INTELLIGENT SYSTEM FOR DIFFERENTIAL DIAGNOSIS AND MONITORING …€¦ · INTELLIGENT SYSTEM FOR DIFFERENTIAL DIAGNOSIS AND MONITORING OF PATIENTS AFTER CARBON MONOXIDE POISONING Gulchin

European Journal of Research in Medical Sciences Vol. 3 No. 2, 2015 ISSN 2056-600X

Progressive Academic Publishing, UK Page 13 www.idpublications.org

INTELLIGENT SYSTEM FOR DIFFERENTIAL DIAGNOSIS AND MONITORING OF

PATIENTS AFTER CARBON MONOXIDE POISONING

Gulchin G. Abdullayeva

Institute of Control Systems of Azerbaijan National Academy of Sciences, AZERBAIJAN

Nazaket H. Qurbanova

Azerbaijan Medical University, AZERBAIJAN

Irada H. Mirzazadeh

Institute of Mathematics and Mechanics of Azerbaijan National Academy of Sciences, AZERBAIJAN

&

Ulkar R. Naghizade

Practice for Anesthesiology in Esslingen am Neckar, GERMANY

ABSTRACT

According to statistical data, with the development of oil, chemical, gas industries cases of

poisoning caused by toxic substances employed in these branches have become more frequent

recently. A special place among them is occupied by carbon monoxide, poisoning with which has

been growing steadily. This research deals with poisonings caused by carbon monoxide and

chemical substances which are clinically close to carbon monoxide in pre-laboratory situation and

this calls for conducting differential diagnosis. Considering such consequences of similar-

poisonings as myocardial infarction, Parkinson's disease u.a. it is expedient to perform monitoring

of a patient after staying in a stationary hospital which determines optimum time of its performance,

kind and the number of analyses required for developing an intelligent system. This paper proposes

an elaboration of an intelligent information system for differential diagnosis and monitoring in

cases of poisonings with toxic substances using carbon monoxide as an example.

Keywords: Carbon monoxide, differential diagnosis, monitoring, biostatistical methods, intelligent

system.

INTRODUCTION

There exists a certain group of problems in medicine which demand the accuracy of diagnosis and

quickness of the first aid. Poisonings with toxic substances relate to a group of similar problems

where the solution and positive outcome strongly depend on time. Under conditions of speedy and

urgent aid the solution of this problem is significantly complicated when a patient is in a comatose

state. In compliance with statistical data, due to the development of oil, chemical and gas industries

cases of poisonings with toxic substances used in the mentioned branches have become more

numerous recently. It should be particularly emphasized that the number of cafes of carbon

monoxide poisonings has been constantly on the rise. Carbon monoxide or carbon oxide is formed

everywhere if there are conditions of incomplete combustion of substances containing CO. This is a

perfidious enough gas -it has no colour, no taste and almost no smell. Easily penetrating through

lungs into blood it interacts with hemoglobin forming carboxyhemoglobin (HbCO) and blocks the

transfer of oxygen to tissue cells which leads to hypoxia. It is enough to look at the data of 2004

which we have consolidated according to territorial principle (Table 1), in order to assess the

severity of the problem under consideration [1, 2, 3]



Table 1: Carbon monoxide poisoning statistics for 2004

Regions Number of poisonings

North America 179533

Central America 8296

Caribbean basin 1624

South America 125408

Northern Europe 8293

Western Europe 41871

Central Europe 64296

Eastern Europe 87195

South Western Europe 21168

Southern Europe 24636

South Eastern Europe 21325

Middle Asia 1146

Central Asia 20234

East Asia 636315

South West Asia 33938

South Asia 590084

South East Asia 211325

Middle East 75295

North Africa 50372

East Africa 72949

South Africa 30387

Oceania 12219

This problem did not pass by Azerbaijan either Table 2 demonstrates the number of people who

