- Email: [email protected]
Arrhythmia diagnosis by the IBM electrocardiogram analysis program
J. ELECTROCARDIOLOGY 13 (1), 1980, 17-24
Arrhythmia Diagnosis by the IBM Electrocardiogram Analysis Program BY HIDEO MIYAHARA, M.D.,* KYOKO ENDOU, M.D., AKIRA DOMAE, M.D.t a n d TOSHIRO SATO, M.D**
SUMMARY A comparative study was made between r h y t h m diagnosis by computer using the IBM E C G program (Bonner-I), and diagnosis by a physician. 2434 electrocardiograms (ECGs) recorded in Kitasato University Hospital from July to August 1978 were used. Of the 2434 cases the physician made a diagnosis o f sinus r h y t h m in 2185 and of abnormal (other t h a n sinus) r h y t h m in 249. Among the former, 32 cases were erroneously classified into abnormal groups by the computer (false positive), and in 30 out o f the latter 249 cases, the computer failed to detect abnormalities (false negative). In the remaining 219 cases, 33 o f 57 cases o f sinus r h y t h m with VPCs, 19 o f 41 sinus r h y t h m with SVPCs, and 62 o f 64 atrial fibrillation were correctly diagnosed by the computer. In 27 out o f the 219 cases "undetermined r h y t h m " was printed out after a n unsuccessful dominant r h y t h m determination by the computer. Cases o f artificial pacemaker were not recognized by the computer, but were distributed into 7 classification categories. We were o f the opinion t h a t ff the computer program had utilized the information on the duration o f QRS complexes these cases could have been correctly classified. We repeated E C G sampling and analysis in the cases o f complicated arrhythmias, because more valuable information was often supplied by the computer after successive E C G recording. F r o m these results we concluded that the usefulness o f the IBM E C G program was c o n f m n e d concerning the recognition of sinus rhythm, whereas the detection and identification of complex r h y t h m disturbances by the computer program was still in need of improvement. In the computer diagnosis of the conventional twelve lead ECG, the reliability of a r r h y t h m i a diagnosis has been considered to be inferior to t h a t of contour diagnosis, because detection and location of abnormal P waves are still unsatisfactory and the length of record is too short for the analysis of complicated a r r h y t h m i a s . TM We have been using MEPC, a n a u t o m a t e d E C G diagnostic system with the IBM program ~ since September 1977. This report reviews the current status of r h y t h m diagnosis by the c o m p u t e r based on our experience of the MEPC system.
institution of Kitasato University Hospital September, 1977. It consists of two automatic electrocardiograph terminals (Healthtec HT-300), Data receiver, CPU (Interdata 7/16 mini-computer with 10 MB disc, console keyboard and character display) and 2 terminal printers (Extel S/N). The terminal carts and the receiver unit are connected by a cable about 70 meters in length. The IBM ECG program (Bonner-I) is used for measurement and interpretation of ECG signals. The capacity of the system has prevented print-out of the full reference data: only intervals, axes and measurement matrix of ECG parameters are available for detailed analysis. Directly recorded tracings are available at both the terminal carts and the direct writer in the computer room. Physicians were allowed to examine the computer print-outs before making a final judgment. The comparison was confmed to the rhythm distfirbance and excluded the contour information unless this was essential for the arrhythmia diagnosis. From July to August 1978, ECGs of 2963 patients were examined by the computerized electrocardiography system. Among them were 529 taken from patients under 15 years of age and excluded from this study because of the program restriction. ECGs taken from the remaining 2434 patients were used for this study.
MATERIALS AND METHODS The MEPC (Multi-user Electrocardiogram Processing Center) system manufactured by Telemed Corp. was installed in Health Science Center, an affiliated
From the Dept. of Internal Medicine, School of Medicine, Kitasato University, 1, Asamizodai, Sagamihara, Kanagawa, Japan 228 *Associate Professorof Medicine tVisiting Professorof Medicine **Professorof Medicine The costsof publicationofthis article were defrayedin part by the payment of page charges. This article must therefore be hereby marked ~advert/sement"in accordancewith 18 U.S.C.w 1734 solelyto indicate this fact. Reprint requests to: Hideo Miyahara, M.D., Dept. of Internal Medicine, School of Medicine, Kitasato University, 1, Asamizedai, Sagamihara, Kanagawa, Japan 228
RESULTS Sinus r h y t h m . O f 2434 cases which were processed by the M E P C system, 2183 were d i a g n o s e d b y t h e c o m p u t e r as e i t h e r t r u e n o r m a l sinus r h y t h m ( h e a r t r a t e b e t w e e n 60 17
18
MIYAHARA ET AL
Fig. 1. Ventricular premature contraction (VPC) overlooked by the computer. Although a typical VPC was found at the first part of V1-V3 lead set, it could not be identified by the computer (arrows indicate the moment of lead set switching). and 100), true sinus bradycardia (less t h a n 60) or true sinus tachycardia (more t h a n 100). We will use the general term "sinus r h y t h m " for these three kinds of r h y t h m s with no other abnormalities. Some cases were incorrectly diagnosed as sinus r h y t h m by the computer. Three cases of ventricular premature contractions (VPCs) and six cases of supraventricular premature contractions (SVPCs) were missed by the computer because the ectopic events happened to appear just before or after the s w i t c h i n g m o m e n t f r o m one l e a d s e t to another (Fig. 1). In addition to these, ten cases of occasional SVPCs, three cases of VPCs (two with fully compensatory pause, one interpolated) and five cases of wandering pacemaker were missed by the computer (Table 1). Low atrial or junctional pacemaker. The IBM program criteria for the low atrial or junctional pacemaker (LAR) are: each QRS complex driven by a P wave with an abnormal angle or short P-R interval; no extra P waves; all P waves with about the same P-R interval; and heart rate less t h a n 100 beats per minute. Thirty-one cases were classified into this category by the computer. In 27 of them, including two cases of left atrial r h y t h m , the physician agreed with the computer in the diagnosis of LAR. Of the r e m a i n i n g four cases, the physician made a diagnosis of sinus r h y t h m in three cases and sinus r h y t h m with VPCs in one case. On the other hand, the
