Medical Informatics
ISSN: 0307-7640 (Print) (Online) Journal homepage: http://www.tandfonline.com/loi/imif18
Disturbances of impulse formation: an expert system for ECG interpretation M. S. Habashi & M. A. Abdel-Bary To cite this article: M. S. Habashi & M. A. Abdel-Bary (1991) Disturbances of impulse formation: an expert system for ECG interpretation, Medical Informatics, 16:1, 29-41, DOI: 10.3109/14639239109025293 To link to this article: http://dx.doi.org/10.3109/14639239109025293
Published online: 12 Jul 2009.
Submit your article to this journal
Article views: 2
View related articles
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=imif18 Download by: [UNSW Library]
Date: 26 April 2016, At: 07:04
MED. INFORM.
(1991),
VOL.
16,
NO.
1, 29-41
Disturbances of impulse formation: an expert system for ECG interpretation M. S. HABASHI? and M. A. ABDEL-BARYI
t Faculty of Engineering, Electronics and Computer Department;
Downloaded by [UNSW Library] at 07:04 26 April 2016
$' Faculty of Medicine, Pharmacology Department, Ain-Shams University, Cairo, Egypt Abstract. This paper describes an expert system (ES) that aids in interpretation of some disturbances of impulse formation from electrocardiographic records. The system consists of a user interface, a knowledge base, an inference engine and an explanation facility. I t is implemented using Turbo PROLOG and uses the built-in interpreter for goal proving or disproving. The user interface gets information about the case by interrogation through multiple choice or Yes/No questions. The response is processed and stored in a dynamic database. After the interview the processed data are stored in a permanent file for subsequent calls. The knowledge base contains domain rules of the If-Then variety. The inference engine supports the logic-based method of knowledge organization, which is controlled by backward-chaining. The explanation facility is able to give reasons for any fact in the dynamic database. The main diagnosis, the diagnostic criteria and the algorithm used are explained and illustrated with examples. Sample outputs of the system are also given. Keywords: Artijcial intelligence; Expert system; Prolog; Electrocardiography; Dysrhythmia.
1. Introduction T h e heart is merely a muscular pump, the operation of which depends on electrical triggering of individual cells, more or less simultaneously. T h e triggering stimulus is initiated in the sinoatrial node and spreads to involve the whole heart. The pattern of spread is shown in figure 1. At resting state the myocardial cells are in a polarized state. External interference by the triggering stimulus converts the cell into a depolarized state. This depolarization is self-propagating to involve the whole myocardium and is responsible for heart contraction. Repolarization follows and the heart relaxes again (figure 2). Electrocardiography (ECG) is the recording of the summations of the previously mentioned electrical events by leads applied to the body surface. Each lead has a
( 7 ) Lt
v
( 5 ) Lt BB ( 6 ) PF
Figure 1. Order of spread of electrical activity to different chambers of the heart. Abbrmiations: SAN, sinoatrial node; AVN, atrioventricular node; AVB, atrioventricular bundle; BB, bundle branch; PF, purkinje fibres; V, ventricle; Lt, left; Rt, right. 0307-7640/93 $3.00 0 1991 Taylor & Francis Ltd.
M . S . Habashi and M . A . Abdel-Bary
30
-----
1,1 >p+.;;; t t t t t
Myocardial
t t t t t
---------Stimulus
(4
(4
(b)
Downloaded by [UNSW Library] at 07:04 26 April 2016
Figure 2. Electrical events in myocardial cells. ( a ) Polarization: excess positive charges on the outside and negative charges on the inside of the cell membrane. ( b ) Depolarization: reversal of the normal polarity. (c) Repolarization: resumption of the normal polarity.
I1 (60)
I11 (120) AVP ( S O )
Electrical axes of frontal ECG leads.
Figure 3.
-
ve \
\
\ \
\ \
-
ve
,__----------
:v7 Magnitude of the record in lead I
\\
t
ve lead I
Cardiac Vector
/
Magnitude of the record in lead I1 t Figure 4.
ve lead I1
T h e relation of electrical cardiac vector to lead axes.
recording electrode and a reference electrode. T h e line connecting the two electrodes is known as the lead axis (figure 3). T h e magnitude of voltage recording depends on the direction of the electrical activity in relation to the lead's axis (figure 4). T h e ECG is the record obtained by continuous tracing of the voltage during the cardiac cycle. Activities in various parts of the heart appear as recognizable waves. For example P-wave represents atrial depolarization, QRS-wave represents ventricular depolarization and T-wave represents ventricular repolarization (figure 5).
Disturbances of impulse formation
31
R
Q
Downloaded by [UNSW Library] at 07:04 26 April 2016
Figure 5.
B
The various waves of the ECG.
ECG is the most commonly applied technique in investigating heart problems. According to Rowlands [l] it is probably true that no field of study in clinical medicine is as littered with the intellectual corpus of those who have sought to master it and have failed as is the field of electrocardiography’. One possible reason for that is the interdependency of the electrocardiographic rules. Another reason is that the subject of ECG is based on empiricism to a great degree, i.e. the change in the E C G in a given pathological condition indicates the presence of that abnormality, not because it is possible by reasoning to predict the changes but because observation of a large number of subjects shows a high correlation between the given ECG changes and that pathological state. Though scientific explanation has been grafted onto the emperical framework, empiricism remains predominant. A third reason for complexity is the enormous range of possible appearances for the normal ECG. The previously mentioned difficulties have led us to think of the application of artificial intelligence (AI) [ 2 4 ] to ECG interpretation. T h e name of A1 was coined to the new phase in the computer revolution which started in the early 1970s. This phase was considered a breakthrough in the field of computer science. In the last 20 years the work of A1 scientists has spanned the expert system (ES) concept [5-7] a trial to simulate a real expert with the computer. A real expert is a person who, because of training experience, can normally do things the rest of us cannot. Knowledge engineers, in co-operation with domain experts, started to use the computer in handling extensive, high-quality specific knowledge (expertise) about some narrow problem area to create very specialized solutions. Broadly speaking, the artificial expert has many advantages over the real expert. First: permanency-artificial experience, unlike human experience, does not fade with disuse and does not need continuous practice and rehearsal to maintain proficiency. Second: consistency-it gives reproducible results in identical situations and is not affected by emotional factors like forgetfulness. Third: easy to transfer-while transferring knowledge from one human to another (by the education process) is laborious and expensive, the transfer of artificial experience is an easier process. Fourth: affordable-low cost. Fifth: documentation-reasons could be given for the achieved decisions. Now ES have already gained popularity and applications in so many fields (agriculture, chemistry, physics, engineering, law, military sciences, etc.). Concerning medicine, few attempts were made in the early 1970s to apply ES to medical problems (e.g. ARAMIS [8] for rheumatic diseases, MYCIN [9, 101 for infectious diseases). However, in the 1980s medical ES started to spread and to involve the various branches of medicine. In the cardiovascular field many ES were introduced.
