Docstoc

Diagnosis of Heart Disease based on Ant Colony Algorithm

Document Sample
Diagnosis of Heart Disease based on Ant Colony Algorithm Powered By Docstoc
					                                                            (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                               Vol. 11, No. 5, May 2013


           Diagnosis of Heart Disease based on Ant Colony Algorithm
                                                 Fawziya Mahmood Ramo
                                                  Computer Science Department
                                          College of Computer Science and Mathematics
                                                       Mosul University
                                                         Mosul, Iraq


Abstract - The use of artificial intelligence method in medical
analysis is increasing, this is mainly because the effectiveness                          II.   RELATED WORKS
of classification and detection systems has improved in a great           Sengur A. and Ibrahim T. in 2008 designed artificial
deal to help medical experts in diagnosing. In this paper, we           immune system and fuzzy K-NN algorithm to determine the
investigate the performance of an Heart disease diagnosis is a          heart value disorders from the Doppler heart sounds. The
complicated process and requires high level of expertise, the           proposed system is a better clinical application a specially for
work include a novel method for diagnosing eight heart                  earlier survey of population [6 ].
disease (Atrial Fibrillation, Ventricle Strikes, Bigemeny,                    Ramteke R. and Manza R. in 2010 provided expert
Ventricular Tanchycardia, Ventricular fibrillation, Third               system for diagnosing of heart disease using support vector
Degree Heart Block, R on T phenomenon and normal) using                 machine and feed forword backpropagation technioque gives
Ant Colony System(ACS) based on ECG (Electrocardiogram),                less appropriate result for medical persecription for heart
blood oxygen and blood pressure . The experiment show that              disease patient[7].
the proposed method achieves high performance with a heart                  Usha Rani in 2011 analyzed heart disease data set by
diseases classification accuracy of 92.5%.                              using Neural Network approach to increase the efficiency of
                                                                        the classification process parallel approach is also adopted in
                      I.   INTRODUCTION                                 the training phase [8].
                                                                           Jyothi Singaraju and Vanisree in 2011 decision support
       The use of computer in medical applications has                  system has been proposed for diagnosis of congenital heart
 increased dramatically. Computerized image processing                  disease, the system designed by using MATLAB GUI feature
 techniques have been used to improve the picture quality,              with the implementation of back propogation [9].
 images can be analyzed to highlight areas of interest or to               Sameh Ghwanmeh in 2012 provided a decision support
 extract meaningful diagnostic features that can provide                system to classify the heart disease mitral stenosis,aortic
 objective evidence to aid the human decision making                    stenosis and ventricular septal defect. Series of experiment
 process[1].                                                            have been conducted using real medical data to test the
          artificial intelligent technique (i.e., fuzzy logic,          performance and accuracy [10].
 neural networks, genetic algorithms, Ant Colony algorithm
 and expert systems) has particular computational properties                               III-Medical Background
 that make it suited for a particular type of problems,                    In emergency departments and intensive care doctor needs
 there are great advantages in their synergistic utilization            to monitor continuous and intensive follow-up of a number of
 [2][3].                                                                variables and the patient's vital signs are in fact many and
   Today there is a synergy beginning to form among                     varied and differ from satisfactory state to another is the most
 neural networks, ant algorithm               and      genetic          important of these variables[11]:
 algorithms.This synergy has been variously called Soft                 A - The average number of heart Pulse Rate per minute (60-
 Computing[4].Soft Computing is an area of computing                    100 beats per minute for a person of normal).
 allowing imprecision, uncertainty and partial truth to                 B- The average number of times breathing Respiratory Rate
 process and therefore achieves robustness and low solution                per minute (10-15times per minute for a person of normal).
 cost. Hybrid Soft Computing approaches incorporates all                C-Arterial Blood Pressure and is divided into:
 the features from individual fields and, moreover, has the                      1-systolic arterial blood pressure (120-139 mm Hg
 ability to overcome difficulties and limitations that                           for normal human).
 characterize each field. The use of intelligent hybrid                          2-Diastolic Hypertension (Diastolic Blood Pressure)
 systems is growing rapidly with successful applications                         (80-89 mm Hg for normal human).
 in     many areas including process control, robotics,
 manufacturing, medical diagnosis, etc. [4][5].                         D- The level of arterial blood oxygen saturation (95-100%),
                                                                           must not be less than 90% in normal human).




