ARTIFICIAL NEURAL NETWORK APPROACH

FOR OBESITY-HYPERTENSION CLASSIFICATION

Octavian Postolache

, Joaquim Mendes

Instituto de Telecomunicações, Av. Rovisco Pais, 1049-001, Lisboa, Portugal

IDMEC, Faculdade de Engenharia UP, Portugal

Gabriela Postolache

, Pedro Silva Girão

Universidade Atlantica, Escola de Saude, Oeiras, Portugal

Instituto de Telecomunicações, Av. Rovisco Pais, 1049-001, Lisboa, Portugal

Keywords: Obesity-hypertension syndrome, Artificial neural network.

Abstract: One of the newest targets of public health is management of obesity-hypertension. In this paper is presented

the use of an artificial neural network based model for objective classification of obesity-hypertension.

Different neural network architectures as part of hybrid processing scheme including comparators and

competitive processing blocks were developed and tested. The neural network functionality is the

classification of the individuals according to the obesity risks. The results show that the neural network

classifier is consistent with the standard criteria suggested by the obesity and hypertension guidelines.

1 INTRODUCTION

Obesity is rapidly turning into an “epidemic”

afflicting much of the industrialized world. Obesity

is a major risk factor for serious non-communicable

diseases such as cardiovascular disease,

hypertension, stroke, diabetes mellitus and various

forms of cancer. Therefore, it will be one of the

major causes of death, according to the estimation of

the World Health Organization, which suggest that

by 2025 approximately 60% of deaths worldwide

will be caused by circulatory diseases and cancers

(WHO, 2000). The relationship between obesity and

hypertension appears to be non-linear and exists

throughout the non-obese range. Obesity by itself

possibly accounts for 78% and 65% of essential

hypertension in men and women, respectively,

according to data from the Framingham Cohort

(Kannel et al., 1993). Hyperinsulinemia,

hyperleptinemia, hypercortisolemia, renal

dysfunction, altered vascular structure and function,

enhanced sympathetic and renin/angiotensin system

activity, and blunted natriuretic peptide activity

stand out as major contributory mechanisms to

“obesity - hypertension” (Tuck et al., 1981; Hall et

al., 2002; Mansuo et al., 2000; Engeli & Sharma,

2002). Furthermore, according to the European

Society of Hypertension and the European Society

of Cardiology (ESH-ESC) guidelines, hypertension

induces high added risk for target organ damage,

diabetes, or associated clinical conditions (ESH-

ESC, 2003). Moreover, organ damage and

associated clinical condition in obese people

increase with the extent of risk factor clustering

(Narkiewicz, 2006a). Therefore, objective diagnosis

of obesity-hypertension

is an important public health

challenge because of its high frequency and

concomitant risk of cardiovascular and kidney diseases.

Details regarding hypertension risk stratification

have been published: 1999 WHO/ISH Guidelines

(Chalmers et al., 1999), 2003 ESH-ESC Guidelines

(ESH-ESC, 2003), JNC6 of USA (Sheps et al.,

1997) and the Guidelines for the Management of

hypertension of China (Ministry of Health People’s

Republic of China, 1999). In what concerns obesity,

the evidences from the literature show a continuous

relationship between gradation of body mass index

(BMI), waist circumference, waist to hip ratio and

514

Postolache O., Mendes J., Postolache G. and Silva Girão P. (2009).

ARTIFICIAL NEURAL NETWORK APPROACH FOR OBESITY-HYPERTENSION CLASSIFICATION.

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing, pages 514-520

DOI: 10.5220/0001553705140520

 SciTePress

health risk (WHO, 2000; Sowers et al., 2001; Health

Canada, 2003; Lau et al., 2007, Ergun, 2008).

