Computerised Decision System for Diabetes Melitius [618113]

Computerised Decision System for Diabetes Melitius
and associated complications – CODES
Gabriel Spiridon, senior researcher, PhD
IPA SA – R&D Institute for Automation
Bucharest , Romania
[anonimizat]

Dorin Carstoiu, Prof., PhD
University POLITEHNICA of Buc harest
Bucharest, Romania
[anonimizat] Anca Sarbu, Lecturer , MD, PhD
Univ. of Medicine and Pharmacy "Carol D avila"
Bucharest, Romania
[anonimizat]

Florian Ion, researcher
IPA SA – R&D Institute for Automation
Bucharest, Romania
[anonimizat]

Abstract —This Diabetes mellitus (DM) and associated
cardiovascular complications represent two of the major global
healthcare problems. In Romania, the prevalence of DM is
approximately 6% and the prevalence of pre -diabetes states (also
a cardiovascular risk factor) is 8.5%. The risk of complications
decreases considerably [11, 12] by controlling the risk factor s.
Computerized care surveillance technologies are strongly
encouraged also in eHealth Action Plan for 2012 -2020, European
Commission. Improvement of risk prediction for cardiovascular
complications in patients with DM is crucial to the identification
of h igh-risk individuals who could benefit from preventive
measures. There is a realistic necessity of a computerized system
for prevention and management of these risk factors. We
propose in this paper a developing concept of an Intelligent
Clinical Decision Support System for Diabetes Mellitus (DM) and
Related Cardiovascular Complications, based on Artificial
Intelligence Systems. The purpose of the system is to assist
clinicians in establishing the complications risks for DM patients,
to provide recommendat ions and alarms during patients’
investigation, based on Clinical Guidelines and to generate
evolution patterns for DM complications, using a combination of
traditional risk factors, novel biomarkers and imagistic tests.
Keywords —expert systems, neural net works, diabetes mellitus,
biosensors, medical guidelines, medical data, computerized systems
I. INTRODUCTION
There is the necessity to develop an Intelligent Clinical
Decision Support System for DM and related Cardiovascular
complications, using Knowledge B ased Systems and having in
structure an Expert System and a complementary Neural
network, the last one well suited to problems with a high
degree of complexity [5,7] for which there is no algorithmic
solutions or the solutions are too complex for traditional
techniques to determine, being perfectly tailored for the
identification and prediction of the cardiovascular
complications in DM. The purpose of the system is to assist
clinicians with recommendations and alarms during patients’
investigation, based on Clinical Guidelines, to establish the
correct predictions for patients, by defining a set of rules connecting risk factors to specific complications. The system
will provide complete solutions for monitoring the patients’
evolution and their complete set of medical records, to generate
a second medical opinion. An important role of the system is to
assist medical research in discovering new patterns in DM and
Cardiovascular complications, using a combination of
traditional risk factors, as well as novel biomarkers and
imagistic tests. We propose a novel solution, based on a
scientifically -technically sound concept that will be a major
contribution to the knowledge in the field. The solution could
be validated through clinical trials with the direct scientif ic
support of the new proposed system. The impact of the
proposed solution will be in direct benefits to the patients and
medical care system and could be demonstrated through
clinical trials. The system has to be developed according to
international medic al standards – HL7 Clinical Document
Architecture, for storing the patient medical data, HL7
communication protocol to interact with the hospital
information, IEEE 11073 Point -of-care Medical Device
Communication to receive data from the Vital Signs monito r
devices, DICOM, etc. -, in order to facilitate the interconnection
with other medical information systems in operation for to
assure in this mode a high acceptance and repeatability degree.
The proposed system will have the possibility to be used by
different medical units after it’s validated through Final
medical trials. This “external multi- user” function will be
possible using Cloud Computing, in order to assure a large
spreading and a majority accepted solution after system
validated in medical resea rch and real Clinical practice. As
related work , research for artificial intelligence use in medicine
has constantly grown, producing a number of experimental and
commercial systems: INTERNIST, MYCIN,, ONCOCIN,
DXplain, GRIP, UKPDS Risk Engine. Regarding t he use of
artificial intelligence in the diagnosis and management of
diabetes, several decision support systems have been proposed,
based on different approaches. ESDIABETES is an expert
system for diabetes, developed by computer science graduate
students at Texas A&M University Corpus Christi, designed to
help people monitor and control the blood glucose level. Akter