                                                                   77                              http://sites.google.com/site/ijcsis/
                                                                                                   ISSN 1947-5500
                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                  Vol. 11, No. 5, May 2013

  In this work eight type of heart diseases has been                       is called loading matrix and incorporates the orthogonal
diagnosis[12][13][14]:-                                                    vectors P called as loading or principal vectors, which are, in
1-normal :The case where a normal ECG signal and the                       fact, eigenvectors associated with eigen values of the
normal rate of blood pressure(120high,80 low) and blood                    covariance or correlation matrix of X. T is called score matrix,
oxygen (95%)                                                               which is the projection of the original data[16].
2-Atrial Fibrillation: This situation occurs due to the presence
of more than a location within the atria produces electrical
impulses lead to twitter atrium not shrink An bassath natural                             V- Artificial Ant Colony System
and oxygen percentage (90%) and blood pressure                                  An artificial Ant Colony System (ACS) is an agent-based
(156high,95low)                                                            system which simulates the natural behavior of ants and
3-Ventricular strikes: deceased where this situation occurs                develops mechanisms of cooperation and learning. ACS was
when there is a site in one of the ventricles generate electrical          proposed by Dorigo et al. (1999) as a new heuristic to solve
impulses lead to a contraction in the ventricles outside the               combinatorial-optimization problems. This new heuristic
natural harmony, oxygen percentage (92%) and blood pressure                called Ant Colony Optimization (ACO) has been shown to be
(145high,90low)                                                            both robust and versatile – in the sense that it can be applied to
4- R on T phenomenon: Occurrence of ventricular stroke                     a range of different combinatorial optimization problems[17].
accompanied by the phenomenon of interference between the
wave of ventricular blow migrans and natural wave to blow                    The Ant Colony algorithm idea is summarized in the
her previous, oxygen percentage (93%) and blood pressure                   following pseudo code [17][18]:-
(130high,85low)                                                            Set parameters, initialize pheromone trails
5-Bigemeny:Succession occurs between natural strikes and                   while termination condition not met do
migrans ventricular strikes where each pulse followed by                   ConstructAntSolutions
natural a ventricular strike blow and then a normal pulse,                 ApplyLocalSearch (optional)
oxygen percentage (92%) and blood pressure (125high,80low)                 UpdatePheromones
6-Ventricular Tachycardia: Shrink the ventricles in response to            Endwhile
electrical impulses generated from the point of the one
controlled by the pulses generated by this point completely on                 The most interesting contribution of ACS               is the
the ventricles, leading to an acceleration in the heart                    introduction of a local pheromone update in addition to the
characterized this case that the heart be Regular but faster than          pheromone update performed at the end of the construction
the natural, oxygen percentage (85%) and blood pressure                    process. The local pheromone update is performed by all the
(90high,50low)                                                             ants after each construction step. Each ant applies it only to the
7- Ventricular Fibrillation, there is more than one point in the           last class traversed [17][19]
ventricles produce electrical impulses without any tune, the
ventricles stops extroversion does not pump blood from the                        τ       1      φ .τ    φ. τ … … …    2
heart to the main arteries, oxygen percentage (60%) and blood
pressure (0high, 0low)                                                         where ϕ        [0- 1] is the pheromone decay coefficient, and
8- Third Degree Heart Block: interrupted transmission of                   τ is the initial value of the pheromone. The main goal of the
electrical impulses completely between the atria and the                   local update is to diversify the search performed by
ventricles at the atrioventricular node (AV Node), oxygen                  subsequent ants during an iteration by decreasing the
percentage (87%) and blood pressure (90high,60low)                         pheromone concentration on the traversed classs, ants
                                                                           encourage subsequent ants to choose other classs and, hence,
                     IV- Feature Extraction                                to produce different solutions. This makes it less likely that
     The goal of the feature extraction is to extract feature              several ants produce identical solutions during one
from these patterns for reliable intelligent classification[15]. In        iteration[17][19].
this paper to extract the characteristics of ECG using eigen
value matrix is among the most popular methods for extracting              τ          1   p .τ       p. ∆τ    ……(3)
information from raw measured data. It can handle high-
dimensional and correlated data by projecting the data onto a              Where ∆τ              the best solution otherwise zero.
lower dimensional subspace which contains most of the
variance of the original data, the optimal linear transformation                              VI- The proposed approach
of the original data matrix X to determine the minimum                         The proposed system is composed of following stages :-
number of uncorrelated variables that will account for the                       1- Data Configure : create database for eight heart
maximum underlying variance in the data via[16]:                                      disease include medical information about diseases (
T=X P or X=TP T …..(1)                                                                ECG , blood oxygen and blood pressure). As show in
                  