Nowadays, diagnosis of obesity is made mainly

according to body mass index (BMI) and waist

circumference. Unfortunately, comparisons by

ethnicity and sex have revealed that the universal

application of criteria for obesity and central obesity

developed in Caucasians leads to an overestimation

of risk in African Americans and an underestimation

of risk in South Asians (Sumner et al., 2007). Also,

obesity identification by generational trends showed

those generations prior to the "Baby Boomer", who

were not exposed to more recent unhealthy food

consumption patterns as younger people, are less

likely to be obese, in spite of their age (Garavagli &

Synthelabo, 2004). In addition, geography also plays

a role (Garavagli & Synthelabo, 2004). Also, new

Canadian Guidelines on the management and

prevention of obesity in adults and children

emphasised the importance to measure depression

and other mood disorders beyond BMI, waist

circumference and laboratory parameters (as fasting

blood glucose level, total cholesterol, LDL

cholesterol, HDL cholesterol, triglycerides, ratio of

total cholesterol to HDL cholesterol, liver enzyme

levels and urine analysis) (Lau et al., 2007).

Moreover, reverse epidemiology it was shown in

patients characterized by undergoing haemodialysis,

with increased survival of obese patients

(Narkiewicz, 2006b, Salahudeen et al., 2006). To

deal with these issues, much research is needed to

develop improved statistical methods that permit

coherent management, taking into account the global

risk of a patient with obesity-hypertension rather

than to focus solely on biometric variables and blood

pressure values. Objective diagnosis of obesity-

hypertension is important not only to prevent the

progression of obesity and hypertension signalling

the health risk but also for treatment management of

hypertension in obese people (Dentali et al., 2007;

Messerli & Schmieder, 1986, Narkiewicz, 2006a,

Narkiewicz, 2006b) reduction of the anaesthesia-

related mortality (Saravanakumar et al., 2006), and

health costs reduction (WHO, 2008; Lewis & Man,

1999).

The aim of this study now reported was to

evaluate the performance of an artificial neural

network (ANN) for modeling and objective

identification of the obesity-hypertension

physiopathology joining different informations

related to the patient health status. The ANN, a kind

of black box model, shows certain advantages over

other methods for multivariate modeling. The major

advantage used in the present application is that with

sufficient data, an ANN can be trained to learn the

relationship between the inputs (clinical examination

data) and outputs (obesity-hypertension classes)

even if the mechanism of the relationship is

unknown or unclear, as in the present case where

few models are described in literature for

hypertension-obesity stratifications criterion (see

Aneja et al., 2004, Narkiewicz, 2006a, Narkiewicz,

2006b). The ANN ensures great flexibility

associated with computer diagnosis of the

hypertension-obesity sindrome (Ning et al., 2006;

Bidiwala et al., 2004; Lapuerta et al., 1995;

Mangiameli et al., 2004; Orunescu et al., 2004, Poli

et al., 1991). Our application of the ANN permits

better diagnosis and management of the obesity-

hypertension syndrome.

2 METHODS

The problem in diagnosis and management of

obesity is that the relationship between different

items (e.g. laboratory results and/or symptoms) is

not always well established and that there exists a

myriad of exceptions for every rule. A learning

process expert system could be developed using

neural networks for medical decision aid. The

schematic drawing of the model is shown in Fig. 1.

The figure includes a multilayer perceptron neuronal

network classifier (MLP-NN) whose inputs are

expressed by values of clinical examination data.

MLP-NN classifier

COMPETE proc.

& &

ohc

ohca

Figure 1: Obesity-Hypertension Neural Network

Processing Scheme (ohcm – obesity-hypertension classes,

ohca1, ohca2- obesity-hypertension additional classes).

ARTIFICIAL NEURAL NETWORK APPROACH FOR OBESITY-HYPERTENSION CLASSIFICATION

515

The competitive processing block calculates the

binary output (ohc

, ohc

…) using the values

obtained with the MLP-NN classifier. Additional

classes (ohca

, ohca

) are obtained by using the level

of blood-pressure and the hypertension thresholds

(e.g. SBP > 140mmHg, DBP > 90mmHg) and the

results of competitive processing block for the

particular cases of normal and overweight

individuals.

Table 1: Variables range used to describe obesity-

hypertension syndrome class 3.

A brief description of the main blocks of the

obesity-hypertension general classifier is presented

next.

