DECISION SUPPORT AND IMAGE & SIGNAL ANALYSIS IN

HEART FAILURE

A Comprehensive Use Case

Sara Colantonio, Massimo Martinelli, Davide Moroni, Ovidio Salvetti

Inst. of Information Science and Technologies,ISTI-CNR, Pisa, Italy

Francesco Perticone, Angela Sciacqua

Univerity Magna Graecia, Catanzaro, Italy

Franco Chiarugi

Institute of Computer Science (ICS), Foundation for Research and Technology – Hellas (FORTH)

Heraklion, Crete, Greece

Domenico Conforti, Antonio Gualtieri, Vincenzo Lagani

University of Calabria, Cosenza, Italy

Keywords: Decision Support Theory, Echocardiographic Imaging Analysis, Signal Processing.

Abstract: The European STREP project HEARTFAID aims at defining an innovative platform of services able to

intelligently support clinical operators in the daily management of heart failure patients. The core of the

platform intelligence is a Clinical Decision Support System, developed by integrating innovative knowledge

representation techniques and hybrid reasoning methods, and including advanced tools for the analysis of

diagnostic data, i.e. signals and images. Aiming at showing how all these issues are combined in the

HEARTFAID platform, we present a comprehensive use case, centred on echocardiography workflow and

covering the clinical course leading from visit scheduling to therapeutic choices, highlighting the

intervention and the value added by the Clinical Decision Support System.

1 INTRODUCTION

Heart Failure (HF) is a complex clinical syndrome

resulting from any structural or functional cardiac

disorder which impairs the ability of the ventricle to

fill with or eject blood. In its chronic form, HF is

one of the most remarkable health problems for

prevalence and morbidity, especially in the

developed western countries, with a strong impact in

terms of social and economic effects. All these

aspects are typically emphasized within the elderly

population, with very frequent hospital admissions

and a significant increase of medical costs. Recent

studies and experiences have demonstrated that

accurate heart failure management programs, based

on a suitable integration of inpatient and outpatient

clinical procedures, might prevent and reduce

hospital admissions, improving clinical status and

reducing costs.

The European project HEARTFAID (“A

knowledge based platform of services for supporting

medical-clinical management of the heart failure

within the elderly population'” — IST-2005-027107)

aims at defining efficient and effective health care

delivery organization and management models for

the “optimal” management of the care in the field of

cardiovascular diseases.

The HEARTFAID platform (HFP) has been

conceived as an integrated and interoperable system,

able to guarantee an umbrella of services that range

from the acquisition and management of raw data to

the provision of effective decisional support to

288

Colantonio S., Martinelli M., Moroni D., Salvetti O., Perticone F., Sciacqua A., Chiarugi F., Conforti D., Gualtieri A. and Lagani V. (2008).

DECISION SUPPORT AND IMAGE & SIGNAL ANALYSIS IN HEART FAILURE - A Comprehensive Use Case.

In Proceedings of the First International Conference on Health Informatics, pages 288-295

 SciTePress

clinicians. All the functionalities and services

supplied by the entire HFP can be further grouped

into data, decision and end-users macro “contexts”.

The former is devoted to the collection and

management of information, which consists of

biomedical data, acquired from biomedical devices

and structured/unstructured information such as

clinical reports collected during patient

hospitalisation and outpatient visits and within a

homecare program by telemonitoring patients’

conditions. The decision context includes a

knowledge-based Clinical Decision Support System

(CDSS) whose main goal is supporting the HF

health care operators, by making more effective and

efficient all the processes related to diagnosis,

prognosis, therapy and health care personalization of

the HF patients. The latter context provides the

doorway to a multitude of end-user utilities and

services, such as accessing an electronic health

record, querying the CDSS, applying advanced

models and methods for the extraction of new

knowledge, and so forth.

The CDSS represents the core of HFP and has

been carefully designed by combining innovative

knowledge representation formalisms,

robust and

reliable reasoning approaches, based on Machine

Learning and inference methodologies, and

innovative methods for diagnostic images analysis

and biomedical signals processing (VV.AA., 2007).