suffered from carbon monoxide in the city of Baku throughout 2006-2014

Table 2: Consolidated Table of carbon monoxide poisonings in the city of Baku

№ Districts of Baku 2006 2007 2008 2009 2010 2011 2012 2013 2014

1 Narimanov 41 38 38 48 69 121 127 118 105

2 Khatai 77 159 154 82 106 135 192 126 161

3 Sabayil 47 26 36 41 42 85 109 88 139

4 Yasamal 57 64 88 483 118 137 151 132 76

5 Nasimi 20 103 186 122 129 221 237 178 187

6 Nizami - 40 63 54 64 123 171 200 197

7 Binagady 53 70 217 129 190 316 395 378 371

8 Khazar - - 9 9 17 26 36 63 78

9 Surakhany 34 33 59 62 70 141 141 193 185

10 Sabunchi 30 43 111 72 107 147 221 202 158

11 Karadag 42 83 83 86 98 115 232 145 206

12 Total: 401 659 1044 788 1010 1567 2012 1823 1863



LITERATURE REVIEW

While it is quite easy to diagnose carbon monoxide poisoning unmistakably in everyday life this

procedure may become just as complicated in production conditions where a large quantity of toxic

substances are present simultaneously and among the said substances there are ones causing almost

similar symptoms of affection [4,5]. This fact is particularly undesirable when the first and urgent

aid are needed for which time and absence of laboratory data are the main factors. At the pre-

hospital stage, it is important to eliminate generated pathological syndromes and to take measures

for active detoxication of organism. If one manages to reveal a poisoning substance, exactly it

becomes vital to apply antidote therapy (removal of a poison from organism) which is specific for

each toxin [6].

Brain and heart are particularly vulnerable to the action of this poison. Myocardium binds CO

stronger than skeletal muscles which results in a serious lack of oxygen and symptoms of

stenocardia, arrhythmia as well as in cell death markers [7]. People who were through carbon

monoxide poisoning may die of heart attack within some nearest years because of the damage,

which this poisonous substance had caused their cardiac muscle. These are the conclusions made by

researchers from the Heart Institute of Minneapolis who had studied ambulatory cards of patients

undergoing treatment for carbon monoxide poisoning of different degree of severity. In accordance

with the scientists’ data, 37 % of the patients poisoned with carbon monoxide suffered from cardiac

muscle injuries. About one-fourth of them died within 7 years after the poisoning event. In

conformity with professor Timothy Henry’s words, the number of patients in whom cardiac

disturbances due to poisoning had been revealed , surpassed the boldest expectations of the

scientists a great deal (this information is taken from the “Report” on a research published in the

latest issue of Journal of the American Medical Association of 2005). Carbon monoxide can cause

detrimental damage to brain and central nervous system, lead to loss of hearing, eyesight

disturbances, Parkinson’s disease u.a. [10].

From the above said it should be concluded that in cases of toxic substance poisonings it is

obligatory to conduct differential diagnosis and to organize monitoring for subsequent observation

of a patient’s state. Modern information technologies, methods of artificial intelligence and medical

statistics could create tools in the form of software for solving such sophisticated problem. This

paper proposes an elaboration of one intelligent information system performing both differential

diagnosis of poisoning with 15 toxic substance and subsequent monitoring in addition.

SOLUTION

We have studied cases of carbon monoxide poisonings in Baku throughout 2006-20014 and

additionally revealed 14 toxic substances having similar primary clinical pattern witnessed by the

first and emergency aid service before laboratory researches.

.



Fig. 1. Relevancy of CO clinical pattern with 14 toxic substances

The following notations and symbols are used in Fig. 1:

Toxic substances: 1-aniline, 2-atropine, 3-barbiturates, 4-dichlorethane, 5-codein, 6-

pachycarpine, 7-tubaside, 8-phosphororganic compounds, 9-ethyl alcohol, 10-ethylene

glycol, 11-CO-carbon monoxide, 12-tranquilizers, 13-antihistaminic agents, 14-salicylates,

15-cyanides, respectively.