2.1 The Input of the Model

In the present study, a model of the clinical

parameters distribution that is essential for diagnosis

and monitoring was built using published data. The

evidences on relation between hypertension and

obesity are mainly documented on adult people (18-

50 years), overweight, or with class 3 obesity

(BMI>40 kg/m

). However, there are a lack of

studies that may give a thorough view on the main

clinical indicators (see Aneja et al., 2004), which

may better describe the present state of an individual

and possible future evolution of hypertension in

relation with obesity.

Since there were no well-characterized real

datasets available that fit with all obesity classes

described in our study, a simulation study was

proceed. The input data considered was essential

clinical information for obesity-hypertension

association, being expressed by the values of the

following parameters: body mass index (BMI), waist

girth, blood pressure, heart rate, triglyceride,

glucose, high-density lipase (HDL)-cholesterol

(Aneja et al., 2004), total cholesterol (Ai et al., 2000,

Aguilera et al. 2008), low-density lipase (LDL)-

cholesterol (Aguilera et al., 2008, Gupta et al.,

2007). The simulated values for different obesity

classes were adjusted according to the published

data applying uniformly random data distribution for

specific data intervals. The maximum values for

morbid obese group were built taking into account

clinical cases of obesity described in 19

century –

Daniel Lambert (Table 1). The samples sizes were

simulated for 400 to 1600 individuals, 40 to 200

individuals for each class.

The simulated data were not defined using sex-

specific observation points. However, a future study

taking into account the data distribution versus sex

and age will be considered.

Although the parameter settings are not

exhaustive in terms of all physiopathological

plausible situations, the outlined conditions are

reasonable, mainly designed to differentiate obesity

and hypertension features.

2.2 The Output of the Model

The classification of the simulated data were made

according to 2000 WHO (WHO, 2000b), 2003

Health Canada (Health Canada, 2003) and 2003

ESH-ESC Guidelines (ESH, 2003). Presently,

diagnosis of obesity is made mainly according to

BMI index:

2−

= mhBMI

(1)

where m-personal weight and h-personal height.

The average BMI index in normal people is

between 18.5 and 24.9 kg/m

. Obesity is defined as a

BMI>30 kg/m

; morbid obesity is when BMI>35

kg/m

. There is a continuous relationship between

gradation of BMI and health risk and between waist

circumference and health risk. Hypertension is

considered when systolic blood pressure (SBP) is

>140 mmHg and diastolic blood pressure (DBP) is

>90 mmHg. Current guidelines suggest that essential

laboratory investigation for hypertension diagnosis

should include: blood chemistry for fasting glucose,

total cholesterol, HDL-cholesterol, LDL-cholesterol,

triglycerides, urate, creatinine, sodium, potassium,

haemoglobin and haematocrit, decreased creatine

clearance, liver enzyme, the detection of an elevated

urinary excretion of albumin, an electrocardiogram

or echocardiography (Narkiewicz, 2006a, Lau et al.,

2007). For the sake of application in practice, the

input and output of the model was simplified. The

output of the model was defined as: healthy subject

(N), hypertensive with BMI normal (H), overweight

(OW), overweight with hypertension (OWH),

Variables OHC

Normal

Body mass index (kg/m

) 40-85 18.5-24.9

Waist girth

(cm) 150-170 58-88

Waist-to-hip ratio (cm/cm) 1.4-1.6 0.6-0.9

Systolic blood pressure (mmHg) 140-190 90-125

Diastolic blood pressure (mmHg) 90-120 55-84

Heart rate (bpm) 90-140 55-95

Serum triglycerides (mg/dL) 200-350 30-175

Total cholesterol

(µmol/L) 200-350 250-680

Serum HDL cholesterol (mg/dL) 35-50 50-60

Serum LDL cholesterol (mg/dL) 30-40 50-130

Glucose (mg/dL) 90-220 70-110

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

516

obesity class I (OC

), obesity class I-hypertension

(OHC

), obesity class II (OC

), obesity class II-

hypertension (OHC

), obesity class III (OC

), and

obesity class III-hypertension (OHC

2.3 Model Architecture and Training

The type of the neural network used in this work

was multilayer perceptron (MLP-NN). The network

training is based on supervised learning techniques

that were implemented and tested for a better

classification of the assessed persons. Thus, training

algorithms such as Levenberg Marquardt back

propagation (LMBP) and Generalized Delta Rule

were used.