This paper aims at showing how all these issues

are combined within a comprehensive use case,

centred on an echocardiography workflow and

covering the clinical course leading from visit

scheduling to therapeutic choices, highlighting the

intervention and the value added by the CDSS.

In the following sections, we will briefly review

the related literature, then introduce the

HEARTFAID CDSS and its main functionalities,

and, finally, describe the use case and CDSS

interventions.

2 DECISION SUPPORT AND

DATA PROCESSING IN

HEARTFAID

The development of computerized applications for

supporting health care givers is an old but still alive

quest, started more than 45 years ago, in the early

1960s, and with ascending and descending periods

of interest and growth (Greenes, 2007).

A plethora of CDSS has already evolved with

different platforms and architectures, encompassing

a variety of services, from information retrieval and

reporting, scheduling and communications, to cost-

effectiveness, error prevention, safety, and

improvement of health care quality. The most

common realizations include electronic

medical/patient records (Poissant et al., 2005),

computerized alerts and reminders, clinical

guidelines formalizations (GEM, 2003), provider

order entries (Park et al., 2005), diagnostic support,

clinical result interpretation, adverse event

monitoring, shared patient-doctor decision-making

(Wirtz et al., 2006).

The primary task for developing effective CDSS

is to select the corpus of pertinent knowledge and/or

collect and process data to create pertinent

knowledge which is relevant for bringing the health

care to effect. Knowledge representation just

concerns understanding, designing, and

implementing ways of formally coding the

knowledge necessary for deriving other knowledge,

planning future activities, solving problems that

normally require human expertise. Representing

knowledge requires the selection of a suitable

language or formalism and the definition of a

Knowledge Base (KB) built by formalizing clinical

experts’ know-how and guidelines. Usually, the

formalism is symbolic and the KB contains

statements or expressions of one of the following

formalisms: (i) rule based; (ii) frame based; (iii)

network based; and (iv) logic based (Helbig, 2006).

Workflow based representation is also becoming

well-known, especially for modelling guidelines

(Boxwala et al., 2004). Moreover, in recent years

ontologies are emerging as a powerful knowledge

representation formalism which is conceptually

equivalent to the frame based and to first order logic

approach (Bayegan et al., 2002).

The KB is exploited by a reasoning engine

which processes available information for

formulating new conclusions and answering

questions. Inferential reasoning is employed for

inferring new knowledge from the KB by deduction,

induction or abduction.

In some cases, making a decision requires an

investigation on the hidden, complex, often non-

linear correlations among data, together with high-

level analytical processing functions. In such cases,

the knowledge needed for the solution should be

acquired directly from data (inductive knowledge)

and stored in a model (e.g. Artificial Neural

Networks, Support Vectors Machines), able to

induce sub-symbolic knowledge by data-driven

processing. Computational models are also useful

for representing uncertain knowledge, as Bayesian

Networks and Fuzzy theory.

HF routine practice presents several aspects in

which an automatic, computer-based support could

have a favourable impact. Some attempts to support

HF clinical operators have been presented, such as

DECISION SUPPORT AND IMAGE & SIGNAL ANALYSIS IN HEART FAILURE - A Comprehensive Use Case

289

an Electronic Patient Record (Bosmans et al., 2000)

or computerized guidelines (Dassen et al., 2003).

More complex decision support systems have been

developed for suggesting the most appropriate

therapy (e.g. Perlini et al., 1990).

Within HEARTFAID, a careful investigation

about the needs of HF practitioners and the effective

benefits assured by decision support was performed:

four problems have been identified as highly

beneficial of HEARTFAID CDSS point-of-care

intervention. They can be referred as macro domain

problems and listed up as: (i) HF diagnosis, (ii)

prognosis, (iii) therapy planning, and (iv) follow-up.