M . S. Habashi and M . A . Abdel-Bary
32
Downloaded by [UNSW Library] at 07:04 26 April 2016
Here are some examples. (1) A N N A [ l l ] and Digitalis Advisor [12] to assist the physician in administration of digitalis to patients with heart failure. (2) Diagnoser [13] and G A L E N [14] to help in identifying congenital heart diseases. (3) Heart image interpreter [15] to analyse two-dimensional intensity distribution images of the heart. (4) M O D I S [16] and HT-Attending [17] for diagnosing and managing hypertension. ( 5 ) M E C S - A I [18, 191 for diagnosing heart and thyroid diseases. ( 6 ) M I [20] for diagnosing mycordial infarction. (7) M E D I [21] for chest pain diagnosis. (8) P I P [22, 231 for oedema diagnosis. (9) System D [24] for dizziness diagnosis. In this paper we try to put a milestone ES for an E C G interpretation. It interprets the ECG data, diagnoses the clinical abnormality and gives full documentation for the diagnosis. This can aid in training inexperienced clinicians and medical students to decide provisionally on the type of dysrhythmia in a particular patient. T h e diagnosis depends on the human assessment of the E C G parameters and which, though not so accurate, gives flexibility in the interpretation by allowing the user to change his mind about any segment of the ECG. T h e individual patient data can be stored for later inspection and consultation. This ES can be extended later on, o r can be included in a cardiovascular ES.
2. Expert s y s t e m s 2.1. Introduction An expert system (ES) is an intelligent computer program that uses knowledge and inference procedures to solve problems that are difficult enough to require significant human expertise [25-271. It is called system because, in addition, it contains a problem-solving component and a support component. T h e latter helps the user interact with the main program. T h e process of building an ES is often called knowledge engineering (KE). I t typically involves a special form of interaction between the ES builder and one or more human experts in some problem area. T h e heart of an ES is the powerful corpus of knowledge that accumulates during system building. Knowledge is clearly separated from the program and input data. T h e knowledge base (KB), is where the specific domain knowledge resides. It takes the form of facts, which are the basic data elements within the domain; inference rules, which are rules to guide the use of knowledge; beliefs, which are a measure of importance, priority or confidence; and heuristics, which are the rules of thumb or the general guidelines. T h e most useful feature of an E S is the high-level expertise it provides to aid in problem solving. This expertise can represent the best thinking of the top experts in the field, leading to solutions that are imaginative, accurate and efficient. I t can also explain how the ES produces answers for the problem under consideration. I t is a way of justification of decisions arrived at. I n addition, ES generally provides a training facility for key personnel and important staff members in the specified domain.
33
Disturbances of impulse formation
ES normally grows incrementally. T h e evolution normally proceeds from simple to hard tasks by incrementally improving the organization and representation of the system’s knowledge; hence the system itself can assist in the development effort. ES development can be viewed as the following, highly interdependent and overlapping phases: identification, conceptualization, formalization, implementation and testing.
Downloaded by [UNSW Library] at 07:04 26 April 2016
2.2. Expert system components In general, ES can be divided into the main components shown in figure 6 . T h e user interface (UI) allows the user to communicate with the ES. It is responsible for taking the user’s commands and questions as well as formatting information generated by the system. T h e best form for an interface is to use some form of natural language. The explanation facility (EF) gives an explanation of the reasoning used by the ES in solving a particular problem. This normally requires the sequence of knowledge items used during solving the problem. In many ES a tree structure is built as the problem solution unfolds. This is then used during the explanation process. The knowledge base (KB) is where the domain-specific knowledge is accumulated. It has a structure decided by the knowledge engineer (e.g. rules, frames, semantic networks, etc.). T h e working database (WDB) holds the current status of the problem being solved. It maintains all data items supplied by the user and those inferred by the ES for itself. T h e inference engine (IE) is an interpreter that controls the general operation of the ES. It has a powerful pattern matcher (PM) and a control regime (CR). T h e IE Domain Expert & Knowledge Engineer
Uaer
I4 I I
Description:: Advice & of new Explanation I 1 I 1 Caae
4:
Explanations :: New & Analyais : ! Knowledge ( 1I I
II I I I I
9I
b
!Modification 1 to KB
U s e r Interface (UI)
Acquisition Explanation Facility (EF)
Inference Engine
4l---------
Knowledge Base (KB)
A
i
1
Figure 6.
Expert system components.
M . S. Habashi and M . A . Abdel-Bary
Downloaded by [UNSW Library] at 07:04 26 April 2016
34
drives its way through the KB until it finds a data structure that matches the data in the WDB. T h e CR determines how the interpreter looks for the data structure and how any fired data structure should be handled. Normally the CR enables data structure to fire. If at any stage it realizes that the line of reasoning being pursued is false, or not leading to a productive line, it is possible to revert to an earlier state by backtracking up the solution tree to establish a solution via an alternative route, if any. Search is normally performed by depth-Jirst method (DFM) with backtracking around an and/or tree which is generated as the problem solution is developed. An ES may use a forward-driven system which works from higher-level concepts to more detailed information, or a backward-driven system which works from a detailed information to a higher-level concept. Some complicated ES use both strategies combined together. The knowledge acquisition module (KAM) is used only by the KE or the domain expert to enter knowledge, expand, delete, or modify the KB. It should also check for the integrity and consistency of the given data.
2.3. Details of the implemented ES Except for the knowledge acquisition module (KAM), the ES developed contains all other components mentioned above. Though they might differ in some details, they agree in main outlines and perform almost the same functions. T h e details are given below. The knowledge was given as a permanent part. Modifications could be done only offline. The knowledge base (KB) contains inference rules and facts. Inference rules are of the If-Then variety, i.e. I F premise T H E N consequence. The left-hand side of the rule may be a simple or a complex premise. T h e complex premise is a premise consisting of more than one logically ANDed simple premises. If logical ORing is needed, the rules are repeated with the individual simple premises since the language used, Turbo-Prolog, does not support logical ORing in the usual form as in standard Prolog. An example in shown in figure 7. T h e facts in the K B are those needed to judge the validity of the user’s answers, e.g. ‘The patient’s sex belongs to the set {male, female}’. If the user gives unexpected answers, they will be rejected. The working database (WDB) is created during the running of the program. It contains facts about the present case. There are two types of facts. First, those gained by interrogating the user, e.g. rhythm-component-node (it contains information
0 IF
0 IF 0 IF
Figure 7.
the P-wave can’t be identified and the QRS are normal T H E N the AVN is the origin of the rhythm the P-waves lie after the QRS, the P-waves are inverted and the QRS are normal T H E N the AVN is the origin of the rhythm. the P-waves lie before the QRS, the P-waves are inverted and the P-R intervals are short T H E N the AVN is the origin of the rhythm.
If-Then rules showing the same consequent but each has a different premise. Repetition of the rules replaces logical ORing.
35
Disturbances of impulse formation T o prove NSR apply the following rules:
goal
IF SO, RR and N R T H E N NSR
rule 1
IF N P and NP-R T H E N SO
rule 2
IF NP, NP-R, RR and N R T H E N NSR
implied rule from rules 1, 2
Then the questions asked to the user would be: Are NP, N-PR, RR, and N R true?
data interrogated from the user
Downloaded by [UNSW Library] at 07:04 26 April 2016
T h e following abbreviations are used:
SO NSR RR
Sinus origin Normal sinus rhythm Regular rhythm Figure 8.