The MLP-NN classifier architecture includes a

set of three layers. The input layer receives a set of

normalized values associated with obesity-

hypertension clinical examination data that are

delivered by the pre-processing block. The number

of input nodes was included in the 10 to 19 interval.

The hidden layer learns to encode these quantities

and includes sigmoidal neurons (logsigmoid,

tansignoid). During the MLP-NN design a practical

approach concerning the number of hidden neurons

for a short training time (t

train

) and good

classification of the individual in the obesity-

hypertension classes OHclass was performed.

The output layer produces the desired

classification results and is expressed by a number

of linear neurons n

out

(8 to 10 neurons). The values

associated with the output neurons are included in

the [0, 1]. Thus a value near one underlined that an

OHclass was identified, while ‘0’ corresponds to no

OHclass identification. The description equation of

neuronal network classifier is:

()(

outhiddenhiddenhiddenoutout

BBXWfWfY ++=

)

()

(2)

where X is the input and Y is the output, W

hidden

and

out

are the weights of the hidden and output layer

neurons, B

hidden

and B

out

are their biases, f

hidden

is a

sigmoid function for the hidden layer neurons

(logsigmoid and/or tansigmoid in the present

application) and f

out

is a linear function for the output

layer neurons.

Referring to the MLP-NN

classifier

design, different

training algorithms were applied for shorter training

times and accurate classification. Thus, fast

backpropagation algorithms expressed by gradient

descent algorithm with momentum and variable

learning rate, or Levenberg-Marquardt back

propagation algorithm were employed to update the

weights and biases of the net in the training process.

The training process requires a set of samples

expressed, in the present case, by physiopathological

variables associated with the individuals under

obesity-hypertension and corresponding known

OHclass. During training, the weights and biases in

the model are adaptively refined to ensure a relative

optimization of the network performance related to

the classification capability. As performance

measurement functions the sum-square error (SSE)

and mean-squared-error (MSE) were used:

∑

=−=

SSE

MSEnnohohSSE

(3)

where N is the number of input samples,

is the

target output imposed for known individual data and

nnoh

is the network output for given weights and

biases. The

nnoh

values in the [0, 1] interval. In

order to conclude about the obtained OHclass for a

given individual, a competitive processing block

(based on

compet() MATLAB function) transforms

the real values obtained at the MLP-NN outputs in

Boolean values corresponding to one of the

OHclasses.

After weights and biases calculation using the

above mentioned algorithms, a testing set was

employed to validate the neural network

classification capabilities using simulated and real

values.

3 RESULTS

The model of obesity-hypertension classes has been

tested using simulated data and experimental data

from a group of 30 voluntary persons. The main

characteristics of the persons included in the study

are presented in Table 2. The studied group includes

3 subjects with normal BMI, 15 persons with normal

BMI and hypertension, 3 overweight persons, 4

overweight with hypertension, 1 person with obesity

class I, 1 person with obesity class II, 2 hypertensive

subjects with obesity class II and 1 hypertensive

subject with obesity class III.

A set of data corresponding to different obesity-

hypertension classes, normally and overweight

individuals, and the corresponding physio-

pathological parameters were used to train the neural

processing scheme associated to obesity –

hypertension model.

ARTIFICIAL NEURAL NETWORK APPROACH FOR OBESITY-HYPERTENSION CLASSIFICATION

517

Table 2: The main characteristics of the patients included

in the study.