Further detailed decision problems were identified

for specifying these macro domains, focusing as

much as possible on the medical users’ needs;

explicative examples are:

 severity evaluation of heart failure;

 identification of suitable pathways

 planning of adequate, patient’s specific therapy;

 analysis of diagnostic exams

 early detection of patient’s decompensation.

An accurate analysis highlighted that the needed

corpus of knowledge mainly consisted of domain

know-how. Nevertheless, the solution of some of

these problems seemed still debated in the medical

community, due to the lack of validated and assessed

evidences. In such cases, computational models

appeared the best solution for modelling the decision

making extracting knowledge directly from available

data. Moreover, specific processing algorithms were

designed for analyzing diagnostic examinations. In

this perspective, HEARTFAID CDSS was designed

for combining different models of reasoning, as will

be described in the next sections.

2.1 Significance of Signal Acquisition

and Analysis in HF

Electrocardiography (ECG) is one of the very basic

examinations performed in the evaluation and

assessment of HF. According to ESC (2005)

guidelines, the negative predictive value of normal

ECG to exclude left ventricular systolic dysfunction

exceeds 90%.

The most common ECG examinations are the

“Resting ECG” and the “Holter ECG”. While the

latter is more commonly used for the discovery of

rhythm abnormalities and the computation of the

Heart Rate Variability (HRV), the former is more

commonly used for the evaluation of morphological

abnormalities in the PQRST shape.

In both examinations, the first step to be

performed is the QRS detection with the

identification of the time occurrences of each heart

beat. This series of data allows for the evaluation of

the heart rate and is preparatory to the beat

classification for the discrimination between normal

and abnormal beats. This task is usually performed

in the “Holter ECG” reading stations as a starting

point for the arrhythmias’ classification and for the

evaluation of the NN series (time intervals between

consecutive normal beats) that is the input for the

HRV evaluation. In case of “Resting ECG”

examination (typical duration is 10 seconds), the

evaluation of the normal beats allows the normal

beat averaging with the construction of a more

noise-free reference beat that can be used for a better

evaluation of wave durations and amplitudes.

Wave durations and amplitudes are paramount in

the evaluation of the “Resting ECG” (usually with

12 leads) parameters of high significance for the HF

patients like ST depression, QRS and QT durations,

Sokolow-Lyon index for left ventricular

hypertrophy, presence of left or right branch bundle

block and presence of pathological Q waves.

2.2 Significance of Imaging Techniques

and Image Processing in HF

Imaging techniques offer invaluable aid in the

objective documentation of cardiac function,

allowing for the computation of parameters relative

to chamber dimensions, wall thickness, systolic and

diastolic function, regurgitations and pulmonary

blood pressure.

According to ESC (2005), chest X-ray and

echocardiography should be included in the HF

initial diagnostic work-up. Further,

echocardiography will be regularly repeated to

monitor in an objective way the changes in the

clinical course of a HF patient. Additional

techniques, like nuclear imaging and cardiac

magnetic resonance, may be also considered for

particular patients, since they have not been shown

to be superior to echocardiography in the

management of most HF population.

Thus, echocardiography —and in particular 2-D

TransThoracic Echocardiography (TTE) for its

portability and versatility— is the key imaging

technique for the practical management of HF.

The most important measurement performed by

TTE is Left Ventricle Ejection Fraction (LVEF),

which permits to distinguish patients with cardiac

systolic dysfunction from patients with preserved

systolic function. LVEF is given by the normalized

(non-dimensional) difference between left ventricle

End-Diastolic Volume (EDV) and the End-Systolic

volume (ESV). Among different models for the

computation of such volumes, the American Society

of Echocardiography (Lang et al., 2005) suggests the

HEALTHINF 2008 - International Conference on Health Informatics

290

use of the so-called Simpson’s rule, by which the

left ventricle is approximated by a stack of circular

(or elliptical) disks whose centres lie in the major

axis. Simpson’s method, therefore, relies on left

ventricle border tracing. It is well-known that

manual border tracing, besides being time-

consuming, is prone to inter- and intra- observer

variability, and thus is unable to provide a

satisfactory and reproducible measurement of

LVEF.