Latin letters stand for probably detectable symptoms: a-miosis; b-mydriasis; c-play of

pupils; d-synchronous myofibrillations; e-asynchronuos myofibrillations; f-hyperkinesis of

choreoid type; g-rigidity of oceiput muscles; h-asynchronous convulsions; i-epileptiform

convulsive status; j-perspiration of skin; k-drastic cyanosis of skin; l-hyperemia of skin; m-

“marble” appearance of skin; n-bradycardia; o-tachycardia; p-respiratory paralysis with

retained reflexes, q-respiratory paralysis only against the background of areflexia; r-

bronchoreia.

Fig. 1 demonstrates informationally significant symptoms for carbon monoxide drawn in

heavy lines.

We shall display the general structure of an intelligent information system for differential diagnosis

developed in [8, 9] to which monitoring module of carbon monoxide poisonings is added.

CO

3,4,7,9,10,13-

15 1,2,6,13-15

9,12,15

a

b c

d

r

q

p

o

n

m

4,7,10,12-

15

4,10,12-15

12 - 15

1,3-7,9,10,12-

15

е

i j

1,3-

7,9,10,13,15

k

5,6,8,10,13-

15 5, 6

3,6,7-

10,12,13,15

3, 6, 7,9,10, 12-

15

g

f

h

1,8,10,12-

15

10

1,10,15

1,2,3,5,7-10,12-

15

3-5,7-9,12-14

2-10,12-15

l

1,3-7,9,10,12-15



Fig. 2. Structure of the intelligent information system

Fig. 2 shows the scheme of architecture of intelligent information system for differential diagnosis

and monitoring, where

- iB medical teams for fist and urgent aid;

- organization of database (DB) of clinical symptoms of toxic substances;

- generation of knowledge base (KB) on the strength of production rules and frame

representation;

- decision maker which makes decisions on the basis of neuronal network;

- antidote therapy discontinuing or weakening the action of poison on organism. The choice

of antidote is determined by the type and nature of action of substance which caused poisoning, the

effectiveness of use depends on the accuracy of revealing a substance that caused poisoning and

also on the quickness of giving aid;

- generation of electronic health card;

- monitoring block comprising time series methods, modern biostatistical methods,

correlation analysis, regression analysis.

A base of primary symptoms before laboratory clinical pattern of 15 substances under consideration

has been developed (see Table 3)

Call Fi

rst

and

urg

ent

aid

B1,

B2

…, B

n Clinical

symptoms

симптомы

Decision

maker

Recommendations for

urgent aid

(Antidote therapy)

EHC

Exit

Data-

base

Know-ledge base

Ambulatory

Stationary hospital

Monitoring

Time series modern

biostatisticat

methods



Table 3: Differential diagnosis

Clinical

symptoms

An

ilin

Atr

opin

Bar

bit

ura

tes

Dic

hlo

reth

ane

Co

dei

n

Pac

hy

carp

in

Tu

bas

ide

Fo

sp.o

rq.s

ubst

an..

..//

’ic

mad

d.

Eth

yl

alco

hol

Eth

yle

ne

qly

col

Car

bo

n m

on

ox

id

Tra

nq

uil

izer

s

An

tihis

tam

ines

.