Median Average

Age (y) 58.50 (33-87) 60.37

BMI (kg/m

) 24.37 (18.57-46.88) 25.21

Waist girth (cm) 101.00 (67-131) 100.90

SBP (mmHg) 158.10 (112-189) 161.00

DBP (mmHg) 81.50 (56-101) 78.90

Heart rate (bpm) 75 (60-99) 76.67

Triglyceride (mg/dL) 171 (60-300) 163.37

Total cholesterol

(μmol/L)

178 (83-264) 174.27

HDL cholesterol (mg/dL) 44 (21-100) 46.07

Glucose (mg/dL) 84 (68-325) 106.43

The MLP-NN training set was expressed by a

10×400 input matrix and a 10×400 target matrix

while the testing set was expressed by a matrix with

the same dimensions as the training matrix. Both,

training and testing data sets were obtained by

simulation according with the values obtained in

clinical trials. Additionally, a reduced testing set for

clinical trial data expressed by 10×30 input matrix

and 8×30 output matrix was used to test the designed

processing architecture.

Different neural network architectures were

designed and tested. Considering the problem

complexity and the amount of data used for training

and testing, the MLP-NN with hidden layer

characterized by 5 to 15 neurons was employed.

Good results were obtained for n

hidden

=10 logsigmoid

neurons. The associated error curve, during the

training with SSE=0.2 training stop condition, is

shown in Fig. 2.

0 20 40 60 80 100 120 140 160 180

-1

Epoch

Sum-Squared Error

Sum-Squared Network Error for 192 Epochs

Figure 2: The error curve of the MLP-NN during the

training phase (10 logsignoid neurons and SSE=0.2

training stop condition).

The weights and biases values obtained during

the training are used to perform the normal (N),

overweight (OW) and obesity-hypertension (OH)

classification of the individuals using the simulated

testing data available for all of eight hypertension

and obesity classes. Real testing data was available

only for reduced number of classes (N, OW, OC

OHC

and OHC

). Several results concerning the

classification scheme performance are presented in

Table 3 that shows good classification accuracy for

ANN training and testing data sets obtained by

simulation. Thus, considering testing data obtained

by simulation, the output of the neural network

classification scheme is expressed by the

classification histogram (Fig. 3).

Table 3: Classification results for MLP-NN classifier

characterized by 10 logsigmoid hidden neurons.

Class

Total Classification Accuracy

Training Testing

97.5% 95%

100 % 97.5%

100% 95%

OHC

97.5% 72.5%

97.5% 97.5%

OHC

100% 82.5%

97.5% 82.5%

OHC

100% 100%

1 2 3 4 5 6 7 8

classes

Figure 3: N-OW-OH classification histogram (Nt –

number of occurrences).

The accuracy of the ANN classifier for real

clinical data associated with the diagnosis of the 30

volunteers was included in the study and the results

are represented in Table 4. As can be observed in the

table, the used real data include several of the

considered classes caused by limited number of the

data provided by the Hypertension Hospital Service.

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

518

Table 4: Accuracy of the ANN model for classification of

individuals from experimental group (30 individuals).

Class

N OW OC

OHC

Accuracy

27% 57.14% 100% 50% 0%

Considering that the experimental group includes

individuals associated with N and OW classes and

less in the obesity classes (OC1, OHC2, OHC3) the

very low or very high classification success in

several classes is expected. Better results are

expected to be obtained when an extended

experimental data for each OH classes will be used

for the designed ANN classifier.

4 CONCLUSIONS

There is a lot of knowledge on obesity, but

thoroughly view of the phenomenon remains to be

done. The model based on ANN with extended

clinical examination data represents an important

method for classification of individuals with obesity-

hypertension syndrome. A hybrid processing based

on backpropagation neural network and competitive

processing blocks was developed. Results for

simulated and experimental data recommend the

implemented processing scheme as a good classifier

and decision support tool.

Future work will be dedicated to the increase of

the classification accuracy by optimizing the neural

network architecture. Additionally, according to the

cooperation of the Hypertension Hospital unit, real

data for different subjects at different times will be

used to extract important information on

cardiovascular risk level associated with each

obesity-hypertension classe.

ACKNOWLEDGEMENTS

The authors wish to thank Drª. Monica Ferreira

(Hospital Santa Maria of Lisbon) for the support to

the research activity. The research was funded by the

Portuguese Research Foundation - FCT through

PTDC/EEA-ACR/75454/2006 research project.