Image processing techniques may relieve this

problem, by providing automated or, at least, semi-

automated methods for tracing contours of relevant

structures found in an image, an issue called image

segmentation in the specific literature. However, the

segmentation problem for ultrasound images is by

no means trivial, due mainly to low signal to noise

ratio, low contrast, and image anisotropy and

speckle noise (Noble and Boukerroui, 2006).

3 A SIGNIFICANT SCENARIO

HEARTFAID CDSS was designed after a careful

analysis of the problems to be faced and the

expectations of the medical users.

A complete use case was defined for guiding the

development activity of CDSS by considering many

of the integrated services of the platform.

More in detail, we are considering a 65 years old

patient, already enrolled in the HFP, former smoker,

suffering from hypertension for several years. The

patient was enrolled in the HFP six months ago and,

in particular, the telemonitoring services offered by

the platform were activated. At the baseline visit, the

patient referred a slight limitation of physical

activity, since he felt comfortable at rest but ordinary

activity resulted in fatigue and dyspnoea. For these

reasons, the patient was assigned to NYHA class II.

Anamnesis data were also collected, from which it is

known that the patient had an acute myocardial

infarction five years before and he underwent to

aorto-coronary bypass. The patient had a post

ischaemic dilated cardiomyopathy with associated

systolic dysfunction.

The TTE test (performed at baseline evaluation)

showed an LVEF equal to 40%, ESV and EDV

being respectively 114 ml and 190 ml. The left

ventricle end-diastolic diameter was 6.0 cm. The

pharmacological treatment consisted in ACE-

inhibitor, beta-blockers, spironolactone, aspirin and

statin. Neither pulmonary nor systemic congestion

signs were present. Blood examinations of renal

function and electrolytes were normal. During these

six months, the patient has been telemonitored. In

particular, the pharmacological therapy has been

followed with care and no relevant changes have

been detected by the platform.

Suddenly, the patient observes a worsening of

his symptoms, with a marked limitation of physical

activity. After he fills in a periodic questionnaire

suggested by the platform based on Minnesota

questionnaire, the changes in the symptoms are

automatically detected and considered relevant. A

medical visit is suggested by the CDSS, accepted by

the referent physician and immediately scheduled.

At the visit, the NYHA class changes from II to

III. No variations in the signs are observed by the

cardiologist, apart from a slight worsening of blood

pressure (150/90 mmHg) and an increase of 10

beats/min in the heart rate. An ECG is performed

also to confirm the heart rate increment.

The cardiologist, supported by the CDSS,

decides however to evaluate other parameters by

echocardiography. During the TTE examination, the

sonographer acquires images and images sequences

according to a protocol specified by the platform.

Finally the images and the parameters manually

evaluated by the sonographer are stored in the

platform image archive. The reviewing cardiologist

visualizes the echocardiographic images and the

estimated parameters. Left ventricle volume and

ejection fractions are computed again by automatic

methods, exploiting the available image sequences.

These values are compared with the historical data

of the patient. EDV increases to 210 ml, ESV

increases to 145 ml, EF decreased from 40% to 30%.

Mild tricuspidal insufficiency is Doppler-

detected by its regurgitation. By tricuspidal

regurgitation extent, the pressure gradient (mmHg)

between right ventricle and right atrium is measured.

Pulmonary pressure is then estimated. With this aim,

the subcostal view is taken into account, so as to

determine Inferior Vena Cava (IVC) diameter and its

collapsibility index. The pulmonary pressure is

estimated to be 40 mmHg, by using a lookup table

with entries consisting in the tricuspidal gradient,

IVC diameter and collapsibility index. Since this

value indicates a slight pulmonary congestion, the

CDSS suggests the physician to integrate the

pharmacological therapy with diuretics, for example

loop diuretics or thiazides. Further, since there are

no up-to-date information about the renal function

and electrolytes, the CDSS suggests to start with a

safe diuretic dosage and to perform blood

examinations, which are scheduled for few days

later. The physician opts for a loop diuretics therapy,

for quicker beneficial effects.