computer correctly made a diagnosis of LAR with VPCs in one case, but failed to detect LAR with SVPCs in one case. Eetopic premature contractions. Thirty-six cases of sinus r h y t h m with VPCs, 35 cases of sinus r h y t h m with SVPCs, ten cases of sinus rhythm with both VPCs and SVPCs (VPCs+SVPCs) and one case of LAR with VPCs were detected by the computer. The physician agreed with the computer in the diagnosis of 33 of the 36 cases of sinus r h y t h m with VPCs. The remaining three cases were determined to be true sinus r h y t h m , wandering pacemaker or SVPCs. In 19 of 35 cases of the SVPCs, the physician agreed with the computer i n t h e diagnosis; but in the remaining 16, he made the diagnosis of sinus arr h y t h m i a in eight, VPCs in four, wandering pacemaker in two, VPCs+ SVPCs in one, and atrial r h y t h m with SVPCs in one. In only three of the ten cases of VPCs+ SVPCs, did the physician agree with the computer in the diagnosis. Six cases were diagnosed as VPCs and the r e m a i n i n g case was diagnosed as SVPCs. We summarized the comparison of diagnoses of other abnormalities in Tables 2 and 3. Those cases bearing such symbols as *, + and ~ belonged to more t h a n one category. Atrial fibrillation. Seventy-one patients were diagnosed as cases of atrial fibrillation by the computer. Sixty-two of t h e m were given the same diagnosis by the physician, J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
A R R H Y T H M I A D I A G N O S I S BY C O M P U T E R
~
19
TABLE 1 False negatives and false positives in sinus rhythm detection. iagnosis
ComputerDiagn~s ~ Sinus Rhythm (SR)
True
LAR
WP
SR
SR
SR
+
+
Others
VPC SVPC
Total of Computer Diagnosis
2153
0
5
6
16
3
2183
1575
0
3
6
10
0
1594
Sinus Bradycardia
524
0
2
0
5
3
534
Sinus Tachycardia
54
0
0
0
1
0
55
Low Atrial or Junctional
3
27
0
1
0
0
31
28
0
3
50
25
114
220
2184 27
8
57
41
117
2434
Normal Sinus Rhythm
Pacemaker (LAR) Other Abnormalities Total of Physician's Diagnosis
WP: Wandering Pacemaker VPC: Ventricular Premature Contractions SVPC: Supraventricular Premature Contractions
but the remaining nine cases were classified into other categories; they were marked sinus bradycardia (with junctional escape; 1, with first degree AV block; 2, with blocked SVPCs; 1), sinus r h y t h m with f r e q u e n t SVPCs (2 cases), short run of supraventricular tachyc a r d i a (1), s e c o n d d e g r e e AV b l o c k of Wenckebach type (1) and artificial pacemaker (1). Other complex arrhythmias. The computer succeeded in making the diagnoses of first degree AV block for a long P-R interval of 0.33 see and for a ease in which the beginning of the P wave was located close to the end of the preceding T wave. On the other hand, it failed in a case of the low amplitude P wave with first degree AV block with 0.42 sec P-R intervals (Fig. 2). Two cases of junctional rhythm, four of junctional tachycardia, three of atrial flutter, and one of second degree AV block were printed out as the results of computer diagnosis. One case of complete AV block was diagnosed by the computer. In this case, an artificial p a c e m a k e r was d o m i n a t i n g t h e rhythms, whereas regular non-conductive P waves were observed all over the tracing. The IBM r h y t h m program has a classification category for a suspect AV conduction defect, which is identified when atrial and ventricular rates differ significantly, with the atrial rate higher than the ventricular rate and when no other AV conduction defect is J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
detected. In our study, three cases were classified into this category; two were actually artificial pacemaker and one was sinus rhythm with frequent SVPCs. The computer gave a diagnosis of suspect ventricular pacemaker when a given ECG tracing showed an almost constant heart rate; no P waves driving QRS complexes; QRS width more than 120 ms; and heart rate between 60 and 120 beats per minute. All of the eight cases in this category were artificial pacemaker rhythm. If the basic r h y t h m cannot be determined by the program, the computer will print out the tentative diagnosis of undetermined rhythm and leave the final decision to the physician. In this study, such a statement was made in 27 cases. For these cases, the physician's diagnoses were true sinus r h y t h m (6 cases), atrial fibrillation (2), bigeminy due to VPCs (3), freq u e n t S V P C s or V P C s (3), a r t i f i c i a l pacemaker (11) and W P W syndrome (2). One case of reciprocal beats with junctional escape was also classified by the computer into this category. In addition to this case, the physician could detect junctional escape beats in eight E C G tracings, whereas the computer could successfully detect t h e m only in one case. If the input data was considered unsuitable for detailed analysis, the IBM program gave a diagnosis of r inconsistency" and did not produce rhythm interpretation. Ten cases were classified into this
20
MIYAHARA
ET AL
o
o.