Sal

icy

late

s

Cy

anid

es

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Miosis 0 0 ± ± + 0 ± + ± ± ± + ± ± ±

Mydriasie + + ± ± 0 + ± 0 ± ± + ± + + +

“Play of pupils” 0 0 + ± 0 0 ± 0 + ± ± ± ± ± ±

Synchronous

myofibrillation

0 0 0 ± 0 0 0 0 + ± ± ± ± ± ±

Asynchronous

myofibrillation

0 0 0 0 0 + 0 + 0 0 ± ± ± ± ±

Hyperkinesis of

choreoid tipe

+ + 0 0 0 0 0 + ± 0 ± ± + + ±

Riqidity of oc-

ciput muscles

0 0 0 ± 0 0 ± ± ± + + ± ± ± ±

Asynchronous

convulsions

+ ± 0 ± ± ± 0 ± ± + + ± ± ± +

Epilept. con-

vulsive status

± 0 0 0 0 0 + ± 0 ± ± ± ± ± ±

Perspiration of

skin

± 0 ± ± ± ± + ± ± ± ± ± ± ±

Dryness of skin ± + ± ± ± ± ± 0 ± ± ± + ± + ±

Drastic syano-

sis of skin

+ ± ± ± ± ± ± ± ± ± ± ± ± ± ±

Hyper.of skin 0 + ± ± ± 0 ± ± ± + ± ± ± ± +

«Marble» ap-

pearanse of skin

± ± ± + ± + ± ± ± ± ± ± ± ± ±

Bradycardia 0 0 ± 0 0 ± ± + ± ± ± ± ± ± ±

Taxchycardia + + ± + + ± ± ± ± ± ± ± ± + ±

Resp.paralysis

with retained

reflexes

± ± ± ± + + 0 ± 0 ± + ± ± ± ±

Respiratory

paralysis only

against the

backqround of

areflexia

+ + + + ± ± + ± + ± ± + ± ± ±

Bronchoreia ± 0 ± ± ± ± ± + ± ± ± ± ± ± ±



The following notations are taken for Table 3: «+» required presence of symptom, «0»-absence of

this symptom, «+»-situation in which the symptom may be present with minimum action or may

not be observed at all, i.e. they do not dominate when diagnosis is made. These data permitted to

develop a database functioning according to the principle of production rules of the type:

If “prerequisites”-Then “actions”

15,...,2,1;19,...,2,1

,...,,,...,,

,

1121

ji

yyyyyxthen

yxXxif

mjji

jii

the first step-rigorous differentiation

15119,...,2,1

ji

yxthen

YyXxif

ji

ji

the second step-non-rigorous differentiation

15119,...,2,1

ji

yxyxthen

Xxif

jiji

i

the third step-indeterminate differentiation

A two-layered model of neuronal network with 38 entries and 15 exits has been implemented

for differential diagnosis [8, 9].

Fig. 3. Model of neuronal network for differential diagnosis of poisonings.

x1

●

x2

●

x3

●

x4

●

X37

●

X38

●

1

2

3

15

y1

y2

y3

y15



Why are there 38 entries on the layer? We think that for each parameter the presence of two neurons

is advisable. One of them is activated when an indicator-parameter is revealed, the second neuron is

activated when it is absent. The exits are 15 hypotheses, i.e. causes of poisoning. The second layer

of the network can be represented as a sum

38

,...,,

38

,...,,

38

,...,,

/

lkmi psti vrji

yh

iiii xxxy (1)

If the first two sums in (1) are taken as a zero state confirming the existence of hypothesis the third

sum will be considered to be a set corroborating the hypothesis with all kinds of variations. In this

case the third sum performs two functions: all variants of which are different than zero confirm

the hypothesis and the set itself is a set of neuronal network teaching. For example, expression

(1) will be like this for carbon monoxide:

38

33,15,13,3

38

37,35,31,29,27,25,23,21,19,17,11,9,7,5,1

/

i i

yh

iii xxy (2)

When the system performs successfully a poisoning substance is found out and antidote therapy is

made in time after which, if necessary, a patient is taken to a stationary hospital where he has

treatment. As mentioned above, after a stationary hospital it is required to conduct monitoring for a

patient regardless of the degree of affection.