Back to his home, the patient is monitored in the

subsequent days. In particular control of weight,

urine output, blood pressure, symptoms are

scheduled daily. Blood examinations are scheduled

seven days after the beginning of the new treatment.

DECISION SUPPORT AND IMAGE & SIGNAL ANALYSIS IN HEART FAILURE - A Comprehensive Use Case

291

The results of such blood examinations are uploaded

to the platform.

An up-titration table for the diuretics is compiled

by the CDSS, considering in particular symptoms

and electrolytes balance, creatinine clearance, blood

pressure, weight slope and urine output. The CDSS

also suggests to control weight and urine output

daily and to schedule blood examinations weekly. A

visit is also suggested in one month, to appreciate

the response to the therapy. The physician reviews

this program and decides to approve it. After

approval, the up-titration table for diuretics is

automatically sent to the patient.

One week after, telemonitoring evidences

persistence of symptoms; the patient is thus required

to continue the up-titration program for diuretics.

During the subsequent weeks symptoms get better

until the visit. At that visit, the patient refers that

symptoms are relieved. NYHA class is moved back

to II. However, the CDSS suggests the physician to

explore the possible origins of the change in the

symptoms reported in the previous visit (i.e. the

probable cause of heart failure decompensation). In

particular, with the aim of controlling the ischemic

disease, a stress test is scheduled.

3.1 Methods

The CDSS was defined for an overall support of HF

management, facing the main decisional problems of

diagnosis, prognosis, therapy and follow-up, by

using patients’ heterogeneous information (e.g.,

actual status, anamnesis, clinical history, diagnostic

parameters, and clinicians’ evaluation).

Ontologies combined with rules were chosen as

representation formalism, because of the more

suitable and up-to-date methodology for formalizing

the declarative and procedural knowledge derived

from the guidelines and the experts’ know-how.

Actually, ontologies constitute a logic-based

representation which also assures easy re-use and

sharing of knowledge. Moreover, the rule based

approach appeared the more appropriate both to

complete possible representation lacks of ontological

model and to involve more effectively the experts in

the elicitation process. An inference engine was then

devised for the corresponding inferential reasoning

processes, by induction and deduction on the

formalized knowledge for assessing patients’ status,

formulating diagnosis and prognosis, assisting

therapy planning, and monitoring patients.

Computational reasoning models were included

for those difficult HF decision problems, such as

prognosis assessment and early detection of patient’s

decompensation.

The HEARTFAID CDSS architecture was

designed according to a multilevel conceptualization

scheme for distinguishing among

 the knowledge level, corresponding to all the

information needed by the system for performing

tasks, e.g. data, domain knowledge, computational

decision models;

 the processing level, consisting of the system

components that are responsible for tasks

accomplishment by using the knowledge level;

 the end-user application level, including the system

components whose functionalities are specifically

defined for interacting with the user.

More in detail, the CDSS architecture consists of

the following components (Figure 1):

 Domain Knowledge Base, consisting of the domain

knowledge, formalized from the guidelines and of

the clinicians’ know-how;

 Model Base, containing the computational decision

models, signals and images processing methods

and pattern searching procedures;

 Meta Knowledge Base, composed by the strategy

knowledge, i.e. about the organization of CDSS

tasks.

 Brain, the system component endowed with the

reasoning capability, which is divided into the meta

and object level;

 Explanation Facility, providing a means to explain

conclusions taken.

The Brain was modelled by functionally

separating a meta level, devoted to task

accomplishment and organisation, and an object

level, responsible for actually performing tasks, by

reasoning on the computational and domain

knowledge. A Strategy Controller was inserted for

performing the meta level functionalities, by

orchestrating the two components of the object level,

i.e. the Inference Engine and the Model Manager.