.I
E II
.5
i~1 n r ,
~++
Lid
0.
.I +
N
~
,, .,,,
.,,, ~ ~ ~
= ~'
E ~
~o 5=
~ ,~
== ~
~ +
+
=~
C
-
,2,
""
i:
~,
<_.-,~
.~
< +
==,,'I,,-~ ,-,-
~,-,
,.,. , . , _ ++ - ~ m - , -
"
"
J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
ARRHYTHMIA
DIAGNOSIS
BY COMPUTER
21
O
0
._ ~= 1,1.1
-,=~
*
*
I
E
.~_
+~"
~'~
+~
v
"8
.w
0
o
t~
E
E
r
/
9-
~-,
~
=_
=
_
J, E L E C T R O C A R D I O L O G Y ,
•~
VOL
13, N O .
1, 1 9 8 0
~
-
. .~
~
o
o
,,o
=~ o
o
~
,-,~
._
.~v N
~
N
.,.-,
~_ o
~= ~
"
22
M I Y A H A R A ET AL
TABLE 4. Reproducibility of diagnosis of serial recordings Computer Diagnosis First
No.
1 Atrial Fibrillation 2 3 4 5 6
Atrial Atrial Atrial Atrial Atrial
Fibrillation Flutter-fibrillation Flutter-fibrillation Fibrillation Fibrillation
7 8 9
Atrial Fibrillation Atrial Fibrillation Undetermined Rhythm
10 11
Undetermined Rhythm Undetermined Rhythm
12
Undetermined Rhythm
13
Undetermined Rhythm
14 Undetermined Rhythm 15 Suspect Ventricular Pacemaker 16 Normal Sinus Rhythm 17 Probably SB, rule out Heart Block 18 SB + Sinus Arrest + Junct. Escape 19 Low Heart Rate, suspect Heart Block 20 Complete Heart Block 21
Junctional Tachycardia
Physician's Diagnosis
Second Atrial Fibrillation Atrial Fibrillation Atrial Flutter-fibrillation Atrial Fibrillation Atrial Fibrillation Sinus Bradycardia Undetermined Rhythm Undetermined Rhythm Atrial Fibrillation + VPCs Undetermined Rhythm Low Atrial or Junctional Pacemaker Suspect Ventricular Pacemaker Normal Sinus Rhythm + VPCs Sinus Bradycardia Undetermined Rhythm
Atrial Fibrillation Atrial Fibrillation Atriat Fibrillation Atrial Fibrillation Sinus Bradycardia (SB) + SVPCs Sinus Arrhythmia + Junctional Escape Demand Pacemaker Sinus Rhythm + frequent SVPCs Atrial Fibrillation WPW Syndrome Left Atrial Rhythm + VPCs Demand Pacemaker Bigeminy due to VPCs Sinus Bradycardia Electronic Pacemaker
Undetermined Rhythm Probably SB, rule out Heart Block SB + Sinus Arrest
Normal Sinus Rhythm Sinus Bradycardia + Junctional Escape S-A Block + Junctional Escape
Probably SB, rule out Heart Block Suspect Ventricular Pacemaker Junctional Tachycardia
Sinus Bradycardia
category. Five of them were sinus rhythm, whereas the remaining five were artificial pacemaker rhythm. Rhythms classified as wandering pacemaker by the physicians were expected to be interpreted by the program as ~changing P contour." None of the nine records in this category were correctly detected by the program. Although the IBM program includes a logic for d e t e c t i n g electronic pacemaker and demand pacemakers, none of the 29 artificial pacemaker rhythms identified by the physicians were properly detected and classified into these two categories. Of five cases of WPW syndrome, two cases were detected by the computer, whereas one case was incorrectly diagnosed as sinus
Electronic Pacemaker Normal Sinus Rhythm
rhythm with VPCs+ SVPCs, and the remaining two cases were classified into undetermined rhythm without any suggestion concerning WPW syndrome.
DISCUSSION Since our patient population consists to a large extent of subjects with normal sinus rhythm, a substantial saving in electrocardiographer's time could be achieved if the computer program could accurately separate this category from categories with abnormal rhythms. The IBM program falsely classified only 30 records w i t h a b n o r m a l r h y t h m as sinus J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
ARRHYTHMIA DIAGNOSIS BY COMPUTER
23
Fig. 2. Sinus bradycardia with first degree AV block mistaken for atrial fibrillation by the computer. Low P waves could be detected by the physician, but not by the computer.