et. al. present a low -cost automated knowledge- based system
dedicated to self -diagnosis and management of diabetes for
both patients and doct ors. Kalpana and Kumar introduce a
fuzzy expert system for diabetes using fuzzy verdict
mechanism (2011). The purpose of the system is to assist
medical practitioners in establishing the diagnosis of diabetes.
Setlak et. al. propose an intelligent system for automatic
classification of diabetes patients into one of the 4 diabetes
types. Based on background medical information, tests results
and symptoms, the system will decide under which diabetes
category the patient falls. Campos -Delgado et . al. develop ed a
fuzzy -based expert knowledge controller for regulating the
blood glucose level.
Most efforts have been made to develop Clinical Decision
Support Systems (CDSS) for diagnosing diabetes and to
monitor blood glucose level. However, the complications ris k
of DM patients is an important aspect and having a good
anticipation of risk factors can significantly increase their
quality of life and prevent the evolution of their disease. The
CODES system will demonstrate progress beyond the state -of-
the-art by fi lling this gap between current advances in CDSS
and traditional complication risk evaluation methods. This
project will provide medical specialists with a novel tool for a
timely prediction of DM complication factors. Moreover, the
neural network will serve as a research instrument, which will
be trained to discover complication patterns and evolutions in
DM patients. By integrating the decision support system with
the neural network, the CODES system will be able to predict
the evolution of DM in patients, based on their personal history
and evolution patterns. This project will provide an example of
medical research applied into clinical practice (translational
research), thus lifting the barriers between experimental
science and practice in the field of d iabetes and cardiovascular
complications.
II. TECHNICAL AND SCIENTIFIC SOLUTION
DESCRIPTION
We propose the developing concept of an Intelligent
Clinical Decision Support System for Diabetes Mellitus (DM)
and Related Cardiovascular Complications, based on Artif icial
Intelligence Systems. The purpose of the system is to assist
clinicians in establishing the complications risks for DM
patients, to provide recommendations and alarms during
patients’ investigation, based on Clinical Guidelines and to
generate evolut ion patterns for DM complications, using a
combination of traditional risk factors, novel biomarkers and
imagistic tests. The proposed solution will be validated through
clinical trials. The impact of the proposed system will be
reflected in direct benefit s to the patients and medical care
system. The possible architecture of such a system is depicted
in the Fig.1 below .

III. SYSTEM COMPONENTS

The proposed interdisciplinary platform is made out of the
following components:
A. Medical Data Interface (MDI)
MDI will collect medical data from various sources: Vital
Signs Monitor, Continuous glucose monitoring systems, Blood
pressure meters, Bio -Thesiometer, EKG, Blood tests, Patient
medical history, Radiology studies. Information regarding the
following a spects will be collected through the MDI
component:
• Clinical evaluation: Personal and family history
regarding the presence and the evolution of diabetes,
and other known risk factors (obesity, CVD), complete
physical exam, anthropometrical parameters (Hei ght,
weight, waist circumference, waist/hip ratio)
• Monitoring values: blood glucose, blood pressure and
so on, collected from various monitoring devices (vital
signs monitor, sensors etc.)
• Lifestyle information : Physical inactivity, smoking,
alcohol consum ption, nutrition evaluations. These data
will be obtained through a series of validated
questionnaires.
• Biochemical test results: glucose metabolism (HbA1c,
24 h glycemic profile), hepatic tests, lipids profile (total
and fractioned cholesterol, triglyceri des, apolipoproteins
B and A1), inflammatory profile (fibrinogen, CRP,
homocystein, IL -6), micro/macroalbuminuria, markers
of cardiac stress (B -type natriuretic peptide BNP and N –
terminal proBNP), sensitive marker of renal dysfunction
(cystatin C).
• Hormon al investigations: leptin, adiponectin, thyroid
hormones, total testosterone, SHBG, FSH, LH in men
• Psychological examination: eating behavior
disturbances and personality type
• Imagistic procedures results: whole body DXA (to
evaluate visceral adipose tiss ue)
• Cardiac investigations: Cardiac ultrasound, vascular
Doppler (carotid arteries), basal ECG, effort test, sleep
registration
An important aspect to be considered here is the heterogeneity
of the data coming from various sources. Therefore, MDI has
to integrate [ 11, 12] a component for the extract -transform -load
(ETL) process, which will bring all data into a unitary format.
The goal is to obtain data from as many sources as possible, in
order to have a complete picture of the patient’s status and
treatments he is under. This component could be developed in
a modular and scalable manner, in order to easily allow new
sets of medical data to be included.