NR Normal rate NP-R Normal P-R wave NP Normal P-wave
A sample of the inference mechanism.
about the name of the ECG strip, the examined ECG component and information whether the value of the examined component is available). T h e other type of facts are those concluded by the system through the inference engine (IE), e.g. rhythmorigin-node (it contains information about the name of the ECG strip, the origin of the rhythm, and whether the examined condition is proved or disproved). At the end of every session the contents of the WDB are saved for later consultation. T h e inference engine (IE) uses the built-in goal-oriented backward-chaining mechanism available in Turbo-Prolog. T h e system starts with a goal statement and works backwards through inference rules to find the data that establish that goal. An example is given in figure 8. The inference engine contains rules for the following rhythms [28-331:
(1) Sinus rhythm ( a ) Normal sinus rhythm ( b ) Sinus rhythm disturbances (i) Sinus tachycardia (ii) Sinus bradycardia (iii) Sinus irregularity (iv) Sinus arrest (v) Respiratory sinus arrythmia (vi) Sinoatrial block
(2) Atrial rhythm disturbances ( a ) Wandering atrial pacemaker ( b ) Atrial escape rhythm ( c ) Atrial tachycardia (3) Nodal rhythm disturbances ( a ) Nodal tachycardia ( b ) Nodal escape rhythm (c) Accelerated nodal rhythm
(4) Supraventricular tachycardia
M . S . Habashi and M . A. Abdel-Bury
36
T h e user’s interface (UI) is useful when some facts are required from the user, e.g. the conditions given in figure 8. T h e user will be prompted to give the needed data, e.g. ‘Are the P-R intervals within normal?’. T h e UI will read the user’s response and check its validity. In case of errors, error messages will be given. T h e user’s response will be then stored in the WDB. T h e system has also an explanation facility (EF). At the end of every session the E F will search the W D B for the facts that support the diagnosis and gives the user a detailed explanation for the final diagnosis, as will be shown in the next section.
Downloaded by [UNSW Library] at 07:04 26 April 2016
3. Implementation 3 . 1 . Algorithm T h e algorithm for ECG interpretation is discussed in this section. A sample for normal sinus rhythm diagnosis is explained in detail. Other parts of the system work similarly. T h e system works in the following manner:
(1) Obtains the personal history of the patient (if no patient file is already available). (2) Recalls any stored data from the patient’s file to the WDB. (3) Searches the rythm tree. There are 14 rhythm types; each will lead the system to a different subtree. Common conditions are searched first, since they are the most probable ones to happen (e.g. normal sinus rhythm before sinus tachycardia). (4) T o check each one of the previous conditions the system searches for essential diagnostic subconditions. If the subconditions are not already available in the system’s WDB, the user will be prompted to enter them, User’s response is processed and added to the database. (5) At the end of the session the accumulated data are used to update the patient’s file, which is saved for later use. 3 . 2 . Example: Diagnosis of normal sinus rhythm As an illustration, the search tree for normal sinus rhythm is shown in figure 9. To prove normal sinus rhythm, either the left or the right subtree should be fulfilled. This means that the two subtrees are logically ORed. All branches of any subtree should be fulfilled to reach the indicated diagnosis, i.e. the branches of the subtree are logically ANDed. When neither subtree is proven, this implies that the diagnosis of normal sinus rhythm is discarded and another condition will be searched. T h e sequence of traversing the branches of the tree is from left to right. For the logical order of the session questionnaire, some data might be asked forin advance-irrespective of their immediate usefulness. A sample of the output of the session is shown in figure 10. and the file contents generated by the system for the corresponding patient is given in figure 11.
4. Conclusions T h e system has pros and contras which might be summarized as follows.
4.1. Limitations Not all types of rhythm disturbances are included in this phase of the system development. It could easily be expanded by adding the corresponding diagnostic rules for each disease.
37
Disturbances of impulse formation
Downloaded by [UNSW Library] at 07:04 26 April 2016
N o r m a l e i n u s rhythm
Each QRS has a
P-waves precede QRS
Upright P-waves
Each QRS has a P-wave
Sinus rhythm
Similar P-waves
K
Normal
be w e l l aeseeeed
QRS
Figure 9.
Heart r a t e = 80 120
-
Normal P-R intervals
Supraventricular rhythm
P-wave c a n ' t
Regular rhythm
Normal P-R intervalB
Regular rhythm
Search tree for normal sinus rhythm.
Heart rate= 100-120
Rhythm Diagnosis Please give the file name you want to deal with: patient 1 Please complete the following data about the patient’s personal history: name: sex:
Ali male
Birth date: year: 1960 month: 5 day: 4 marital status: single Give the name of the ECG strip to be examined
Downloaded by [UNSW Library] at 07:04 26 April 2016
v2 Can the P-waves be easily identified? Yes Does each QRS have a P-wave? yes
Do the P-waves lie before QRS? yes Are the P-waves of upright contour? yes Are the P-waves similar in shape? yes Are the P-R intervals homogeneous in length all through the strip? Y €5
Are the P-R intervals short? no
Is the P-P cycle length quite regular? yes Calculation o f P-wave rate Please count the no. of cycles in the strip; no. of cycles = 5 Please count the corresponding no. of great squares; no. of great squares = 15
Diagnosis Normal sinus rhythm at a rate of 100 is the most accepted diagnosis of V2 because: (1) T h e rhythm is probably sinus in origin due to the following causes: ( a ) T h e P-wave contour is upright (b) The P-waves are similar in shape (c) T h e P-waves do lie before QRS (d) The P-R interval is not short
(2) The calculated heart rate=100 Figure 10. (a)
Disturbances of impulse formation
39
Rhythm Diagnosis Type the name of the next strip If no further strips -P Type no
L1 Can the P-waves be easily identified? no May the shape of the T-wave that precedes the QRS suggest superimposed P-wave? Yes
Downloaded by [UNSW Library] at 07:04 26 April 2016
Can you identify or suspect a P-wave for every QRS? Yes
Is the QRS complex normal in shape? Yes Are the P-R intervals within normal? Yes
Is the P-P cycle length quite regular? Yes Calculation of QRS rate Please count the no. of cycles in the strip; no. of cycles = S Please count the corresponding no. of great squares; no. of great squares = 10 Did you try to terminate this tachycardia by vagal stimulation? if yes, mention the result: (a) is not tired (b) terminates the attack abruptly (c) fails to terminate the attack (d) terminates the attack gradually
d
Diagnosis Sinus tachycardia at a rate of 150 is the most accepted diagnosis of L1 because:
( 1 ) The rythym is probably supraventricular in origin due to the following causes: (i) the QRS is normal (ii) the T-wave contour may suggest superimposed P-wave
(2) Vagal stimulation terminates the attack gradually (3) T h e calculated heart rate = 1SO Figure 10. (b) Figure 10. A sample session for patient 1. ( a )Strip 1: normal sinus rhythm; (b) strip 2: sinus tachycardia.
M . S . Habashi and M . A . Abdel-Bary
40
rhythm,diagnosis("Ll","nlO2","yes", 150, ["n53",::ne'' "m2"]) r h y t hm-d iag nos i 8 ( " V 2 " " n 1 00" , yes " , 1 00 ,[ " n 5 0 " , m 1 j 1 I'
rhythm_origin("Ll","n53","yes", ["nl", "n6"]) rhythm,orlgin("V2","n5O","yes", ["n4","n3", "n2","n5"11
rhythm-component-node( "V2","no" ,"a",[ "yes"]) rhythm-component-node( "V2","81 "y" [ "yes"]1 rhythm-component-node( "V2","n2","a",[ "yes"3 1 rhythm~component~node~"V2","n4","a",["yes"]) rhythm-component-node( "V2","n3","a" [ "yes"3 ) rhythm-component-node( "V2","s2","y", [ "yes"]) rhythm-component-node( "V2","n5"," b " , "no"] rhyth~component~node("V2","nlO","a", ["yes"]) rhythm-component-node( "Ll ","no"," a " , [ "no"]) ' I ,
Downloaded by [UNSW Library] at 07:04 26 April 2016
rhythm,component~node("Ll","ne","b",
rhythm-component-node( rhythm-component-node( rhythm-component-node( rhythm-component-node( rhythm-component-node(
"L1
'I,
["yes"])
"s3","y" ,["yes"])
"L1 " ,"nl " , "a",[ "yes"]) "L1 "n5"," a ' * ,[ "yes"]) "L1 "nl0" ,"a",["yes"]) "L1 "na", "d" [ "yes"]1 rhythm,rate,node("V2","ml", 100,["ca culated"]) r hythm-rate-node ( L 1 , "m2" ,1 5 0 , [ ca cu 1 ated I ) rhythm-rate-node( "L1 "ml 150, [ " c a culated"]) rhythm-rate-node("Ll","m2", 150,["ca culated"] ) datum( "year",1960) datum( "month",5 ) datum( "day",4) datum-1 ( "name","Ahmed Hasean A1 1 " ) datum-1 ( " s e x " , "male") datum-1 ("marital-status" ,"single") "
'I,
"
' I ,
Figure 1 1 .