rhythm. In 27 of the 30 instances the misclassification took place in categories which are of little or no clinical significance, such as infrequent ectopic complexes and wandering pacemaker. In addition, the number of cases which actually belonged to the sinus r h y t h m category but were incorrectly regarded as abnormal was 32. The low incidence of false positive cases is worthy of note, because it could save the physician's time and labor to re-examine normal ECGs misclassified as abnormal. The high sensitivity (2153/2185 = 98.5%) and specificity (219/249 = 88.0%) of sinus r h y t h m detection was considered to prove the usefulness of the IBM program for our present purpose. The current IBM program could not recognize ectopic events if they appeared at the m o m e n t j u s t before or a f t e r t h e lead set switching. This fact was the major cause of errors in the detection of ectopic events. Another cause of error was misrecognition of SVPCs in sinus r h y t h m as sinus arrhythmia. Because of small sample size of interval data available, they could not be clearly separated. Longer continuous recording exclusively using a single channel might afford a larger sample of interval data. Moreover, it might give more information even in the neighborhood of lead set switching and allow us to obtain higher specificity. On the other hand, the causes of false posiJ. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
tive detection vary. In nine cases, the physician could not find any ectopic events on the t r a c i n g recorded by the t e r m i n a l cart, although the computer detected either SVPCs (eight cases) or VPCs (one case). It was considered t h a t the most probable cause of incompatibility between the physicians' and computer's interpretation was the difficulty of separating SVPCs from sinus a r r h y t h m i a in borderline cases. Relatively low incidence of unidentified ectopic events showed t h a t the direct connection between carts and the central processor successfully decreased the noise during transmission. Discrepancies among P wave recognitions for individual lead sets was another cause of false positive detection and misclassification of true sinus r h y t h m into undetermined r h y t h m (UDR). Improvement of the accuracy of P wave identification would decrease these mistakes. Analysis of the 27 cases classified into the UDR category revealed t h a t an a r r h y t h m i a for which the IBM program did not prepare the proper diagnostic category was observed in only one case. The remaining 26 cases were erroneously diagnosed as UDR, though they should have been classified into other categories (Table 3). Consequently, it was considered t h a t an effective procedure to enhance the utility of the ECG program would be to lessen the number of cases in which the r h y t h m could not be determined. For example, 11 cases of artificial
24
MIYAHARA ET AL
pacemaker r h y t h m were diagnosed as UDR. An artificial pacemaker r h y t h m could be easily detected on an ECG chart by the physician from the characteristic pattern of sharp and tall spikes locating at the beginning of broad QRS complexes. The computer program was also designed to m a k e a diagnosis of the pacemaker r h y t h m by means of the spike artifact. However, the c u r r e n t computerized ECG system usually set a high-cut filter in a terminal cart and limited the band-width of the ECG signal. This device eliminated high frequency noise, but at the same time blunted the spike artifact so t h a t the computer program failed to identify the pacemaker r h y t h m and incorrectly classified 11 cases into UDR. As a tentative measure, before completing an advanced version of the program, we would propose alternative criteria to diagnose the cases of UDR which had wide QRS duration (160 ms or more) as artificial p a c e m a k e r rhythm. Pordy 6 was reluctant to use the findings of a regular heart rate, absent P waves and wide QRS complex as a criteria of electronic pacemaker, because an AV junctional pacemaker with IV block could be classified into this category. It was true, but use of the duration of QRS in UDR seemed to be effective in our hospital for the separation of artificial pacemaker from other abnormalities, because the incidence of electronic pacemaker was far greater t h a n t h a t of the junctional pacemaker with IV block. In fact, 11 cases of artificial pacemaker could be separated from UDR by this criteria, whereas only one case of UDR with wide QRS duration was erroneously diagnosed as artificial pacemaker. We repeated the ECG e x a m i n a t i o n consecutively when a technician found t h a t the computer gave the diagnosis of atrial fibrillation, h e a r t block, suspect of v e n t r i c u l a r pacemaker or undetermined rhythm. In Table 4, we summarized the comparison among the first and second computer diagnosis and the physician's. It was revealed t h a t at least one of the two computer diagnoses agreed with physician's in 11 of 21 cases. We noticed t h a t in five of six cases of undetermined rhythm, the second test gave a more meaningful diagnosis t h a n the first. These observations can be
at least partially explained by the fact t h a t the input signals in two successive ECG records are not identical even though the basic r h y t h m r e m a i n s u n c h a n g e d . It m i g h t be v a l u a b l e to t a k e ECGs r e p e a t e d l y if t h e d i a g n o s i s of u n d e t e r m i n e d r h y t h m w a s printed out. Since information regarding the accuracy of the computer program, especially as to the diagnosis of arrhythmias, is not yet available, cardiologists responsible for the performance of a computerized ECG system have to overread the ECG record of the computer r h y t h m diagnosis (even in the case of an output of simple sinus rhythm). It m i g h t be helpful to accumulate those cases misdiagnosed by the computer for the purpose of analyzing the detail and feeding the r e l e v a n t information back to the distributor of the original program. A special committee consisting of computer engineers, system users, relevant cardiologists, etc. could cooperate in analyzing questionable cases and discuss sending advice to the program distributors. If such a system were established, elaboration of the diagnostic logic would be promoted and more reliable a r r h y t h m i a diagnosis by the computer could be expected.
