intelligent system for differential diagnosis and monitoring …€¦ · intelligent system for...
TRANSCRIPT
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]
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 14 www.idpublications.org
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
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 15 www.idpublications.org
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.
.
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 16 www.idpublications.org
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
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 17 www.idpublications.org
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
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 18 www.idpublications.org
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 ± ± ± ± ± + ± ± ± ± ± ± ±
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 19 www.idpublications.org
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
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 20 www.idpublications.org
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.
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 21 www.idpublications.org
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
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 22 www.idpublications.org
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
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 23 www.idpublications.org
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:
European Journal of Research in Medical Sciences Vol. 3 No. 1, 2015 ISSN 2056-600X
Progressive Academic Publishing, UK Page 24 www.idpublications.org
- 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.