'I,
' I ,
"
"
'I,
The contents of the file generated by the ES for patient 1
T h e user response is evaluated to be either true of false. Addition of probability factors to weigh the user's response is planned to be added to give more flexibility in the diagnosis. T h e user is not able to delete or modify-on-line-any data after file saving. An interactive interface for this task can easily be added to the ES.
4.2. Advantages T h e system is user-friendly. T h e available explanation facility provides a logical trace for the diagnosis. T h e system is easily expandable. T h e system can be used as an ES shell for other diseases in the medical field. T h e system could be used as a training tool for checking the E C G interpretation. Interpreted ECG records for a patient are stored for further recall when needed. References 1 . ROWLANDS, D. J. (1981) Understanding the Electro-cardiogram: A New Approach (London: Cower Medical). 2. RICH,E. (1983) Artificial Intelligence (New York: McGraw-Hill). 3 . NILSSON, N. J . (1982) Principles of Artificial Intelligence (Berlin: Springer-Verlag). 4. WINSTON, P. H. (1984) Artificial Intelligence, 2nd edn (Reading: Addison-Wesley). 5. WATERMAN, D. A. (1986) A guide to Expert Systems (Menlo Park: Addison-Wesley). 6. ANDRIOLE, S. J . (1985) Applications in Artificial Intelligence (Englewood Cliffs: Prentice-Hall. 7. JACKSON, P. (1986) Introduction to Expert Systems (Reading: Addison-Wesley).
Downloaded by [UNSW Library] at 07:04 26 April 2016
Disturbances of impulse formation
41
8. FRIES,J. F. (1972) Time-oriented patient records and a computer data-bank. Journal ofthe American Medical Association, 222 (12), 1536-1542. 9. BUCHANAN, B. G., and SHORTCLIFFE, E. H . (1984) Rule-Based Expert Systems. The M Y C Z N Experiment of the Stanford Heuristic Programming Project (Reading: Addison-Wesley). S. G., BUCHANAN, B. G., MERIGAN, T . C., and COHEN,S. N. (1973) An 10. SHORTLIFFE, E. H., ALIXINE, artificial intelligence program to advise physicians regarding antimicrobial therapy. Computers and Biomedical Research, 6, 544-560. 11. SILVERMAN, H . (1974) A digitalis therapy advisor. MZT Technical Report T R - 1 4 3 , December. W. R. (1977) A digitalis therapy advisor with explanations. M Z T Technical Report T R 12. SWARTOUT, 176, February. F., MOLLER. J . , and PRIETULA, M . (1981) Expertiseand 13. JOHNSON, P. E., DURAN, A. S., HASSERBROCK, error in diagnostic reasoning, Cognitive Science, 5, 235-283. 14. THOMPSON, W. B., JONSON, P. E., and MOEN.J. B. (1983) Recognition-based diagnostic reasoning. Proceedings ZJCAZ, 83. 236-238. 15. NIEMANN, H., BUNKE, H., HOFMANN, I., and SAGERER, G. (1984) Diagnostic inferences from image sequences-a knowledge based approach. Proceedings of the First Conference on Artificial Intelligence Applications (New York: IEEE Computer Society). V. A,, KOVRIGIN, 0. V., and SMOL’YANINOV, N. D. (1982) An expert for 16. BOKHUA, N. K., GELOVANI, diagnostic of various forms of atrial hypertension, Tekh. Kibern. [Engineering Cybernetics], 20(6), 106-111. 17. MILLER,P. L., and BLACK,H. R. (1984) Medical plan-analysis by computer: Critiquing the pharmacologic management of essential hypertension. Computers and Biomedical Research, 17,38-54. S., KOYAMA, T., MINAMIKAWA, T . , and YASAKA, T. A. (1978) A rule-based physicians’ 18. KAIHARA, consultation system for cardiovascular diseases. Proceedings of the International Conference on Cybernetics and Society (New York: IEEE), 85-88. 19. KAIHARA, S., and KOYAMA, T. (1982) Medical consultation system with practical requirementsdevelopment of MECS-AI. Computer Science and Technologies (Japan Annual Review in Electronics, Computer and Telecommunications), vol. 4 (Amsterdam: North-Holland). C. A. (1983) An efficient scheme for time-dependent 20. KASTNER, J . K., WESS,S. M., and KULIKOWSKI, consultation systems. Proceedings MEDZNFO-83, Fourth World Conference on Medical Informatics (Amsterdam: North-Holland), 619-622. 21. PUPPE,F., and PUPPE,B. (1983) Overview on MEDI: A Heuristic diagnosis system with an efficient control structure. Report SEKZ-83-02 (Universitaet Kaiserslautern, West Germany: Fachbereich Informatik). 22. PAUKER, S. G., GORRY, G . A,, KASSIRER, J. P., and SCHWARTZ, W. B. (1976) Towards the stimulation of clinical cognition: taking a present illness by computer. American Journal ofMedicine, 60,981-996. 23. KULIKOWSKI, C. A. (1980) Artificial intelligence methods and systems for medical consultation. ZEEE Transactions on Pattern Analysis and Machine Intelligence, 2(S), 464476. 24. REGGIA,J. A., NAU,D. S., and WANG,P. Y. (1983) Diagnostic expert systems based on a set converting model. Znternationat Journal of Machine Studies, 19, 437460. E. A. (eds) (1982) TheHandbookof Artificialfntelligence, 25. BARR, A,, COHEN,P. R., and FEIGENBAUM. 3 vols (London: Pitman). 26. KOWALSKI, R. (1979) Logic for Problem Solving (New York: American Elsevier. 27. KOWALSKI, R. (1979) Algorithm = Logic Control. A C M Communications, 22(7), July. 28. BRAUNWALD, E. (1980) Heart Disease (Philadelphia: W. B. Saunders). 29. CORDAY, E., and IRVING,D . W. (1961) Disturbances of Heart Rate, Rhythm and Conduction (Philadelphia: W. B. Saunders). 30. FLEMING, J. S. (1979) Introduction to the Electro-cardiogram (New Jersey: Update). 31. HURST,J. W., and MYEBURG, R. J. (1975) Introduction to Electrocardiography, 2nd edn (New York: McGraw-Hill). 32. PHILLIPS,R. E., and FEENEY,M. K. (1980) The Cardiac Rhythm, A Systematic Approach to Interpretation, 2nd edn (Philadelphia: W. B. Saunders). 33. SCHERE, D., and COHEN,J . (1 968) Atrioventricular rhythm, Progress in Cardiovascular Disease. 34. WEISS,S. M., and KUI.IKOWSKI, C. A. (1 983) A Practical Guide to Designing Expert Systems (London: Chapman & Hall). 35. WIT,A.L., WEISS,M. B., BERKOWITZ, W. D., ROSEN,K. M., STEINER, C.,and DAMATO,A. N. (1970) Patterns of atrioventricular conduction in human heart, Circulation Research, 27 (3), 345-359.