REFERENCES

World Health Organization. WHO. 2000.

http://search.who.int/search?q=2025%2C+obesity&bt

nG=Search&entqr=0&output=xml_no_dtd&sort=date

%3AD%3AL%3Ad1&Search=Search&ie=utf8&client

=WHO&ud=1&site=default_collection&oe=UTF-

8&proxystylesheet=WHO

Kannel, W.B., Garrison, R.J., Dannenberg, A.L. 1993.

Secular blood pressure trends in normotensive

persons. Am Heart J, 125:1154-1158.

Tuck, M.L., Sowers, J., Dornfield, L., Kledzik, G.,

Maxwell, M. 1981. The effect of weight reduction on

blood pressure plasma rennin activity and plasma

aldosterone level I obese patients. N Eng J Med.

304:930-933.

Hall J.E., Crook, E.D., Jones, D.W., Wofford, M.R.,

Dubbert, P.M. 2002. Mechanisms of obesity-

associated cardiovascular and renal disease. Am J Med

Sci. 324:127-137.

Mansuo, K., Mikami, H., Ogihara, T., Tuck, M.L. 2000.

Weight gain-induced blood pressure elevation.

Hypertension. 35:1135-1140.

Engeli, S. Sharma, A.M. 2002. Emerging concepts in the

pathophysiology and treatment of obesity-associated

hypertension. Curr Opin Cardiol. 17:355-359.

European Society of Hypertension. Guidelines Committee.

2003. European Society of Hypertension - European

Society of Cardiology guidelines for the management

of arterial hypertension”, J Hypertens. 21:1011-1053,

http://www.eshonline.org/documents/2003_guidelines.

pdf

Narkiewicz, K. 2006a. Diagnosis and management of

hypertensionin obesity. Obesity Reviews. 7(2):155-

162.

Chalmers, J., MacMahon, S., Mancia, G. et al, 1999. 1999

World Health Organization-International Society of

Hypertension Guidelines for the management of

hypertension. J Hypertension. 17:151-183.

Sheps, S.G., Black, H.R., Cohen, J.D. et al, 1997. The

sixth report of the joint national committee on

prevention, detection, evaluation and treatment of

high blood pressure: the JNC 6 report. NIH

Publication.

Ministry of Health People’s Republic of China, China

Hypertension League, Drafting Committee for The

Guideline. 1999. Guidelines for the management of

hypertension of China (in Chinese).

Sowers, J.R., Epstein, M., Frohlich, ED. 2001. Diabetes,

hypertension, and cardiovascular disease: an update.

Hypertension. 37(4):1053-1059.

Health Canada. 2003. Canadian guidelines for body

weight classification in adults, http:/ / www.hc-

sc.gc.ca/ fn-an/ alt_formats/ hpfb-dgpsa/ pdf/ nutrition/

weight_book-livres_des_poids_e.pdf.

Lau, D.C.W., Douketis, J.D., Morrison, K.T., Hramiak,

I.M., Sharma, A.M., Ur, E. 2007. 2006 Canadian

clinical practice guidelines on the management and

prevention of obesity in adults and children. CMAJ.

176(8):S1-S10.

Ergun U. 2008. The classification of obesity disease in

logistic regression and neural network methods.

Current Cardiovascular Risk Reports. 1(2): 97-101.

Sumner, A.E., Ricks, M., Sen, S., Frempong, B.A. 2007.

How current Guidelines for obesity underestimate risk

ARTIFICIAL NEURAL NETWORK APPROACH FOR OBESITY-HYPERTENSION CLASSIFICATION

519

in certain ethnicities and overestimate risk in others.

Current Cardiovascular Risk Reports. 1(2):97-101.

Garavaglia, S.B., Synthelabo, S. 2004. Generational trends

in obesity in the United States: analysis with a wavelet

coefficient self-organizing map. Proceedings. 2004

IEEE International Joint Conference on Neural

Networks. 1:769-774.

