action support systems in relation to illnesses

according to region (Gosselin and Lebel, 2005), and

systems which aid the differential diagnostic in

“erythemato-squamous” form, incorporating

classification algorithms for trees and other similar

methods (Güvenir, and Emeksizb, 2000). Possibly

the most interesting and most similar application in

relation to the current system is that developed by

(Faria, Fernandes and Perdigoto, 2008), which

propose a specific type of monitoring for old people

or people with mental illnesses such as Alzheimers,

which using a mobile system ensures that the person

can be geographically located in any place in the

WEBIST 2009 - 5th International Conference on Web Information Systems and Technologies

292

case that he needs help, or physically reached by

someone else by means of a button which is

designed to deal with such cases.

Besides these types of geolocation systems,

currently there is a scare number of systems in

medicine which provide this type of functionality.

These types of systems are in fact more advanced in

other domains, such as tourism. There are

localization systems for destinations or touristic

monuments to create personalized information for

travelers (Kalczyński, 2001), and of course the well-

known tools such as Google, Google Earth and

Google Maps, as previously detailed. This makes the

incorporation of such tools in medical applications a

challenging future field.

5 CONCLUSIONS AND FUTURE

RESEARCH

The current paper has outlined a system for

recommendation of medical diagnostics which uses

an ontology, logical inference, and in particular,

introduces the feature of detection of medication that

may interact with other medicines and other signs

like allergies. An additional step which has been

taken is to provide a framework for geolocation of

doctors, an architecture which is currently under

development.

A further research step which is envisaged is to

also allow users to introduce other complementary

parameters into the system to construct a more

specific system that allows the user to obtain more

precise results.

ACKNOWLEDGEMENTS

This work is supported by the Spanish Ministry of

Industry, Tourism, and Commerce under the project

SONAR (TSI-340000-2007-212), GODO2 (TSI-

020100-2008-564) and SONAR2 (TSI-020100-

2008-665), under the PIBES project of the Spanish

Committee of Education & Science (TEC2006-

12365-C02-01) and the MID-CBR project of the

Spanish Committee of Education & Science

(TIN2006-15140-C03-02).

REFERENCES

Bühler E.R., Localización Geoespacial con Google Maps.

Solo Programadores 143, pp 36-41, 2006

Faria S., Fernandes T.R. and F.S. Perdigoto., Mobile web

server for elderly people monitoring. ISCE 2008, IEEE

International Symposium on Consumer Electronics,

2008

Gladys M., Casas C., Grau R. and A. Milagros., Técnicas

de clustering para la detección de epidemias, 1999

Gosselin P., Lebel G., Rivest S. and D. Monique, The

Integrated System for Public Health Monitoring of

West Nile Virus (ISPHM-WNV): a real-time GIS for

surveillance and decision-making. International

Journal of Health Geographics., 2005

Grabler F., Agrawala M., Sumner Robert W., and M.

Pauly., Automatic generation of tourist maps. ACM

Transaction on Graphics, 27 (3), Article 100, 2008

Güvenir H.A. and N. Emeksizb.,An expert system for the

differential diagnosis of erythemato-squamous

diseases. Expert Systems with Applications, 18 (1), pp.

43-49., 2008

Ishibuchi H., Nakashima T., and T.Murata., Performance

Evaluation of Fuzzy Classifier Systems for

Multidimensional Pattern Classification Problems.

IEEE Trans. SMC–B 29, pp. 601–618., 1999

Kalczyński P., Paprzycki M., Fiedorowiczl I.,

Abramowicz W. and C. Maria., Personalized traveller

information system, Proceedings of the 5th

International Conference Human-Computer

Interaction, 2001

Muller Ross M., Kyusuk C., Croke K. G. and K. Mensah

Edward., Geographic Information Systems in Public

Health and Medicine. Journal of Medical Systems 28

(3), pp. 215-221., 2004

Rodriguez A. 2007. ODDX: Ontology Based Differential

Diagnosis. Final Year Project, Computer Science

Department, University of Oviedo, Oviedo, Spain.

Rodriguez A., Mencke M., Alor Hernandez G., Posada

Gomez R. and J. M. Gomez., Medboli: Medical

Diagnosis Based on Ontologies and Logical Inference.

The Third International Conference on Digital

Society, ICDS 2009, February 1 - 7, 2009, Cancun,

Mexico (accepted, to appear)., 2008

Roubos J.A., M. Setnes., Compact Fuzzy Models through

Complexity Reduction and Evolutionary Optimization.

In FUZZ-IEEE, pp. 762-767., 2000

MEDFINDER - Using Semantic Web, Web 2.0 and Geolocation Methods to Develop a Decision Support System to Locate

Doctors

293