+
ISSN: 0307-7640 (Print) (Online) Journal homepage: http://www.tandfonline.com/loi/imif18
Disturbances of impulse formation: an expert system for ECG interpretation M. S. Habashi & M. A. Abdel-Bary To cite this article: M. S. Habashi & M. A. Abdel-Bary (1991) Disturbances of impulse formation: an expert system for ECG interpretation, Medical Informatics, 16:1, 29-41, DOI: 10.3109/14639239109025293 To link to this article: http://dx.doi.org/10.3109/14639239109025293
Published online: 12 Jul 2009.
Submit your article to this journal
Article views: 2
View related articles
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=imif18 Download by: [UNSW Library]
Date: 26 April 2016, At: 07:04
MED. INFORM.
(1991),
VOL.
16,
NO.
1, 29-41
Disturbances of impulse formation: an expert system for ECG interpretation M. S. HABASHI? and M. A. ABDEL-BARYI
t Faculty of Engineering, Electronics and Computer Department;
Downloaded by [UNSW Library] at 07:04 26 April 2016
$' Faculty of Medicine, Pharmacology Department, Ain-Shams University, Cairo, Egypt Abstract. This paper describes an expert system (ES) that aids in interpretation of some disturbances of impulse formation from electrocardiographic records. The system consists of a user interface, a knowledge base, an inference engine and an explanation facility. I t is implemented using Turbo PROLOG and uses the built-in interpreter for goal proving or disproving. The user interface gets information about the case by interrogation through multiple choice or Yes/No questions. The response is processed and stored in a dynamic database. After the interview the processed data are stored in a permanent file for subsequent calls. The knowledge base contains domain rules of the If-Then variety. The inference engine supports the logic-based method of knowledge organization, which is controlled by backward-chaining. The explanation facility is able to give reasons for any fact in the dynamic database. The main diagnosis, the diagnostic criteria and the algorithm used are explained and illustrated with examples. Sample outputs of the system are also given. Keywords: Artijcial intelligence; Expert system; Prolog; Electrocardiography; Dysrhythmia.
1. Introduction T h e heart is merely a muscular pump, the operation of which depends on electrical triggering of individual cells, more or less simultaneously. T h e triggering stimulus is initiated in the sinoatrial node and spreads to involve the whole heart. The pattern of spread is shown in figure 1. At resting state the myocardial cells are in a polarized state. External interference by the triggering stimulus converts the cell into a depolarized state. This depolarization is self-propagating to involve the whole myocardium and is responsible for heart contraction. Repolarization follows and the heart relaxes again (figure 2). Electrocardiography (ECG) is the recording of the summations of the previously mentioned electrical events by leads applied to the body surface. Each lead has a
( 7 ) Lt
v
( 5 ) Lt BB ( 6 ) PF
Figure 1. Order of spread of electrical activity to different chambers of the heart. Abbrmiations: SAN, sinoatrial node; AVN, atrioventricular node; AVB, atrioventricular bundle; BB, bundle branch; PF, purkinje fibres; V, ventricle; Lt, left; Rt, right. 0307-7640/93 $3.00 0 1991 Taylor & Francis Ltd.
M . S . Habashi and M . A . Abdel-Bary
30
-----
1,1 >p+.;;; t t t t t
Myocardial
t t t t t
---------Stimulus
(4
(4
(b)
Downloaded by [UNSW Library] at 07:04 26 April 2016
Figure 2. Electrical events in myocardial cells. ( a ) Polarization: excess positive charges on the outside and negative charges on the inside of the cell membrane. ( b ) Depolarization: reversal of the normal polarity. (c) Repolarization: resumption of the normal polarity.
I1 (60)
I11 (120) AVP ( S O )
Electrical axes of frontal ECG leads.
Figure 3.
-
ve \
\
\ \
\ \
-
ve
,__----------
:v7 Magnitude of the record in lead I
\\
t
ve lead I
Cardiac Vector
/
Magnitude of the record in lead I1 t Figure 4.
ve lead I1
T h e relation of electrical cardiac vector to lead axes.
recording electrode and a reference electrode. T h e line connecting the two electrodes is known as the lead axis (figure 3). T h e magnitude of voltage recording depends on the direction of the electrical activity in relation to the lead's axis (figure 4). T h e ECG is the record obtained by continuous tracing of the voltage during the cardiac cycle. Activities in various parts of the heart appear as recognizable waves. For example P-wave represents atrial depolarization, QRS-wave represents ventricular depolarization and T-wave represents ventricular repolarization (figure 5).
Disturbances of impulse formation
31
R
Q
Downloaded by [UNSW Library] at 07:04 26 April 2016
Figure 5.
B
The various waves of the ECG.
ECG is the most commonly applied technique in investigating heart problems. According to Rowlands [l] it is probably true that no field of study in clinical medicine is as littered with the intellectual corpus of those who have sought to master it and have failed as is the field of electrocardiography’. One possible reason for that is the interdependency of the electrocardiographic rules. Another reason is that the subject of ECG is based on empiricism to a great degree, i.e. the change in the E C G in a given pathological condition indicates the presence of that abnormality, not because it is possible by reasoning to predict the changes but because observation of a large number of subjects shows a high correlation between the given ECG changes and that pathological state. Though scientific explanation has been grafted onto the emperical framework, empiricism remains predominant. A third reason for complexity is the enormous range of possible appearances for the normal ECG. The previously mentioned difficulties have led us to think of the application of artificial intelligence (AI) [ 2 4 ] to ECG interpretation. T h e name of A1 was coined to the new phase in the computer revolution which started in the early 1970s. This phase was considered a breakthrough in the field of computer science. In the last 20 years the work of A1 scientists has spanned the expert system (ES) concept [5-7] a trial to simulate a real expert with the computer. A real expert is a person who, because of training experience, can normally do things the rest of us cannot. Knowledge engineers, in co-operation with domain experts, started to use the computer in handling extensive, high-quality specific knowledge (expertise) about some narrow problem area to create very specialized solutions. Broadly speaking, the artificial expert has many advantages over the real expert. First: permanency-artificial experience, unlike human experience, does not fade with disuse and does not need continuous practice and rehearsal to maintain proficiency. Second: consistency-it gives reproducible results in identical situations and is not affected by emotional factors like forgetfulness. Third: easy to transfer-while transferring knowledge from one human to another (by the education process) is laborious and expensive, the transfer of artificial experience is an easier process. Fourth: affordable-low cost. Fifth: documentation-reasons could be given for the achieved decisions. Now ES have already gained popularity and applications in so many fields (agriculture, chemistry, physics, engineering, law, military sciences, etc.). Concerning medicine, few attempts were made in the early 1970s to apply ES to medical problems (e.g. ARAMIS [8] for rheumatic diseases, MYCIN [9, 101 for infectious diseases). However, in the 1980s medical ES started to spread and to involve the various branches of medicine. In the cardiovascular field many ES were introduced.