Narkiewicz, K. 2006b. Obesity and hypertension-the issue

is more complex than we thought. Nephrol Dial

Transplant. 21(2):264-267.

Salahudeen, A.K. 2006. The obesity paradox as it relates

to survival and hypertension in dialysis patients.

Nephrol Dial Transplant. 21(6):1729-1729.

Dentali, F., Sharma, A.M., Douketis, J.D. 2005.

Management of hypertension in overweight and obese

patients: A practical guide for clinicians. Curr

Hypertens Rep. 7(5):330-336.

Messerli, F.H., Schmieder, R.E. 1986. Use of diuretic

agents in obese or black patients with systemic

hypertension. Am J Cardiol. 58(2):11A-14A.

Saravanakumar, K., Sudarsan, G.R., Cooper, G.M. 2006.

The challenges of obesity and obstetrics anaesthesia.

Curr Opin Obstet Ginec. 8(6):631-635.

World Health Organization, WHO, 2008.

http://www.who.int/dietphysicalactivity/publications/f

acts/obesity/en/

Lewis, K.K., Man, L.H. 2007. Overweight and obesity in

Massachusetts: epidemic, hype or policy opportunity?

Policy Brief, The Massachussetts Health Policy

Forum. http.//masshealthpolicyforum.bradeis.edu.

Aneja, A., El-Atat,

F., McFarlane, S.I., Sowers, J.R. 2004.

Hypertension and obesity. Rec Progr in Hormone Res.

59:169-205.

Ning, G. Su, J., Li, Y., Wang, X., Li, C., Yan, W.,Zheng

X. 2006. Artificial neural network based model for

cardiovascular risk stratification in hypertension. Med

Biol Eng Comput. 44:202-208.

Bidiwala, S., Pittman, T. 2004. Neural network

classification of pediatric posterior fossa tumors using

clinical and imaging data. Pediatr Neurosurg. 40(1):8-

15.

Lapuerta, P., Azen, S.P., LaBree, l. 1995. Use of neuronal

network in predicting the risk of coronary artery

disease. Comp Biomed Res. 28:38-52.

Mangiameli, p., West, D., Rampal, R. 2004. Model

selection for medical diagnosis decision support

systems. Decision support Syst. 36(3):247-259.

Orunescu, E., Bagnasco, M., Salmaso, C. Altrinetti, V.,

Bernasconi, D, DelMonte, P., Pesce, G., arugo, m.,

Mela, G.S. 2004. Use of an artificial neural network

to predict Graves’ disease outcome within 2 years of

drug withdrawal. Eur J Clin Invest 34(3):210-217.

Poli, R., Cagnoni, S., Livi, R., Coppi, G., Vali, G. 1991. A

neural network expert system for diagnosing and

treatinghypertension. Computer 24(3):64-71.

Mangiameli, P. West, D., Rampal, R. 2004. Model

selection for medical diagnosis decision support

systems. Decision Support Syst. 36(3):247-259.

Ai, M., Tanaka, A., Ogita, K., Sekine, M., Numano, F.,

Numano, F., Reaven, G.M. 2000. Relationship

between hyperinsulinemia and remnant lipoprotein

concentrations in patients with impaired glucose

tolerance. J Clin Endocrinol Metab. 85(10):3557-

3560.

Aguilera, C.M., Gil-Campos, M., Cañete, R., Gil, A. 2008.

Alterations in plasma and tissue lipids associated with

obesity and metabolic syndrome. Clin Sci (Lond).

114(3):183-193.

Gupta, R., Rastogi, P., Sarna, M., Gupta, V.P., Sharma,

S.K., Kothari, K. 2007. Body-mass index, waist-size,

waist-hip ratio and cardiovascular risk factors in urban

subjects. J Assoc Physicians India. 55: 621-627.

World Health Organization. WHO. 2000b. Obesity:

Preventing and Managing the Global Epidemic. WHO

Obesity Technical Report Series No. 894 World

Health Organization, Geneva, Switzerland.

BIOSIGNALS 2009 - International Conference on Bio-inspired Systems and Signal Processing

520