REFERENCES 1. WILLEMS,J L AND PIPBERGER,H V: Arrhythmia detection by digital computer. Comp Biomed Res 5:263, 1972 2. LEBLANC,A R AND ROBERGE,F A: III. Present state of arrhythmia analysis by computer. Can Med Assoc J 108:1239, 1973 3. MEYER, J, MERX, W, HEINRICH, K W AND EFFERT, S: Computer-analyse des Elektrokardiograms mit verschiedenen Programmen II. Rhythmusanalyse, Dtsch med Wschr 99:1294, 1974 4. CACERES,C A: Limitations of the Computer in electrocardiographic interpretation, Am J Cardiol 38:362, 1976 5. Health Care Support/Electrocardiogram (ECG) Analysis Program, Physician's Guide, IBM Corp., 1974 6. PORDY, L: Computer Electrocardiography: Present Status and Criteria, Futura Publ. Co., New York, 1977
J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
Arrhythmia Diagnosis by the IBM Electrocardiogram Analysis Program BY HIDEO MIYAHARA, M.D.,* KYOKO ENDOU, M.D., AKIRA DOMAE, M.D.t a n d TOSHIRO SATO, M.D**
SUMMARY A comparative study was made between r h y t h m diagnosis by computer using the IBM E C G program (Bonner-I), and diagnosis by a physician. 2434 electrocardiograms (ECGs) recorded in Kitasato University Hospital from July to August 1978 were used. Of the 2434 cases the physician made a diagnosis o f sinus r h y t h m in 2185 and of abnormal (other t h a n sinus) r h y t h m in 249. Among the former, 32 cases were erroneously classified into abnormal groups by the computer (false positive), and in 30 out o f the latter 249 cases, the computer failed to detect abnormalities (false negative). In the remaining 219 cases, 33 o f 57 cases o f sinus r h y t h m with VPCs, 19 o f 41 sinus r h y t h m with SVPCs, and 62 o f 64 atrial fibrillation were correctly diagnosed by the computer. In 27 out o f the 219 cases "undetermined r h y t h m " was printed out after a n unsuccessful dominant r h y t h m determination by the computer. Cases o f artificial pacemaker were not recognized by the computer, but were distributed into 7 classification categories. We were o f the opinion t h a t ff the computer program had utilized the information on the duration o f QRS complexes these cases could have been correctly classified. We repeated E C G sampling and analysis in the cases o f complicated arrhythmias, because more valuable information was often supplied by the computer after successive E C G recording. F r o m these results we concluded that the usefulness o f the IBM E C G program was c o n f m n e d concerning the recognition of sinus rhythm, whereas the detection and identification of complex r h y t h m disturbances by the computer program was still in need of improvement. In the computer diagnosis of the conventional twelve lead ECG, the reliability of a r r h y t h m i a diagnosis has been considered to be inferior to t h a t of contour diagnosis, because detection and location of abnormal P waves are still unsatisfactory and the length of record is too short for the analysis of complicated a r r h y t h m i a s . TM We have been using MEPC, a n a u t o m a t e d E C G diagnostic system with the IBM program ~ since September 1977. This report reviews the current status of r h y t h m diagnosis by the c o m p u t e r based on our experience of the MEPC system.
institution of Kitasato University Hospital September, 1977. It consists of two automatic electrocardiograph terminals (Healthtec HT-300), Data receiver, CPU (Interdata 7/16 mini-computer with 10 MB disc, console keyboard and character display) and 2 terminal printers (Extel S/N). The terminal carts and the receiver unit are connected by a cable about 70 meters in length. The IBM ECG program (Bonner-I) is used for measurement and interpretation of ECG signals. The capacity of the system has prevented print-out of the full reference data: only intervals, axes and measurement matrix of ECG parameters are available for detailed analysis. Directly recorded tracings are available at both the terminal carts and the direct writer in the computer room. Physicians were allowed to examine the computer print-outs before making a final judgment. The comparison was confmed to the rhythm distfirbance and excluded the contour information unless this was essential for the arrhythmia diagnosis. From July to August 1978, ECGs of 2963 patients were examined by the computerized electrocardiography system. Among them were 529 taken from patients under 15 years of age and excluded from this study because of the program restriction. ECGs taken from the remaining 2434 patients were used for this study.
MATERIALS AND METHODS The MEPC (Multi-user Electrocardiogram Processing Center) system manufactured by Telemed Corp. was installed in Health Science Center, an affiliated
From the Dept. of Internal Medicine, School of Medicine, Kitasato University, 1, Asamizodai, Sagamihara, Kanagawa, Japan 228 *Associate Professorof Medicine tVisiting Professorof Medicine **Professorof Medicine The costsof publicationofthis article were defrayedin part by the payment of page charges. This article must therefore be hereby marked ~advert/sement"in accordancewith 18 U.S.C.w 1734 solelyto indicate this fact. Reprint requests to: Hideo Miyahara, M.D., Dept. of Internal Medicine, School of Medicine, Kitasato University, 1, Asamizedai, Sagamihara, Kanagawa, Japan 228
RESULTS Sinus r h y t h m . O f 2434 cases which were processed by the M E P C system, 2183 were d i a g n o s e d b y t h e c o m p u t e r as e i t h e r t r u e n o r m a l sinus r h y t h m ( h e a r t r a t e b e t w e e n 60 17
18
MIYAHARA ET AL