Fig. 1 – Intelligent Clinical Decision Support System Architecture

B. Clinical Decision Support System for Cardiovascular
Complications in Diabetes (CDSSCCD)
CDSSCCD is a de cision support and screening tool [3, 4, 6] ,
which is aimed to assist medical specialists in their decision
making process regarding DM patients and the complications
risk factors they are facing. The system is connected to the
MDI in order to receive biom edical information about patients.
It could contain two Artificial Intelligence Systems:
• An Expert system, which will generate predictions
regarding the cardiovascular complications risk for each
investigated patient.
• A Neural Network, which will be tra ined using datasets
provided by medical partners . The neural network role
is to generate decision trees on the basis of which new
pieces of knowledge are generated, thus obtaining a
self-training system. Other approaches based on the
algorithm ID3, C4.5, C 5 etc. are possible.
IV. EXPERT SYSTEM DESCRI PTIO N
The expert system has the following main components [ 8,
3]:
• Knowledge base, which stores the connections between
data, concepts, and statistical probabilities, in order to
allow the reasoning part of the syste m to perform an
accurate evaluation of a potential problem. Knowledge
bases go beyond "if – then" statements ; they define
associative relationships among concepts and statistical
information about the probability of certain solutions.
The knowledge base of the expert system will
interconnect data from the following sources:
computerized medical guidelines for diabetes mellitus
and cardiovascular pathology and medical knowledge
provided by medical specialists. Moreover, the outputs
generated by the neural network will be used as input
for the knowledge base. Thus, the knowledge base will
constantly increase its reasoning capacity.
• Facts base, which stores data collected from patients.
The inputs of the facts base is provided through the
MDI component, and col lects the patients’ data
gathered from various sources.
• Inference engine, the “brain” of the expert system,
designed and implemented to reason upon the
information from the facts base, correlated with the
knowledge base, in order to produce new conclusions
and perspectives. The output produced by the inference
engine consists of risk prediction for cardiovascular
complications in DM patients, medical
recommendations – as second opinion, alerts – to inform
the doctor about special situations, such as interf erences
between 2 prescribed medications.
V. THE NEURAL NETWORK
A special component of the system is the neural
network [5,7] , which is aimed at producing DM complications
evolution patterns and serves as a medical research tool. The
inputs of the neural netwo rk will be: a) Clinical evaluation: Personal and family history
regarding the presence and the evolution of diabetes, and other
known risk factors (obesity, CVD), complete physical exam,
anthropometrical parameters (Height, weight, waist
circumference, waist/ hip ratio)
b) Lifestyle factors : Physical inactivity, smoking,
alcohol consumption, nutrition evaluations (validated
questionnaires)
c) Biochemical test: glucose metabolism (HbA1c, 24 h
glycemic profile), hepatic tests, lipids profile (total and
fractioned chole sterol, triglycerides, apolipoproteins B and
A1), inflammatory profile ( fibrinogen, CRP, homocystein, IL –
6), micro/macroalbuminuria, markers of cardiac stress (B -type
natriuretic peptide BNP and N -terminal proBNP), sensitive
marker of renal dysfunction (c ystatin C),
d) Hormonal investigations : leptin, adiponectin, thyroid
hormones, total testosterone, SHBG, FSH, LH in men
e) Psychological examination: eating behavior
disturbances and personality type
f) Imagistic procedures: whole body DXA (to evaluate
visceral adipose tissue)
g) Cardiac ultrasound, vascular Doppler (carotid
arteries), basal ECG, effort test, sleep registration
Before being integrated with the expert system, the neural
network has to undergo the learning phase. Once an input is
provided to the neural network, a corresponding target
response has to be defined at the output. An error will be
established based on the difference between the desired output
and the real system output. The error metrics is looped back to
the system, which must automatically adjust the parameters
weight, in order to obtain the desired output. In our case, the
desired output is a set of well- defined rules which establish the
influence of the input factors to the cardiovascular
complications of diabetes mellitus and evolution p atterns and
models for complications. One of the main purposes of the
neural network is to optimize the response time for obtaining
the desired output. The neural network will be designed to have
at least 3 layers and will be trained using medical datasets
provided by the medical partners of the project. Moreover,
there is a significant scientific research factor related to the
neural network [3,10] . An important milestone in the
development of the proposed solution will consist in
identifying the most adeq uate structure of the neural network,
as well as to optimize the learning algorithm used to train the
network. Thus, the medical research community will benefit
from a set of clear results, which will enhance future growth of
the knowledge base, and theref ore will lead to an optimized
inference engine and clinical decision support system.
The integration between the Expert system and the neural
network facilitates data interoperability in the knowledge base.
Their cooperation involves several interactions:
• The neural network provides significant decisional data
to the expert system;
• The expert system controls the learning process of
neural network;

• The expert system processes the neural network data
and correlates them with the computerized medical
guideline s and other specific medical information.
The below figure depicts the System components, with
their associated main functionalities.