where X            indicates a matrix of n observations and p                         table(1)
variables, measured about their means P=[P P … P ]  R    




                                                                      78                                     http://sites.google.com/site/ijcsis/
                                                                                                             ISSN 1947-5500
                                                              (IJCSIS) Interna
                                                              (                              of            ence and Information Security,
                                                                             ational Journal o Computer Scie
                                                                                                                          o.
                                                                                                                Vol. 11, No 5, May 2013


                                                                             sitivity and spe
                                                                       1. Sens              ecificity
                                 o               isease
           Table (1) : Data base of eight heart di                         To test the perfo                             lassification,
                                                                                            formance of a d i s e a s e cl
                                                                             alues will be u
                                                                        six va                            egative rate (FN
                                                                                             used: False Ne              N%), False
  D
  Disease name     CG
                  EC image     Oxy
                                 ygen    High        Low
                                                                              ve rate (FP%), sensitivity, specific
                                                                        Positiv                                          city,Positive
                                Ra
                                 ate     blood      blood               predic
                                                                             ctive value (P  PP%) and N  Negative predi  ictive value
                                        pressure   pressure             (NP%%). The perfo                 ted
                                                                                            ormance is test on a datab   base of 80
  Normal
  N
                                 95
                                  5       120        80
                                                                                            e
                                                                        cases. The database contains featu               e.
                                                                                                          ures of disease Cases are
                                                                              fied as normal/ negative (
                                                                        classif                                          se/
                                                                                                          (N) or diseas positive
                                                                             0].             vity
                                                                        (P)[20 For sensitiv and specif                   ,
                                                                                                          ficity analysis, we use the
  A
  Atrial
  Fibrillation
  F                              90
                                  0       156        95                      wing expression
                                                                        follow               ns[6]:
                                                                                      PP
                                                                              vity=
                                                                       Sensitiv              ……(6)
                                                                                        PP FN
  Ventricular
  V                                                                                       NP
  strikes                        92
                                  2       145        90
                                                                             icity=
                                                                       Specifi           ………(7)
                                                                                         …
                                                                                   NP FP
                                                                             (2)         proposed system performance
                                                                       Table ( shows the p             m           e.
  R    on   T                                                                                                     proposed system
                                                                                          Table(2) performance of p
  phenomenon
  p                              93
                                  3       130        85
                                                                                                itivity
                                                                                            Sensi                       91%
                                                                                                ificity
                                                                                            Speci                       100%
  B
  Bigemeny
                                 92
                                  2       125        80
                                                                                                  N%
                                                                                                 FN                     6%
                                                                                                  P%
                                                                                                 FP                     0%
  V
  Ventricular                     5
                                 85        90        50
  T
  Tachycardia                                                                                     P%
                                                                                                 PP                     64%
                                                                                                  P%
                                                                                                 NP                     10%
  V
  Ventricular                    60
                                  0        0          0
  Fibrillation
  F