Fig. 1. Ventricular premature contraction (VPC) overlooked by the computer. Although a typical VPC was found at the first part of V1-V3 lead set, it could not be identified by the computer (arrows indicate the moment of lead set switching). and 100), true sinus bradycardia (less t h a n 60) or true sinus tachycardia (more t h a n 100). We will use the general term "sinus r h y t h m " for these three kinds of r h y t h m s with no other abnormalities. Some cases were incorrectly diagnosed as sinus r h y t h m by the computer. Three cases of ventricular premature contractions (VPCs) and six cases of supraventricular premature contractions (SVPCs) were missed by the computer because the ectopic events happened to appear just before or after the s w i t c h i n g m o m e n t f r o m one l e a d s e t to another (Fig. 1). In addition to these, ten cases of occasional SVPCs, three cases of VPCs (two with fully compensatory pause, one interpolated) and five cases of wandering pacemaker were missed by the computer (Table 1). Low atrial or junctional pacemaker. The IBM program criteria for the low atrial or junctional pacemaker (LAR) are: each QRS complex driven by a P wave with an abnormal angle or short P-R interval; no extra P waves; all P waves with about the same P-R interval; and heart rate less t h a n 100 beats per minute. Thirty-one cases were classified into this category by the computer. In 27 of them, including two cases of left atrial r h y t h m , the physician agreed with the computer in the diagnosis of LAR. Of the r e m a i n i n g four cases, the physician made a diagnosis of sinus r h y t h m in three cases and sinus r h y t h m with VPCs in one case. On the other hand, the
computer correctly made a diagnosis of LAR with VPCs in one case, but failed to detect LAR with SVPCs in one case. Eetopic premature contractions. Thirty-six cases of sinus r h y t h m with VPCs, 35 cases of sinus r h y t h m with SVPCs, ten cases of sinus rhythm with both VPCs and SVPCs (VPCs+SVPCs) and one case of LAR with VPCs were detected by the computer. The physician agreed with the computer in the diagnosis of 33 of the 36 cases of sinus r h y t h m with VPCs. The remaining three cases were determined to be true sinus r h y t h m , wandering pacemaker or SVPCs. In 19 of 35 cases of the SVPCs, the physician agreed with the computer i n t h e diagnosis; but in the remaining 16, he made the diagnosis of sinus arr h y t h m i a in eight, VPCs in four, wandering pacemaker in two, VPCs+ SVPCs in one, and atrial r h y t h m with SVPCs in one. In only three of the ten cases of VPCs+ SVPCs, did the physician agree with the computer in the diagnosis. Six cases were diagnosed as VPCs and the r e m a i n i n g case was diagnosed as SVPCs. We summarized the comparison of diagnoses of other abnormalities in Tables 2 and 3. Those cases bearing such symbols as *, + and ~ belonged to more t h a n one category. Atrial fibrillation. Seventy-one patients were diagnosed as cases of atrial fibrillation by the computer. Sixty-two of t h e m were given the same diagnosis by the physician, J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
A R R H Y T H M I A D I A G N O S I S BY C O M P U T E R
~
19
TABLE 1 False negatives and false positives in sinus rhythm detection. iagnosis
ComputerDiagn~s ~ Sinus Rhythm (SR)
True
LAR
WP
SR
SR
SR
+
+
Others
VPC SVPC
Total of Computer Diagnosis
2153
0
5
6
16
3
2183
1575
0
3
6
10
0
1594
Sinus Bradycardia
524
0
2
0
5
3
534
Sinus Tachycardia
54
0
0
0
1
0
55
Low Atrial or Junctional
3
27
0
1
0
0
31
28
0
3
50
25
114
220
2184 27
8
57
41
117
2434
Normal Sinus Rhythm
Pacemaker (LAR) Other Abnormalities Total of Physician's Diagnosis
WP: Wandering Pacemaker VPC: Ventricular Premature Contractions SVPC: Supraventricular Premature Contractions
but the remaining nine cases were classified into other categories; they were marked sinus bradycardia (with junctional escape; 1, with first degree AV block; 2, with blocked SVPCs; 1), sinus r h y t h m with f r e q u e n t SVPCs (2 cases), short run of supraventricular tachyc a r d i a (1), s e c o n d d e g r e e AV b l o c k of Wenckebach type (1) and artificial pacemaker (1). Other complex arrhythmias. The computer succeeded in making the diagnoses of first degree AV block for a long P-R interval of 0.33 see and for a ease in which the beginning of the P wave was located close to the end of the preceding T wave. On the other hand, it failed in a case of the low amplitude P wave with first degree AV block with 0.42 sec P-R intervals (Fig. 2). Two cases of junctional rhythm, four of junctional tachycardia, three of atrial flutter, and one of second degree AV block were printed out as the results of computer diagnosis. One case of complete AV block was diagnosed by the computer. In this case, an artificial p a c e m a k e r was d o m i n a t i n g t h e rhythms, whereas regular non-conductive P waves were observed all over the tracing. The IBM r h y t h m program has a classification category for a suspect AV conduction defect, which is identified when atrial and ventricular rates differ significantly, with the atrial rate higher than the ventricular rate and when no other AV conduction defect is J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
detected. In our study, three cases were classified into this category; two were actually artificial pacemaker and one was sinus rhythm with frequent SVPCs. The computer gave a diagnosis of suspect ventricular pacemaker when a given ECG tracing showed an almost constant heart rate; no P waves driving QRS complexes; QRS width more than 120 ms; and heart rate between 60 and 120 beats per minute. All of the eight cases in this category were artificial pacemaker rhythm. If the basic r h y t h m cannot be determined by the program, the computer will print out the tentative diagnosis of undetermined rhythm and leave the final decision to the physician. In this study, such a statement was made in 27 cases. For these cases, the physician's diagnoses were true sinus r h y t h m (6 cases), atrial fibrillation (2), bigeminy due to VPCs (3), freq u e n t S V P C s or V P C s (3), a r t i f i c i a l pacemaker (11) and W P W syndrome (2). One case of reciprocal beats with junctional escape was also classified by the computer into this category. In addition to this case, the physician could detect junctional escape beats in eight E C G tracings, whereas the computer could successfully detect t h e m only in one case. If the input data was considered unsuitable for detailed analysis, the IBM program gave a diagnosis of r inconsistency" and did not produce rhythm interpretation. Ten cases were classified into this
20
MIYAHARA
ET AL
o
o.