M . S. Habashi and M . A . Abdel-Bary
32
Downloaded by [UNSW Library] at 07:04 26 April 2016
Here are some examples. (1) A N N A [ l l ] and Digitalis Advisor [12] to assist the physician in administration of digitalis to patients with heart failure. (2) Diagnoser [13] and G A L E N [14] to help in identifying congenital heart diseases. (3) Heart image interpreter [15] to analyse two-dimensional intensity distribution images of the heart. (4) M O D I S [16] and HT-Attending [17] for diagnosing and managing hypertension. ( 5 ) M E C S - A I [18, 191 for diagnosing heart and thyroid diseases. ( 6 ) M I [20] for diagnosing mycordial infarction. (7) M E D I [21] for chest pain diagnosis. (8) P I P [22, 231 for oedema diagnosis. (9) System D [24] for dizziness diagnosis. In this paper we try to put a milestone ES for an E C G interpretation. It interprets the ECG data, diagnoses the clinical abnormality and gives full documentation for the diagnosis. This can aid in training inexperienced clinicians and medical students to decide provisionally on the type of dysrhythmia in a particular patient. T h e diagnosis depends on the human assessment of the E C G parameters and which, though not so accurate, gives flexibility in the interpretation by allowing the user to change his mind about any segment of the ECG. T h e individual patient data can be stored for later inspection and consultation. This ES can be extended later on, o r can be included in a cardiovascular ES.
2. Expert s y s t e m s 2.1. Introduction An expert system (ES) is an intelligent computer program that uses knowledge and inference procedures to solve problems that are difficult enough to require significant human expertise [25-271. It is called system because, in addition, it contains a problem-solving component and a support component. T h e latter helps the user interact with the main program. T h e process of building an ES is often called knowledge engineering (KE). I t typically involves a special form of interaction between the ES builder and one or more human experts in some problem area. T h e heart of an ES is the powerful corpus of knowledge that accumulates during system building. Knowledge is clearly separated from the program and input data. T h e knowledge base (KB), is where the specific domain knowledge resides. It takes the form of facts, which are the basic data elements within the domain; inference rules, which are rules to guide the use of knowledge; beliefs, which are a measure of importance, priority or confidence; and heuristics, which are the rules of thumb or the general guidelines. T h e most useful feature of an E S is the high-level expertise it provides to aid in problem solving. This expertise can represent the best thinking of the top experts in the field, leading to solutions that are imaginative, accurate and efficient. I t can also explain how the ES produces answers for the problem under consideration. I t is a way of justification of decisions arrived at. I n addition, ES generally provides a training facility for key personnel and important staff members in the specified domain.
33
Disturbances of impulse formation
ES normally grows incrementally. T h e evolution normally proceeds from simple to hard tasks by incrementally improving the organization and representation of the system’s knowledge; hence the system itself can assist in the development effort. ES development can be viewed as the following, highly interdependent and overlapping phases: identification, conceptualization, formalization, implementation and testing.
Downloaded by [UNSW Library] at 07:04 26 April 2016
2.2. Expert system components In general, ES can be divided into the main components shown in figure 6 . T h e user interface (UI) allows the user to communicate with the ES. It is responsible for taking the user’s commands and questions as well as formatting information generated by the system. T h e best form for an interface is to use some form of natural language. The explanation facility (EF) gives an explanation of the reasoning used by the ES in solving a particular problem. This normally requires the sequence of knowledge items used during solving the problem. In many ES a tree structure is built as the problem solution unfolds. This is then used during the explanation process. The knowledge base (KB) is where the domain-specific knowledge is accumulated. It has a structure decided by the knowledge engineer (e.g. rules, frames, semantic networks, etc.). T h e working database (WDB) holds the current status of the problem being solved. It maintains all data items supplied by the user and those inferred by the ES for itself. T h e inference engine (IE) is an interpreter that controls the general operation of the ES. It has a powerful pattern matcher (PM) and a control regime (CR). T h e IE Domain Expert & Knowledge Engineer
Uaer
I4 I I
Description:: Advice & of new Explanation I 1 I 1 Caae
4:
Explanations :: New & Analyais : ! Knowledge ( 1I I
II I I I I
9I
b
!Modification 1 to KB
U s e r Interface (UI)
Acquisition Explanation Facility (EF)
Inference Engine
4l---------
Knowledge Base (KB)
A
i
1
Figure 6.
Expert system components.
M . S. Habashi and M . A . Abdel-Bary
Downloaded by [UNSW Library] at 07:04 26 April 2016
34
drives its way through the KB until it finds a data structure that matches the data in the WDB. T h e CR determines how the interpreter looks for the data structure and how any fired data structure should be handled. Normally the CR enables data structure to fire. If at any stage it realizes that the line of reasoning being pursued is false, or not leading to a productive line, it is possible to revert to an earlier state by backtracking up the solution tree to establish a solution via an alternative route, if any. Search is normally performed by depth-Jirst method (DFM) with backtracking around an and/or tree which is generated as the problem solution is developed. An ES may use a forward-driven system which works from higher-level concepts to more detailed information, or a backward-driven system which works from a detailed information to a higher-level concept. Some complicated ES use both strategies combined together. The knowledge acquisition module (KAM) is used only by the KE or the domain expert to enter knowledge, expand, delete, or modify the KB. It should also check for the integrity and consistency of the given data.
2.3. Details of the implemented ES Except for the knowledge acquisition module (KAM), the ES developed contains all other components mentioned above. Though they might differ in some details, they agree in main outlines and perform almost the same functions. T h e details are given below. The knowledge was given as a permanent part. Modifications could be done only offline. The knowledge base (KB) contains inference rules and facts. Inference rules are of the If-Then variety, i.e. I F premise T H E N consequence. The left-hand side of the rule may be a simple or a complex premise. T h e complex premise is a premise consisting of more than one logically ANDed simple premises. If logical ORing is needed, the rules are repeated with the individual simple premises since the language used, Turbo-Prolog, does not support logical ORing in the usual form as in standard Prolog. An example in shown in figure 7. T h e facts in the K B are those needed to judge the validity of the user’s answers, e.g. ‘The patient’s sex belongs to the set {male, female}’. If the user gives unexpected answers, they will be rejected. The working database (WDB) is created during the running of the program. It contains facts about the present case. There are two types of facts. First, those gained by interrogating the user, e.g. rhythm-component-node (it contains information
0 IF
0 IF 0 IF
Figure 7.
the P-wave can’t be identified and the QRS are normal T H E N the AVN is the origin of the rhythm the P-waves lie after the QRS, the P-waves are inverted and the QRS are normal T H E N the AVN is the origin of the rhythm. the P-waves lie before the QRS, the P-waves are inverted and the P-R intervals are short T H E N the AVN is the origin of the rhythm.
If-Then rules showing the same consequent but each has a different premise. Repetition of the rules replaces logical ORing.
35
Disturbances of impulse formation T o prove NSR apply the following rules:
goal
IF SO, RR and N R T H E N NSR
rule 1
IF N P and NP-R T H E N SO
rule 2
IF NP, NP-R, RR and N R T H E N NSR
implied rule from rules 1, 2
Then the questions asked to the user would be: Are NP, N-PR, RR, and N R true?
data interrogated from the user
Downloaded by [UNSW Library] at 07:04 26 April 2016
T h e following abbreviations are used:
SO NSR RR
Sinus origin Normal sinus rhythm Regular rhythm Figure 8.