  Third Degree
  T                                                                                         uracy
                                                                       2. Classification accu
  H
  Heart Block                    87
                                  7        90        60                   The classification accuracy is a common me       ethod that is
                                                                             n              ecognition appl
                                                                       used in the pattern re               lications. The classification
                                                                              cy
                                                                       accurac for the expe                 n             of
                                                                                            eriment is taken as the ratio o the number
  Features extract
2-F              tion                                                  of sam                ly
                                                                             mples correctl classified to the total number of
                 res          G            gen    trix
    Extract Featur from ECG by using eig values mat                    samples[6]. Table (3) shows the per rformance para ameters.
afte segmentation process to ge disease pulse.
   er            n            et            e
                                                                              Table (3) obtaine performance parameters of proposed system.
                                                                              T               ed
                                                                             sease name
                                                                           Dis                   S
                                                                                                 Samples         Correect        incorrect         The
  Disease Classif
3-D             fication                                                                         n
                                                                                                 number       classifica
                                                                                                                       ation               on
                                                                                                                               classificatio    accuracy%
                m                          ation process a
    Ant algorithm was used in the classifica             and                    Normal               10            10               0              100
                e(1) are used as entries fo algorithm a
the data in table                          or            and
                                                                             rial
                                                                           Atr Fibrillation          10            9                1              90
  ecify paramete to ant colon as following :
spe             ers         ny              g
                                                                             ntricular strikes
                                                                           Ven                       10            10               0              100
   t
Ant number=10 a              er
                 ants ,Numbe of test sampl  le=80                          Ron henomenon
                                                                             nT                      10            10               0              100
  obability thresh
Pro                          x             000
                 hold=0.9 ,Max iteration =10                               Big
                                                                             gemeny                  10            8                2              80
  =0.001 , φ =0
Τ0=              0.1
                                                                             ntricular
                                                                           Ven                       10            9                1              90
                                                                             chycardia
                                                                           Tac
∆τ                ………(4)
                  …                                                        Venntricular              10            9                1              90
                                                                           Fibrillation

                                                                             ird
                                                                           Thi Degree                10            9                1              90
dis        currentv
                  value           rvalue   ……(5)
                             center                                          art
                                                                           Hea Block

                                                                                Total                80            74               6              92.5
   I-            ce
VII Performanc evaluation methodsm
   We have u                    t
                used different methods f                 nce
                                             for performan
  aluation of clas
eva                             h            s.          ods
                 ssification of heart diseases These metho
                on accuracy, sensitivity and specific
are classificatio                                        city           VIII- Conclusions
measures.The des                ese           ill
                 scription of the methods wi be given in tthe             Ade              l                          gh
                                                                             equate medical information leads to a hig diagnostic
   lowing sub sec
foll            ctions[6].                                                   cy.
                                                                       accurac In this res               ve
                                                                                           search we hav proposed a system for
                                                                       comput              osis of heart d
                                                                              terized diagno                          d
                                                                                                         disease, based on medical




                                                                  79                                       http://sites.google.com/site/ijcsis/
                                                                                                           ISSN 1947-5500
                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                              Vol. 11, No. 5, May 2013