.I
E II
.5
i~1 n r ,
~++
Lid
0.
.I +
N
~
,, .,,,
.,,, ~ ~ ~
= ~'
E ~
~o 5=
~ ,~
== ~
~ +
+
=~
C
-
,2,
""
i:
~,
<_.-,~
.~
< +
==,,'I,,-~ ,-,-
~,-,
,.,. , . , _ ++ - ~ m - , -
"
"
J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
ARRHYTHMIA
DIAGNOSIS
BY COMPUTER
21
O
0
._ ~= 1,1.1
-,=~
*
*
I
E
.~_
+~"
~'~
+~
v
"8
.w
0
o
t~
E
E
r
/
9-
~-,
~
=_
=
_
J, E L E C T R O C A R D I O L O G Y ,
•~
VOL
13, N O .
1, 1 9 8 0
~
-
. .~
~
o
o
,,o
=~ o
o
~
,-,~
._
.~v N
~
N
.,.-,
~_ o
~= ~
"
22
M I Y A H A R A ET AL
TABLE 4. Reproducibility of diagnosis of serial recordings Computer Diagnosis First
No.
1 Atrial Fibrillation 2 3 4 5 6
Atrial Atrial Atrial Atrial Atrial
Fibrillation Flutter-fibrillation Flutter-fibrillation Fibrillation Fibrillation
7 8 9
Atrial Fibrillation Atrial Fibrillation Undetermined Rhythm
10 11
Undetermined Rhythm Undetermined Rhythm
12
Undetermined Rhythm
13
Undetermined Rhythm
14 Undetermined Rhythm 15 Suspect Ventricular Pacemaker 16 Normal Sinus Rhythm 17 Probably SB, rule out Heart Block 18 SB + Sinus Arrest + Junct. Escape 19 Low Heart Rate, suspect Heart Block 20 Complete Heart Block 21
Junctional Tachycardia
Physician's Diagnosis
Second Atrial Fibrillation Atrial Fibrillation Atrial Flutter-fibrillation Atrial Fibrillation Atrial Fibrillation Sinus Bradycardia Undetermined Rhythm Undetermined Rhythm Atrial Fibrillation + VPCs Undetermined Rhythm Low Atrial or Junctional Pacemaker Suspect Ventricular Pacemaker Normal Sinus Rhythm + VPCs Sinus Bradycardia Undetermined Rhythm
Atrial Fibrillation Atrial Fibrillation Atriat Fibrillation Atrial Fibrillation Sinus Bradycardia (SB) + SVPCs Sinus Arrhythmia + Junctional Escape Demand Pacemaker Sinus Rhythm + frequent SVPCs Atrial Fibrillation WPW Syndrome Left Atrial Rhythm + VPCs Demand Pacemaker Bigeminy due to VPCs Sinus Bradycardia Electronic Pacemaker
Undetermined Rhythm Probably SB, rule out Heart Block SB + Sinus Arrest
Normal Sinus Rhythm Sinus Bradycardia + Junctional Escape S-A Block + Junctional Escape
Probably SB, rule out Heart Block Suspect Ventricular Pacemaker Junctional Tachycardia
Sinus Bradycardia
category. Five of them were sinus rhythm, whereas the remaining five were artificial pacemaker rhythm. Rhythms classified as wandering pacemaker by the physicians were expected to be interpreted by the program as ~changing P contour." None of the nine records in this category were correctly detected by the program. Although the IBM program includes a logic for d e t e c t i n g electronic pacemaker and demand pacemakers, none of the 29 artificial pacemaker rhythms identified by the physicians were properly detected and classified into these two categories. Of five cases of WPW syndrome, two cases were detected by the computer, whereas one case was incorrectly diagnosed as sinus
Electronic Pacemaker Normal Sinus Rhythm
rhythm with VPCs+ SVPCs, and the remaining two cases were classified into undetermined rhythm without any suggestion concerning WPW syndrome.
DISCUSSION Since our patient population consists to a large extent of subjects with normal sinus rhythm, a substantial saving in electrocardiographer's time could be achieved if the computer program could accurately separate this category from categories with abnormal rhythms. The IBM program falsely classified only 30 records w i t h a b n o r m a l r h y t h m as sinus J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
ARRHYTHMIA DIAGNOSIS BY COMPUTER
23
Fig. 2. Sinus bradycardia with first degree AV block mistaken for atrial fibrillation by the computer. Low P waves could be detected by the physician, but not by the computer.