Under monitoring in case of poisonings we shall imply a system of collection, storage and analysis

of a small amount of required parameters and their indicators for making current diagnosis and

further prognosis on the patient's state of health on the whole. The result of parameter monitoring is

a body of measured values of parameters obtained on continuously adjacent to one another time

intervals during which the values of the parameters do not change appreciably. A principal

distinction of current state monitoring from that of parameters is the presence of an interpreter for

measured parameters in the terms of state-an expert system of supporting decisions on a patient's

state after a certain interval of time.

Monitoring performs several organizational functions:

it reveals the state of critical or being in the process of change conditions in the status of a

patient for whom a plan of future measures will be worked out;

it provides data on the previous state giving feedback will be worked out; relating to earlier

successes and failures of a definite policy or programs;

it checks on the conformity with regulations and contractual obligations;

A need in monitoring in a stated problem is determined by a doctor and it depends on the degree of

poisoning. Periods may vary in the range of a week, month, quarter, six months, year. To organize

monitoring we shall add a module to the intelligent system for differential diagnosis (see Fig. 2).

Monitoring will be carried out by the following mathematical methods:

1. Time series or dynamics series-is a statistical material on the significance of some

parameters (of one parameter in the simplest case) of a process being studied which is

collected in different moments of time. Each unit of a statistic material is called

measurement or readout. For each readout the time of measuring or number of measurement

in succession must be given in time series.

2. Time series analysis. Time series analysis presents a body of mathematico-statistical

methods of analysis intended for revcaling the time series structure or for their prediction.



Detailed discussion of these methods can be found in the following researches: Anderson

(1976), Bocks and Jenkins (1976), Kendall (1984), Kendall and Ord (1990),Montgomery,

Johnson, and Gardiner (1990), Pankratz (1983), Shumway (1988), Vandaele (1983), Walker

(1991), Wei (1989).

Two main objects of time series analysis exist:

determination of series nature;

forecasting (prediction of future values of time series from present and previous values).

Prediction of future values of time series is used for effective decision-making. Correction of the

obtained prediction is made for the sake of refinement of the obtained long-term forecasts with

consideration for the influence of seasonal or spasmodic character of development of a phenomenon

under study. For time series analysis we have employed parametrical and non-parametrical methods

of mathematical statistics from which we shall name Fisher’s F-criterion for comparing two or more

totalities (as, for example, in analysis of variance; Kraskell-Wallace’s criterion which is a non-

parametrical alternative for one-dimensional (intergroup) analysis of variance. It is employed for

comparison of two or more retrievals and checks null hypotheses according to which various

retrievals were taken from one and the same distribution or from distributions with the same

medians (see Siegel & Castellan, 1988); non-parametrical Wilcockson’s criterion is an enhancement

of two-retrieval Wilcockson’s rank sums criterion; Friedman’s criterion is a non-parametrical

analogue of analysis of variance of repeated measurements, it is used for analysis of repeated

measurements relating to one and the same individual. With the help of Friedman’s criterion we

check null hypothesis that diverse methods of treatment give practically the same results.

E.g., Fig. 4 demonstrates a window of menu choice for statistical analysis, Fig. 5 displays individual

fragments of protein study in a patient A. who was being observed between 2012 and 2014 (36

measurements in all). All computations and analysis are performed in СТАТИСТИКА

(STATİSTİCS) package.

Fig. 4. Choise of methods of time series analysis



Fig. 5. Fragments of time series analysis for protein during 3 years under normal distribution

Further a graph of situation analysis at exponential smoothing without consideration for trend and

seasonal constituent is given in Fig. 6 (for smoothing constituent 𝛼 = 0,1; 𝛼 = 0,5; 𝛼 = 0,9).