NR Normal rate NP-R Normal P-R wave NP Normal P-wave
A sample of the inference mechanism.
about the name of the ECG strip, the examined ECG component and information whether the value of the examined component is available). T h e other type of facts are those concluded by the system through the inference engine (IE), e.g. rhythmorigin-node (it contains information about the name of the ECG strip, the origin of the rhythm, and whether the examined condition is proved or disproved). At the end of every session the contents of the WDB are saved for later consultation. T h e inference engine (IE) uses the built-in goal-oriented backward-chaining mechanism available in Turbo-Prolog. T h e system starts with a goal statement and works backwards through inference rules to find the data that establish that goal. An example is given in figure 8. The inference engine contains rules for the following rhythms [28-331:
(1) Sinus rhythm ( a ) Normal sinus rhythm ( b ) Sinus rhythm disturbances (i) Sinus tachycardia (ii) Sinus bradycardia (iii) Sinus irregularity (iv) Sinus arrest (v) Respiratory sinus arrythmia (vi) Sinoatrial block
(2) Atrial rhythm disturbances ( a ) Wandering atrial pacemaker ( b ) Atrial escape rhythm ( c ) Atrial tachycardia (3) Nodal rhythm disturbances ( a ) Nodal tachycardia ( b ) Nodal escape rhythm (c) Accelerated nodal rhythm
(4) Supraventricular tachycardia
M . S . Habashi and M . A. Abdel-Bury
36
T h e user’s interface (UI) is useful when some facts are required from the user, e.g. the conditions given in figure 8. T h e user will be prompted to give the needed data, e.g. ‘Are the P-R intervals within normal?’. T h e UI will read the user’s response and check its validity. In case of errors, error messages will be given. T h e user’s response will be then stored in the WDB. T h e system has also an explanation facility (EF). At the end of every session the E F will search the W D B for the facts that support the diagnosis and gives the user a detailed explanation for the final diagnosis, as will be shown in the next section.
Downloaded by [UNSW Library] at 07:04 26 April 2016
3. Implementation 3 . 1 . Algorithm T h e algorithm for ECG interpretation is discussed in this section. A sample for normal sinus rhythm diagnosis is explained in detail. Other parts of the system work similarly. T h e system works in the following manner:
(1) Obtains the personal history of the patient (if no patient file is already available). (2) Recalls any stored data from the patient’s file to the WDB. (3) Searches the rythm tree. There are 14 rhythm types; each will lead the system to a different subtree. Common conditions are searched first, since they are the most probable ones to happen (e.g. normal sinus rhythm before sinus tachycardia). (4) T o check each one of the previous conditions the system searches for essential diagnostic subconditions. If the subconditions are not already available in the system’s WDB, the user will be prompted to enter them, User’s response is processed and added to the database. (5) At the end of the session the accumulated data are used to update the patient’s file, which is saved for later use. 3 . 2 . Example: Diagnosis of normal sinus rhythm As an illustration, the search tree for normal sinus rhythm is shown in figure 9. To prove normal sinus rhythm, either the left or the right subtree should be fulfilled. This means that the two subtrees are logically ORed. All branches of any subtree should be fulfilled to reach the indicated diagnosis, i.e. the branches of the subtree are logically ANDed. When neither subtree is proven, this implies that the diagnosis of normal sinus rhythm is discarded and another condition will be searched. T h e sequence of traversing the branches of the tree is from left to right. For the logical order of the session questionnaire, some data might be asked forin advance-irrespective of their immediate usefulness. A sample of the output of the session is shown in figure 10. and the file contents generated by the system for the corresponding patient is given in figure 11.
4. Conclusions T h e system has pros and contras which might be summarized as follows.
4.1. Limitations Not all types of rhythm disturbances are included in this phase of the system development. It could easily be expanded by adding the corresponding diagnostic rules for each disease.
37
Disturbances of impulse formation
Downloaded by [UNSW Library] at 07:04 26 April 2016
N o r m a l e i n u s rhythm
Each QRS has a
P-waves precede QRS
Upright P-waves
Each QRS has a P-wave
Sinus rhythm
Similar P-waves
K
Normal
be w e l l aeseeeed
QRS
Figure 9.
Heart r a t e = 80 120
-
Normal P-R intervals
Supraventricular rhythm
P-wave c a n ' t
Regular rhythm
Normal P-R intervalB
Regular rhythm
Search tree for normal sinus rhythm.
Heart rate= 100-120
Rhythm Diagnosis Please give the file name you want to deal with: patient 1 Please complete the following data about the patient’s personal history: name: sex:
Ali male
Birth date: year: 1960 month: 5 day: 4 marital status: single Give the name of the ECG strip to be examined
Downloaded by [UNSW Library] at 07:04 26 April 2016
v2 Can the P-waves be easily identified? Yes Does each QRS have a P-wave? yes
Do the P-waves lie before QRS? yes Are the P-waves of upright contour? yes Are the P-waves similar in shape? yes Are the P-R intervals homogeneous in length all through the strip? Y €5
Are the P-R intervals short? no
Is the P-P cycle length quite regular? yes Calculation o f P-wave rate Please count the no. of cycles in the strip; no. of cycles = 5 Please count the corresponding no. of great squares; no. of great squares = 15
Diagnosis Normal sinus rhythm at a rate of 100 is the most accepted diagnosis of V2 because: (1) T h e rhythm is probably sinus in origin due to the following causes: ( a ) T h e P-wave contour is upright (b) The P-waves are similar in shape (c) T h e P-waves do lie before QRS (d) The P-R interval is not short
(2) The calculated heart rate=100 Figure 10. (a)
Disturbances of impulse formation
39
Rhythm Diagnosis Type the name of the next strip If no further strips -P Type no
L1 Can the P-waves be easily identified? no May the shape of the T-wave that precedes the QRS suggest superimposed P-wave? Yes
Downloaded by [UNSW Library] at 07:04 26 April 2016
Can you identify or suspect a P-wave for every QRS? Yes
Is the QRS complex normal in shape? Yes Are the P-R intervals within normal? Yes
Is the P-P cycle length quite regular? Yes Calculation of QRS rate Please count the no. of cycles in the strip; no. of cycles = S Please count the corresponding no. of great squares; no. of great squares = 10 Did you try to terminate this tachycardia by vagal stimulation? if yes, mention the result: (a) is not tired (b) terminates the attack abruptly (c) fails to terminate the attack (d) terminates the attack gradually
d
Diagnosis Sinus tachycardia at a rate of 150 is the most accepted diagnosis of L1 because:
( 1 ) The rythym is probably supraventricular in origin due to the following causes: (i) the QRS is normal (ii) the T-wave contour may suggest superimposed P-wave
(2) Vagal stimulation terminates the attack gradually (3) T h e calculated heart rate = 1SO Figure 10. (b) Figure 10. A sample session for patient 1. ( a )Strip 1: normal sinus rhythm; (b) strip 2: sinus tachycardia.
M . S . Habashi and M . A . Abdel-Bary
40
rhythm,diagnosis("Ll","nlO2","yes", 150, ["n53",::ne'' "m2"]) r h y t hm-d iag nos i 8 ( " V 2 " " n 1 00" , yes " , 1 00 ,[ " n 5 0 " , m 1 j 1 I'
rhythm_origin("Ll","n53","yes", ["nl", "n6"]) rhythm,orlgin("V2","n5O","yes", ["n4","n3", "n2","n5"11
rhythm-component-node( "V2","no" ,"a",[ "yes"]) rhythm-component-node( "V2","81 "y" [ "yes"]1 rhythm-component-node( "V2","n2","a",[ "yes"3 1 rhythm~component~node~"V2","n4","a",["yes"]) rhythm-component-node( "V2","n3","a" [ "yes"3 ) rhythm-component-node( "V2","s2","y", [ "yes"]) rhythm-component-node( "V2","n5"," b " , "no"] rhyth~component~node("V2","nlO","a", ["yes"]) rhythm-component-node( "Ll ","no"," a " , [ "no"]) ' I ,
Downloaded by [UNSW Library] at 07:04 26 April 2016
rhythm,component~node("Ll","ne","b",
rhythm-component-node( rhythm-component-node( rhythm-component-node( rhythm-component-node( rhythm-component-node(
"L1
'I,
["yes"])
"s3","y" ,["yes"])
"L1 " ,"nl " , "a",[ "yes"]) "L1 "n5"," a ' * ,[ "yes"]) "L1 "nl0" ,"a",["yes"]) "L1 "na", "d" [ "yes"]1 rhythm,rate,node("V2","ml", 100,["ca culated"]) r hythm-rate-node ( L 1 , "m2" ,1 5 0 , [ ca cu 1 ated I ) rhythm-rate-node( "L1 "ml 150, [ " c a culated"]) rhythm-rate-node("Ll","m2", 150,["ca culated"] ) datum( "year",1960) datum( "month",5 ) datum( "day",4) datum-1 ( "name","Ahmed Hasean A1 1 " ) datum-1 ( " s e x " , "male") datum-1 ("marital-status" ,"single") "
'I,
"
' I ,
Figure 1 1 .