Moving from the design to the development

activity, the use case is being used as a real scenario

for implementing the above architecture.

Figure 1: The general view of the HEARTFAID CDSS

architecture – dashed arrows correspond to reference to

the ontologies, while the others denote a direct

communication.

HEALTHINF 2008 - International Conference on Health Informatics

292

The required CDSS interventions consist in the

following services, listed in order of occurrence in

the workflow:

 interpretation of telemonitored data and

assessment of patient’s status;

 visit scheduling;

 suggestion of new diagnostic examinations;

 analysis of imaging examinations;

 interpretation of diagnostic findings;

 suggestion of therapy changes.

The necessary pieces of knowledge have been

identified as mainly symbolic and an elicitation

process has been performed for their formalization

in the Domai1n KB. Specific algorithms for

extracting a number of useful parameters from the

echo images have been developed and inserted into

the Model Base.

In particular, the use case highlights the

interventions of other components of HFP which

hold important roles in assuring the effectiveness of

the support services, e.g. the agenda for scheduling

new visits or examinations. Actually, HFP was

conceived for consciously distributing the work load

among the various components. A sketch of the

platform with the components that interact with the

CDSS is shown in Figure 2.

An EHR module was inserted for suitably

organizing, visualizing and managing patients’ data,

stored into the platform Repository. In particular, a

dedicated Repository for storing examination images

was conceived in accordance to the DICOM

standard. An Agenda module was included for

managing patients’ care planning, e.g., scheduling

new visits, prescribing new examinations and so

forth. The User Interface was designed as a

fundamental component, responsible for all the

interactions and communications with the users.

The different components of the platform were

seen as resources, by virtualizing the operations

required for their management. When involved, the

different components are dynamically integrated for

supplying sophisticated but much flexible

applications. The responsible for guaranteeing

integration and interoperability among all the HFP

components was defined as the platform

Middleware, which includes all the adapters

necessary for the virtualization. For simplifying the

provision of different services, a Service Controller

(SC) was comprised for managing platform events

and invoking the other components.

Figure 2: A sketch of HFP with the components that

interact with the CDSS.

In this perspective, the CDSS component was

designed as a resource able to offer a number of

functionalities and to interact with the other

resources for performing its tasks. Each decision-

making problem has to be translated into a request

or a class of requests committed to the CDSS, which

is then activated on-demand. The system handles

every request according to a specific policy encoded

in the Meta KB, interacting, when necessary, with

the other platform components.

A brief (and partial), discursive description about

how the scenario has been mapped onto a workflow

of HFP services is useful for understanding its

implementation.

Description of the Mapping into the HFP.

The patient answers a questionnaire through his

web-based user interface and sends the information

to the HFP that checks eventual missing values.

Then the Service Controller stores this information

into the repository, gets historical data and

opportunely invokes the specific CDSS service

responsible for handling the request.

The CDSS analyzes data and answers supplying

the current patient’s status, i.e. worsening of

symptoms, and a set of suggested actions the

clinician should undertake, i.e. schedule a new visit,

change the NYHA class or change the therapy and

so forth. Then the SC stores CDSS results into the

repository.

When the doctor on duty logs in the HFP, the list

of patients is displayed ordered by their severity

status and the timestamp of the last related event.

Once the patient is selected, the change in his status

is shown along with the list of suggested actions, for

instance as a list of operations that can be selected.

He then approves the schedule of the visit and the

SC forwards the request to the Agenda component

that opportunely records it and informs the patient.

During the visit, the physician inserts his

observations into the patient’s record and decides to

approve the change of the NYHA class: he selects

the corresponding action within the list and the SC

DECISION SUPPORT AND IMAGE & SIGNAL ANALYSIS IN HEART FAILURE - A Comprehensive Use Case

293

takes care of registering the change in the patient’s

record. An ECG is then performed acquired by the

platform and processed with QRS detection and

classification algorithms in order to produce noise-

free reference beats for all the 12 leads. The

availability of graphical tools for the signal display

and measurements (ruler) allows the cardiologist for

an accurate and reliable evaluation of the significant

morphological parameters. Then the information

obtained by the ECG is inputted and a request is sent

by the SC to the CDSS, which suggests performing

an echocardiography as displayed in the

recommended actions list.