7

1 2

3 4

5 6

8



Fig. 6. Data analysis at exponential smoothing

A blue line in Fig. 6 indicates the true level of a signal itself, a red line shows values of deviations

(residuals) of the true levels from the smoothed ones, a green line is used for the smoothed values

(smoothed series). Analysis of the obtained results has demonstrated the following: the less is a

smoothing constant value, the less smoothed values vary. At a small value of 1,0 smoothed

values differ a great deal from the true levels of the initial time series. In the general case the

smoothing at a small responds to such fluctuations or turning points weakly. But when a constant

is equal to 9 we get a much lesser smoothing effect, smoothed values, however, follow the true

values through the whole length of the initial time series to a greater extent. A constant 5,0

gives an intermediate effect between the first two variants. I.e., when time series contains an

unimportant irregular component it is advisable to use big constants.

The produced example has demonstrated that the state of the patient A. did not become stable (in

relation to protein level) and further treatment and monitoring are needed. The developed software

has been successfully tested on real cards of patients throughout 2006-2014 at the Central Station of

the First and Emergency Medical Aid of the city of Baku.

RESULTS

Finally, the following conclusions can be made:



- cases of carbon monoxide poisonings have been investigated, pre-laboratory signs of poisoning

are singed out which are in many respects similar to poisonings with some chemical substances

having toxic properties;

- the above said is a justification for conducting differential diagnosis;

- consequences of carbon monoxide poisoning are investigated and cases of myocardial infarctions,

damage of central nervous system etc. are shown which is a justification for organizing monitoring

in subsequent period;

- multi-module architecture of the intelligent information system is proposed-the system is based on

modern information technologies, methods of artificial intelligence and biostatistics;

- an algorithm of teaching neuronal network for differential diagnosis is given;

- an example of investigating one parameter by means of methods of time series analysis is

provided;

- the developed system was tested in a medical institution.

REFERENCE

1. Iqbal S, Clower JH, Boehmer TK, Yip FY, Garbe P. Carbon monoxide-related

hospitalizations in the US: evaluation of a web-based query system for public health

surveillance.// Public Health Rep 2010;125:423—32

2. Clower JH, Hampson NB, Iqbal S, Yip FY. Recipients of hyperbaric oxygen treatment for

carbon monoxide poisoning and exposure circumstances. J Emerg Med. In press 2011.

3. US Census Bureau, International Data Base, 2004

4. Bova A.A., Gorokhov S.S. War Toxicology and Toxicology of Extreme Situations. Minsk.

BSMU, 2005.

5. Spirighins L., Chambers J. Emergency Medicine. Diagnosis and Treatment of Urgent States.

M., Medical Literature, 2006.

6. Abdullayeva G.G. et. al. Conseption of Constructing an Emergency System for Differential

Diagnosis of Internal Diseases // News of NAS of Azerbaijan, volume XXII, No. s. 2-3,

2002.

7. Robin F.Hurley, Ramona O.Hopkins, Erin D. Bigler, Catherine H.Taber. Use of Functional

Diagnosis in Carbon Monoxide Poisonings // Journal of Neuropsychology and Clinical

Neuroscience-2001; 13:. 157-160.

8. Abdullayeva G.G., Kurbanova N.G., Mirzazadeh I.H. Intelligent Information System for

Differential Diagnosis of Poisonings with Toxic Substances // Information Technologies.

M.: 2013, No 10 (206)

9. Abdullayeva G.G., Mirzazadeh I.H., Naghiyev R.H., Naghizade U.R. Intelligent Information

System of Diagnosis and Monitoring Application in the emergency medical assistance for

Poisonings by Toxic Substances// Journal of Health and Medical Informatics (JHMI):2013,

4 Volume 2 Issue, 7 pp.

10. World Health Organization. Environmental Health Criteria 213 - Carbon Monoxide (Second

Edition). WHO, Geneva, 1999; ISBN 92 4 157213 2 (NLM classification: QV 662). ISSN

0250-863X.

Documents

INTELLIGENT SYSTEM FOR DIFFERENTIAL DIAGNOSIS AND MONITORING …€¦ · INTELLIGENT SYSTEM FOR DIFFERENTIAL DIAGNOSIS AND MONITORING OF PATIENTS AFTER CARBON MONOXIDE POISONING Gulchin