'I,
' I ,
"
"
'I,
The contents of the file generated by the ES for patient 1
T h e user response is evaluated to be either true of false. Addition of probability factors to weigh the user's response is planned to be added to give more flexibility in the diagnosis. T h e user is not able to delete or modify-on-line-any data after file saving. An interactive interface for this task can easily be added to the ES.
4.2. Advantages T h e system is user-friendly. T h e available explanation facility provides a logical trace for the diagnosis. T h e system is easily expandable. T h e system can be used as an ES shell for other diseases in the medical field. T h e system could be used as a training tool for checking the E C G interpretation. Interpreted ECG records for a patient are stored for further recall when needed. References 1 . ROWLANDS, D. J. (1981) Understanding the Electro-cardiogram: A New Approach (London: Cower Medical). 2. RICH,E. (1983) Artificial Intelligence (New York: McGraw-Hill). 3 . NILSSON, N. J . (1982) Principles of Artificial Intelligence (Berlin: Springer-Verlag). 4. WINSTON, P. H. (1984) Artificial Intelligence, 2nd edn (Reading: Addison-Wesley). 5. WATERMAN, D. A. (1986) A guide to Expert Systems (Menlo Park: Addison-Wesley). 6. ANDRIOLE, S. J . (1985) Applications in Artificial Intelligence (Englewood Cliffs: Prentice-Hall. 7. JACKSON, P. (1986) Introduction to Expert Systems (Reading: Addison-Wesley).
Downloaded by [UNSW Library] at 07:04 26 April 2016
Disturbances of impulse formation
41
8. FRIES,J. F. (1972) Time-oriented patient records and a computer data-bank. Journal ofthe American Medical Association, 222 (12), 1536-1542. 9. BUCHANAN, B. G., and SHORTCLIFFE, E. H . (1984) Rule-Based Expert Systems. The M Y C Z N Experiment of the Stanford Heuristic Programming Project (Reading: Addison-Wesley). S. G., BUCHANAN, B. G., MERIGAN, T . C., and COHEN,S. N. (1973) An 10. SHORTLIFFE, E. H., ALIXINE, artificial intelligence program to advise physicians regarding antimicrobial therapy. Computers and Biomedical Research, 6, 544-560. 11. SILVERMAN, H . (1974) A digitalis therapy advisor. MZT Technical Report T R - 1 4 3 , December. W. R. (1977) A digitalis therapy advisor with explanations. M Z T Technical Report T R 12. SWARTOUT, 176, February. F., MOLLER. J . , and PRIETULA, M . (1981) Expertiseand 13. JOHNSON, P. E., DURAN, A. S., HASSERBROCK, error in diagnostic reasoning, Cognitive Science, 5, 235-283. 14. THOMPSON, W. B., JONSON, P. E., and MOEN.J. B. (1983) Recognition-based diagnostic reasoning. Proceedings ZJCAZ, 83. 236-238. 15. NIEMANN, H., BUNKE, H., HOFMANN, I., and SAGERER, G. (1984) Diagnostic inferences from image sequences-a knowledge based approach. Proceedings of the First Conference on Artificial Intelligence Applications (New York: IEEE Computer Society). V. A,, KOVRIGIN, 0. V., and SMOL’YANINOV, N. D. (1982) An expert for 16. BOKHUA, N. K., GELOVANI, diagnostic of various forms of atrial hypertension, Tekh. Kibern. [Engineering Cybernetics], 20(6), 106-111. 17. MILLER,P. L., and BLACK,H. R. (1984) Medical plan-analysis by computer: Critiquing the pharmacologic management of essential hypertension. Computers and Biomedical Research, 17,38-54. S., KOYAMA, T., MINAMIKAWA, T . , and YASAKA, T. A. (1978) A rule-based physicians’ 18. KAIHARA, consultation system for cardiovascular diseases. Proceedings of the International Conference on Cybernetics and Society (New York: IEEE), 85-88. 19. KAIHARA, S., and KOYAMA, T. (1982) Medical consultation system with practical requirementsdevelopment of MECS-AI. Computer Science and Technologies (Japan Annual Review in Electronics, Computer and Telecommunications), vol. 4 (Amsterdam: North-Holland). C. A. (1983) An efficient scheme for time-dependent 20. KASTNER, J . K., WESS,S. M., and KULIKOWSKI, consultation systems. Proceedings MEDZNFO-83, Fourth World Conference on Medical Informatics (Amsterdam: North-Holland), 619-622. 21. PUPPE,F., and PUPPE,B. (1983) Overview on MEDI: A Heuristic diagnosis system with an efficient control structure. Report SEKZ-83-02 (Universitaet Kaiserslautern, West Germany: Fachbereich Informatik). 22. PAUKER, S. G., GORRY, G . A,, KASSIRER, J. P., and SCHWARTZ, W. B. (1976) Towards the stimulation of clinical cognition: taking a present illness by computer. American Journal ofMedicine, 60,981-996. 23. KULIKOWSKI, C. A. (1980) Artificial intelligence methods and systems for medical consultation. ZEEE Transactions on Pattern Analysis and Machine Intelligence, 2(S), 464476. 24. REGGIA,J. A., NAU,D. S., and WANG,P. Y. (1983) Diagnostic expert systems based on a set converting model. Znternationat Journal of Machine Studies, 19, 437460. E. A. (eds) (1982) TheHandbookof Artificialfntelligence, 25. BARR, A,, COHEN,P. R., and FEIGENBAUM. 3 vols (London: Pitman). 26. KOWALSKI, R. (1979) Logic for Problem Solving (New York: American Elsevier. 27. KOWALSKI, R. (1979) Algorithm = Logic Control. A C M Communications, 22(7), July. 28. BRAUNWALD, E. (1980) Heart Disease (Philadelphia: W. B. Saunders). 29. CORDAY, E., and IRVING,D . W. (1961) Disturbances of Heart Rate, Rhythm and Conduction (Philadelphia: W. B. Saunders). 30. FLEMING, J. S. (1979) Introduction to the Electro-cardiogram (New Jersey: Update). 31. HURST,J. W., and MYEBURG, R. J. (1975) Introduction to Electrocardiography, 2nd edn (New York: McGraw-Hill). 32. PHILLIPS,R. E., and FEENEY,M. K. (1980) The Cardiac Rhythm, A Systematic Approach to Interpretation, 2nd edn (Philadelphia: W. B. Saunders). 33. SCHERE, D., and COHEN,J . (1 968) Atrioventricular rhythm, Progress in Cardiovascular Disease. 34. WEISS,S. M., and KUI.IKOWSKI, C. A. (1 983) A Practical Guide to Designing Expert Systems (London: Chapman & Hall). 35. WIT,A.L., WEISS,M. B., BERKOWITZ, W. D., ROSEN,K. M., STEINER, C.,and DAMATO,A. N. (1970) Patterns of atrioventricular conduction in human heart, Circulation Research, 27 (3), 345-359.
+