Inference on Patients’ Data. In order to include

information derived from the use case, starting from

a preliminary ontology mainly corresponding to a

structured terminology of the domain, we began to

develop a new ontology by inserting relevant

properties, classes and relations for a coherent and

comprehensive formalization, also in accordance to

standard medical ontologies, such as the Unified

Medical Language System (UMLS, 2007). An

excerpt of this new ontology is shown in Figure 3.

Figure 3: Some relevant classes and properties of the

ontology.

A careful elicitation activity was performed for

formalizing a set of rules, founded on the developed

ontology and to be encoded in the KB. An elicited rule

which is used for therapy suggestion is: “If a patient has

Left Ventricle Ejection Fraction <= 40% and he is

asymptomatic and is assuming ACE Inhibitors or ARB)

and he had a myocardial infarction then a suggestion for

the doctor is to give the patient Betablockers”.

Echocardiography Images Analysis. A prototype

module for the computation of LVEF has been

developed (Barcaro et al., 2007). The module is able

to process an apical-view sequence of the heart (the

so-called two- and four-chamber views) in order to

identify the left ventricle cavity in every frame of the

sequence. This segmentation stage is accomplished

augmenting a variational formulation of level set

methods with mimetic criteria for contour

initialization (see Figure 4).

Figure 4: Segmented left ventricle cavity at end-systole.

After segmentation, the left ventricle volume is

computed as a function of time by applying

monoplane Simpson's method. Then ESV and EDV

are regarded as the minimum and maximum

respectively of the volume time-course. Finally

LVEF is obtained by the simple formula:

EDV ESV

LVEF

EDV

−

4 RESULTS AND CONCLUSIONS

A number of tools and instruments are available for

developing the CDSS according to the design

specifications. The key factors that were taken into

account for defining an up-to-date system were

accordance to standards, efficiency and robustness.

We investigated several technologies, with

particular attention to the Semantic Web field, since

it offers various tools for building ontological

models, knowledge bases and reasoning on them.

Moreover, the platform was conceived for web

applications developed in Java. For selecting the

appropriate tools, we carefully analyzed the W3C

recommendations along with the performance,

compatibility and maintenance of the same tools.

As to the knowledge representation formalism,

we selected the Web Ontology Language (OWL,

2007) for defining the ontologies, since it is the

actual de-facto standard semantic markup language

for this task.

The ontology has been built using the two editors

Protégé and Swoop. For defining the rules of the

KB, we chose SWRL (2007), the Semantic Web

Rule Language combining OWL and Rule Mark-up

Language, which is a submission to W3C that

extends the set of OWL axioms to include Horn-like

HEALTHINF 2008 - International Conference on Health Informatics

294

rules. For realizing the reasoning component, Jena

(McBride, 2001) was selected as a Java

programmatic environment that includes OWL, a

language for querying ontologies, SPARQL (2007),

and a rule-based inference engines. In particular, for

improving the reasoning capability of the latter, we

also used Bossam (2007) and Pellet (2007). An

example of the rules we are developing in SWRL is

shown in Figure 5 as it has been defined in Protégé.

Figure 5: A rule developed in SWRL.

Prototypical methods for processing the echo

images were realized implementing the various

procedures in Matlab.

Future activities consist in finalizing the platform

implementation by concluding the realization of the

Domain KB, the algorithms contained in the Model

Base and the Brain, in particular of its meta level for

integrating all the object models and the interface.

ACKNOWLEDGEMENTS

This research work is supported by the European

Community, under the Sixth Framework

Programme, Information Society Technology – ICT

for Health, within the STREP project HEARTFAID

(IST-2005-027107), 2006-2009.