information s u c h a s ( E C G i m a g e , blood oxygen and         [13]Delores M. Pluto and Martha M. Phillips,2004,”Policy
blood pressure). To increase the efficiency of diagnosis                 and Environmental Indicators for Heart Disease and Stroke
extraction features from ECG by using e i g e n v a u l e                Prevention: Data Sources in Two States “,National Center
m a t r i x achieve high classification accuracy. Ant Colony             for Chronic Disease Prevention and Health, Volume: 1
algorithm has an efficient and accurate classification. Our              Issue: 2.
future work would employ more medical information and                 [14] O.K. Rybak and Ya.P. Dovgalevsky,2010,”Multifactor
experiment different additional heart disease types.                      discriminant analysis of electrocardiograms at screening
                                                                          patients with ischemic heart disease “, Saratov Journal of
                                                                          Medical Scientific Research, ISSN: 19950039 , Volume: 6
                         Reference                                        Issue: 1 Pages: 076-081.
                                                                      [15] R. C. Gonzalez and R. E. Woods, "Digital Image Processing
[1]Edward Wilson,(2002), An Overview of IntelligentSystems                " Pearson Education International, 2008
    echnologies for Process Optimization”, Myrtle Beach,SC.           [16]Hesam Komari Alaei,           Karim Salahshoor                and
    Symposium sponsored by the Iron and Steel Society’s                   Hamed Komari Alae,2013,” A new integrated on-line
    Computer Applications Committee.                                      fuzzy clustering and segmentation methodology with
[2] . NEGNEVITSKY, 2011, “Artificial Intelligence A Guide                 adaptive approach for process monitoring and fault
    to ntelligent Systems”, Pearson Educatio.                             detection and diagnosis”, soft computing, Volume 17,
[3] Rios, L.and Chaimowicz, L.,2009,”An Artificial                        issue3,PP 345-362.
    Inteligence for penTTD”, IEEE DOI:10.1109, Pages:52-63.           [17] Marco Dorigo, Mauro Birattari, and Thomas
[4]Bob       Waterbury,    (2002),    “Artificial  Intelligence           Stutzle,2006,” Ant Colony Optimization(Artificial Ant as
    Expands Frontiers in Asset Management”, Technical                     a     Computational     Intelligence      Technique)”,IEEE
    report: ISU CS-TR 95-01.                                              COMPUTATIONAL INTELLIGENCE MAGAZINE
[5]James J.Buckley,Esfandiar Eslami,(2003),“Advances in              [18]F. J. Vasko, J. D. Bobeck, M. A. Governale, D. J. Rieksts,
    Soft computing”,A pringer-Verlag Company, Physica-                   J. D. Keffer, 2011, "A statistical analysis of parameter
    Verlag Heidelberg.                                                   values for the rank-based ant colony optimization for the travelling
[6] Sengur A. and Ibrahim T. ,2008,” A hybrid method based               salesperson problem" Journal of the operational Research Society, Vol.
                                                                         62, No. 6, pp. 1169-1176, by Palgrave Macmillan.
    on artificial immune system and fuzzy K-NN algorithm for         [19] M. Mavrovouniotis, S. Yang, 2010, "Ant Colony Optimization with
    diagnosis of heart value diseases”, Expert System and                Immigrants Schemes in Dynamic Environments", Springer-Verlag Berlin
    applications, Elsevier puplisher.                                    Heidelberg PPSN XI, Part II, LNCS 6239, pp. 371-380.
[7] Ramteke R. and Manza R. ,2010,”Diagnosis and Medical             [20]Nada N. Saleem and Khalid N. Saleem,2009,”Design Hybrid Intelligent
                                                                         System for Transitional Bladder Cell Carcinoma Diagnosis”, Raf.J. of
    prescription of heart disease using support vector machine           Comp. & Math’s., Vol.6, No.1.
    and feed forward back propagation technique”, (IJCSE)
    International Journal on Computer Science and
    Engineering,Vol. 02, No. 06, 2010, 2150-2159
[8] Usha Rani,2011,” ANALYSIS OF HEART DISEASES
    DATASET USING NEURAL NETWORK APPROACH”,
    International Journal of Data Mining & Knowlclass
    Management Process (IJDKP) Vol.1, No.5, September
    2011.
[9]Jyothi Singaraju and Vanisree K,2011,” Decision Support
    System for Congenital Heart Disease Diagnosis based on
    Signs and Symptoms using Neural Networks“,International
    Journal of Computer Applications (0975 – 8887),Volume
    19– No.6, April 2011.
[10] Sameh Ghwanmeh,2012,” Applying Advanced NN-based
    Decision Support Scheme for Heart Diseases
    Diagnosis”,International Journal of Computer Applications
    (0975 – 8887) Volume 44– No.2, April 2012 37 .
[11]      Azhya      N.,2001,"Washington        Manual     Medical
    Therapentics",30th Edition, Lippincott Williams & Wilkins,
    USA, pages 195-215-627-751.
[12] Dr.Shihab, M.S. Shahina," An Intelligent Agent-Based
    Medication and Emergency System", Proceeding of ICCTA06,
    2nd IEEE International Conference of Information and
    Communication technology from theory to application, 24
    April 2006, Vol:2, PP: 1213-1215.




                                                               80                                   http://sites.google.com/site/ijcsis/
                                                                                                    ISSN 1947-5500