rhythm. In 27 of the 30 instances the misclassification took place in categories which are of little or no clinical significance, such as infrequent ectopic complexes and wandering pacemaker. In addition, the number of cases which actually belonged to the sinus r h y t h m category but were incorrectly regarded as abnormal was 32. The low incidence of false positive cases is worthy of note, because it could save the physician's time and labor to re-examine normal ECGs misclassified as abnormal. The high sensitivity (2153/2185 = 98.5%) and specificity (219/249 = 88.0%) of sinus r h y t h m detection was considered to prove the usefulness of the IBM program for our present purpose. The current IBM program could not recognize ectopic events if they appeared at the m o m e n t j u s t before or a f t e r t h e lead set switching. This fact was the major cause of errors in the detection of ectopic events. Another cause of error was misrecognition of SVPCs in sinus r h y t h m as sinus arrhythmia. Because of small sample size of interval data available, they could not be clearly separated. Longer continuous recording exclusively using a single channel might afford a larger sample of interval data. Moreover, it might give more information even in the neighborhood of lead set switching and allow us to obtain higher specificity. On the other hand, the causes of false posiJ. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980
tive detection vary. In nine cases, the physician could not find any ectopic events on the t r a c i n g recorded by the t e r m i n a l cart, although the computer detected either SVPCs (eight cases) or VPCs (one case). It was considered t h a t the most probable cause of incompatibility between the physicians' and computer's interpretation was the difficulty of separating SVPCs from sinus a r r h y t h m i a in borderline cases. Relatively low incidence of unidentified ectopic events showed t h a t the direct connection between carts and the central processor successfully decreased the noise during transmission. Discrepancies among P wave recognitions for individual lead sets was another cause of false positive detection and misclassification of true sinus r h y t h m into undetermined r h y t h m (UDR). Improvement of the accuracy of P wave identification would decrease these mistakes. Analysis of the 27 cases classified into the UDR category revealed t h a t an a r r h y t h m i a for which the IBM program did not prepare the proper diagnostic category was observed in only one case. The remaining 26 cases were erroneously diagnosed as UDR, though they should have been classified into other categories (Table 3). Consequently, it was considered t h a t an effective procedure to enhance the utility of the ECG program would be to lessen the number of cases in which the r h y t h m could not be determined. For example, 11 cases of artificial
24
MIYAHARA ET AL
pacemaker r h y t h m were diagnosed as UDR. An artificial pacemaker r h y t h m could be easily detected on an ECG chart by the physician from the characteristic pattern of sharp and tall spikes locating at the beginning of broad QRS complexes. The computer program was also designed to m a k e a diagnosis of the pacemaker r h y t h m by means of the spike artifact. However, the c u r r e n t computerized ECG system usually set a high-cut filter in a terminal cart and limited the band-width of the ECG signal. This device eliminated high frequency noise, but at the same time blunted the spike artifact so t h a t the computer program failed to identify the pacemaker r h y t h m and incorrectly classified 11 cases into UDR. As a tentative measure, before completing an advanced version of the program, we would propose alternative criteria to diagnose the cases of UDR which had wide QRS duration (160 ms or more) as artificial p a c e m a k e r rhythm. Pordy 6 was reluctant to use the findings of a regular heart rate, absent P waves and wide QRS complex as a criteria of electronic pacemaker, because an AV junctional pacemaker with IV block could be classified into this category. It was true, but use of the duration of QRS in UDR seemed to be effective in our hospital for the separation of artificial pacemaker from other abnormalities, because the incidence of electronic pacemaker was far greater t h a n t h a t of the junctional pacemaker with IV block. In fact, 11 cases of artificial pacemaker could be separated from UDR by this criteria, whereas only one case of UDR with wide QRS duration was erroneously diagnosed as artificial pacemaker. We repeated the ECG e x a m i n a t i o n consecutively when a technician found t h a t the computer gave the diagnosis of atrial fibrillation, h e a r t block, suspect of v e n t r i c u l a r pacemaker or undetermined rhythm. In Table 4, we summarized the comparison among the first and second computer diagnosis and the physician's. It was revealed t h a t at least one of the two computer diagnoses agreed with physician's in 11 of 21 cases. We noticed t h a t in five of six cases of undetermined rhythm, the second test gave a more meaningful diagnosis t h a n the first. These observations can be
at least partially explained by the fact t h a t the input signals in two successive ECG records are not identical even though the basic r h y t h m r e m a i n s u n c h a n g e d . It m i g h t be v a l u a b l e to t a k e ECGs r e p e a t e d l y if t h e d i a g n o s i s of u n d e t e r m i n e d r h y t h m w a s printed out. Since information regarding the accuracy of the computer program, especially as to the diagnosis of arrhythmias, is not yet available, cardiologists responsible for the performance of a computerized ECG system have to overread the ECG record of the computer r h y t h m diagnosis (even in the case of an output of simple sinus rhythm). It m i g h t be helpful to accumulate those cases misdiagnosed by the computer for the purpose of analyzing the detail and feeding the r e l e v a n t information back to the distributor of the original program. A special committee consisting of computer engineers, system users, relevant cardiologists, etc. could cooperate in analyzing questionable cases and discuss sending advice to the program distributors. If such a system were established, elaboration of the diagnostic logic would be promoted and more reliable a r r h y t h m i a diagnosis by the computer could be expected.
REFERENCES 1. WILLEMS,J L AND PIPBERGER,H V: Arrhythmia detection by digital computer. Comp Biomed Res 5:263, 1972 2. LEBLANC,A R AND ROBERGE,F A: III. Present state of arrhythmia analysis by computer. Can Med Assoc J 108:1239, 1973 3. MEYER, J, MERX, W, HEINRICH, K W AND EFFERT, S: Computer-analyse des Elektrokardiograms mit verschiedenen Programmen II. Rhythmusanalyse, Dtsch med Wschr 99:1294, 1974 4. CACERES,C A: Limitations of the Computer in electrocardiographic interpretation, Am J Cardiol 38:362, 1976 5. Health Care Support/Electrocardiogram (ECG) Analysis Program, Physician's Guide, IBM Corp., 1974 6. PORDY, L: Computer Electrocardiography: Present Status and Criteria, Futura Publ. Co., New York, 1977
J. ELECTROCARDIOLOGY, VOL. 13, NO. 1, 1980