REFERENCES

Barcaro, U., Moroni, D., Salvetti, O. (2007) Left ventricle

segmentation in ultrasound sequences for the

automatic computation of ejection fraction. In

OGRW07 Open Gern Russ Work on Pat Rec & Im An.

Bayegan, E. Nytro, O. Grimsmo, A. (2002). Ontologies

for knowledge representation in a computer-based

patient record. In ICTAI 2002, 14th IEEE Int. Conf. on

Tools with Artificial. IEEE.

Bosmans, J.E., Dassen, W.R.M., van der Velde, E.T.,

Dijk, W.A., Spruijt, H.J., Baljon M.H. (2000).

Towards ICT Support for Treatment of Congestive

Heart Failure. Computers in Cardiology, 27, 641-644.

Bossam (2007) http://bossam.wordpress.com/

Boxwala, A.A., Peleg, M., Tu, S. et al. (2004) GLIF3: a

representation format for sharable computer-

interpretable clinical practice guidelines. J. Biomed

Inform., 37(3), 147-161.

Dassen, W.R.M., Gorgels, A.P.M., Berendsen, A., Dijk,

W.A., de Clercq, P.A., Hasman, A., Baljon M.H.

(2003). Guideline Assessment and Implementation in

Congestive Heart Failure. Computers in Cardiology,

30, 331-334.

ESC (2005) Swedberg, K. et al., “Guidelines for the

diagnosis and treatment of Chronic Heart Failure: full

text (update 2005)”, European Heart J, 45 pages.

GEM (2003) E2210-02 Standard Specification for

Guideline Elements Model (GEM) Book of Standards

Volume: 14.01. ASTM International.

Greenes, D. (2007). Clinical Decision Support: The Road

Ahead. Academic Press.

Helbig, H. (2006) Knowledge Representation and the

Semantics of Natural Language, Springer, Berlin,

Heidelberg, New York.

Lang, R.M. et al., (2005). Recommendation for Chamber

Quantification.. J Am Soc Echocardiogr, 18, 1440-

1463.

McBride, B. (2001) Jena: Implementing the RDF Model

and Syntax Specification. In WWW2001, Semantic

Web Workshop.

Noble, J.A., & Boukerroui, D. (2006). Ultrasound image

segmentation: a survey. In IEEE Trans. Med. Imag.,

25, 987-1010.

OWL (2007) http://www.w3.org/TR/owl-features/

Park, R.W., Shin, S.S., Choi, Y.I., Ahn, J.O., Hwang, S.C.

(2005). Computerized physician order entry and

electronic medical record systems in Korean teaching

and general hospitals: results of a 2004 survey. J Am

Med Inform Assoc. 12(6):642-7.

Pellet (2007) http://pellet.owldl.com/

Perlini, S. et al. (1990) Treatment of chronic heart failure:

an expert system advisor for general practitioners.

Acta Cardiol., 45(5), 365-78.

Poissant, L., Pereira, J., Tamblyn, R. Kawasumi, Y.

(2005). The impact of electronic health records on

time efficiency of physicians and nurses: a systematic

review. J Am Med Inform Assoc. 12 (5), 505-16.

SPARQL (2007) http://www.w3.org/TR/rdf-sparql-query/

UMLS (2007). http://www.nlm.nih.gov/research/umls/

SWRL (2007) http://www.w3.org/Submission/SWRL/

VV.AA. (2007). Functional Specifications of Data

Processing and Decision Support Services.

Deliverable D15, HEARTFAID IST-2005-027107.

Wirtz, V., Cribb, A., Barber, N. (2006). Patient–doctor

decision-making about treatment within the

consultation—A critical analysis of models. Social

Science & Medicine. 62, 1, 116-124.

DECISION SUPPORT AND IMAGE & SIGNAL ANALYSIS IN HEART FAILURE - A Comprehensive Use Case

295