Fig. 1 – System components description

VI. ORIGINAL AND INNOVATI VE CONTRIBUTIONS
Some key factors in the original contributions brought by
the proposed system are:
• The proposal of an Intelligent Clinical Decision Support
System for assessing, evaluating and predicting
cardiovascular complications in DM patients;
• Setting up and training an artificial ne ural network
based on genetic tests, nutritional factors and various
lifestyle indicators;
• Identifying the best suited training algorithm for the
neural network;
• Establishing a set of rules between various risk factors,
genetic tests, medical history, nutr itional and lifestyle
factors and the degree and severity of cardiovascular
complications in DM patients;
• Generating predictive models and evolution patterns for
cardiovascular complications for DM patients;
• Proposal for extending the medical guidelines, b y
extending the basic rules they define with new sets of
rules assessed through the neural network and expert
system;
• Decreasing the prevalence of cardiovascular
complications of DM will lead to an increase in average
life expectancy, in quality of life in dicators (increasing
healthy life years), maintaining social and economic integration of patients with DM and cardiovascular
diseases;
• Assessing the performance of new markers in the
evaluation of cardiovascular risks.

REFERENCES

[1] Dorin Carstoiu, Alexandr a Cernian, “Analyzing and tracking the
evolution of rehabilitation treatment for patients with locomotory
deficiencies ” (Chapter 18 pag.) Handbook of Research on Developments
in E-Health and Telemedicine: Technological and Social Perspectives,
Edited by: M aria Manuela Cunha, Polytechnic Insti¬tute of Cavado and
Ave, Portugal; Antonio Jose Tava¬res, Polytechnic Institute of Cavado
and Ave, Portugal; Ricardo J. Simoes, Polytechnic Institute of Cavado
and Ave, and University of Minho, Portugal,, ISBN: 978 -1-61520 -670-
4, IGI Global 701 E. Chocolate Avenue Hershey, PA 17033, USA
[2] Dorin Carstoiu, Alexandra Cernian and Adriana Olteanu, “ Biomedical
Data Correlation And Reuse In Analyzing The Efficiency ”, The 2008
IEEE International Conference on Information Reuse an d Integration,
Las Vegas july 2008, p. 127 -130.
[3] Spiridon, G., Udr escu,F., Dichiu, D., “Using Artificial Neural Network
to Monitor Children's Gait Improvement ” – eGeH ’08 eGOVERNMENT
& eHEALTH 5TH INTERNATIONAL CONFERENCE AND
EXHIBITION – Desio, Milano, J une 2008; Co- chairman for eHealth
section.
[4] Spiridon, G., Dorobantu, M., Fruntelata, A., Firican, S., Bis triceanu, D.,
Zinca A., “Interdisciplinary Developments for Clinical Decision Support
Systems ” EMMIT 2007 – Euro – Mediterranean Medical Informatics and
Telemedicine 3rd International Conference, May 3 -5, 2007, Mangalia,
Romania – ISBN 978 -973-739-423-1.
[5] Bishop CM. “Neural Networks for Pattern Recognition ” Clarendon
Press, Oxford, 1995.
[6] Spiridon, G., Bistriceanu, D., Firican, S., Zinca A., – SAPHIRE –
“Intelligent Healthcare Monitoring Based on a Semantically Enriched
Interoperability Platform ” Clinical Guideline Models – Methods and
Technologies for Creating Clinical Guidelines, 7th European Conference
E- COMM -LINE 2006, Bucharest, September 18 -19, ISBN 973-
88046 -0-4; ISBN 978 – 973-88046 -0-9;
[7] R. Savitha, S. Suresh, N. Sundararajan – “A meta -cognitive learning
algortithm for a fully compex -valued Relaxation Network ” Neural
Network, The Official Journal of the International Neural network
Society, volume 32, August 2012;
[8] Carstoiu, D., Olteanu A., “ Knowledge Based Systems ”, Politehnica
Press, 2002
[9] ***, Neural Network Journal; 2012 Special Issues, Selected Papers
from IJCNN – International Journal Conferences of Neural Network
Society; volume 32, August 2012 .
[10] Lucila Machadob “Neural Networks in medicine” ; Decision Systems
Group Brigham and Women’s Hospital – 2002. The dynamics of life
expectancy over the last six decades in elderly people with diabetes
Ioacara S, Guja C, Fica S, Ionescu -Tirgoviste C.. Diabetes Res Clin
Pract. 2012. DOI 10.1016/j.diabres.2012.10.020
[11] Ioacara S, Guja C, Ionescu -Tirgoviste C, Fica S, Sabau S, Radu S, al..
“Improvements in life expectancy in adult type 2 diabetes patients in the
last six decades Diabetes Res Clin Pract. 2011 ”;92(3 ):400 -4
[12] Popescu AC, Antonini -Canterin F, Enache R, Nicolosi GL, Piazza R,
Faggiano P, Cassin M, Dimulescu D, Ginghină C, Popescu BA. “Impact
of associated significant aortic regurgitation on left ventricular
remodeling and hemodynamic impairment in severe aortic v alve
stenosis” . Cardiology. 2013;124(3):174 -81