The University of Glasgow (Uni-G) ECG Analysis Program
PW Macfarlane, B Devine, E Clark
University of Glasgow, Section of Cardiology and Exercise Medicine, Royal Infirmary, Glasgow
The University of Glasgow 12/15 lead ECG analysis
program has been in continuous development for over 20
years. It has been adapted to meet the needs of different
users and keep abreast of changes in terminology as well
as new morphological features described in the literature.
It is applicable to neonates as well as adults and takes
account of racial variation in wave amplitudes. It has a
capability for comparing serially recorded ECGs using
one of two different approaches. The many varying
features of the software have led to the introduction of the
descriptor Uni-G (unique) ECG analysis program.
Methods for the analysis of electrocardiograms using
automated techniques were first investigated in the
University of Glasgow in the late 1960’s. The earliest of
waveforms recorded in
simultaneously, whether they were from X, Y, Z
orthogonal leads or carefully selected groups of three
leads from the 12-lead ECG, e.g. I, aVF, V1 (1,2). At
the end of the 1970’s, a decision was made to move to
develop a 12-lead ECG analysis program where all leads
were recorded simultaneously.
electrocardiograph was designed and built locally (3) and
with this, ECGs could be collected in digital form to
permit further development of software for analysis and
interpretation. Indeed, this instrument was capable of
acquiring 11 leads simultaneously so that a complete 12-
lead plus an orthogonal 3-lead ECG could be obtained
Throughout the 1980’s, there was a major effort to
collect databases from apparently healthy individuals of
all ages from birth onwards and various publications
presenting these data have appeared previously (4, 5).
Full details will be published in a new edition of
Comprehensive Electrocardiology (6). Diagnostic
criteria evolved therefrom in a variety of ways and a
comprehensive 12-lead ECG analysis program was
introduced for worldwide interpretation of ECGs (7).
The University of Glasgow (Uni-G) ECG
interpretation program is based on an analysis of 8 or 11
simultaneously recorded leads acquired at 500 samples
per second. The first stage in analysis is to apply a 50 Hz
or 60 Hz notch filter to remove AC interference if this has
not already been done by the electrocardiograph itself.
Thereafter, methods for detection of excessive artefact
are used and if leads are found to have an unacceptable
quality of recording, the five seconds in which this is
found, i.e. the first or second half of the recording is set to
be a continuous value. It was found that it could be
beneficial to retain five seconds of a lead given that noise
very often occurs in short, one or two second bursts.
The next stage in the analytical process is QRS
detection and typing. Effectively, a function based on a
combination of available leads is formed from which
putative QRS complexes are determined. Thereafter,
wave typing is undertaken using an iterative process
whereby the first complex in Lead I is compared with the
second to look for any differences. The technique is
extended to include all complexes in this lead and then
repeated for four other leads, as often it is only one or two
leads which clearly show an aberrantly conducted
A complex selection procedure then decides which
class of beat will be selected for averaging and
subsequent interpretation. At this stage, cognisance has
to be taken of whether or not any beats are paced and
although the software itself has routines for detecting and
removing pacemaker stimuli, this is best achieved by
front end processing with signals sampled at a much
higher rate, e.g. 8,000 samples per second within the
electrocardiograph firmware itself. If this is done, a list
of pacemaker spike locations is forwarded to the Uni-G
program and the spike artifacts are removed from the
The program has optional approaches to computing the
average QRS cycle including a simple mean, a weighted
mean and a median beat. In different commercial
versions, manufacturers may utilise their own proprietary
0276−6547/05 $20.00 © 2005 IEEE
Computers in Cardiology 2005;32:451−454.
software for beat averaging if desired.
Different approaches to finding fiducial points have
been tried, including a simple form of threshold crossing
to a more complex template matching technique.
Ultimately, a combination of these approaches has been
adopted where, for example, QRS onset was found to
perform best with respect to a noisy test set using a
threshold technique. On the other hand, T-end
performed best using a template matching method. All
QRST amplitudes are referred to QRS onset as are P
wave measurements, which represents a departure from
an early approach where a straight line was fitted between
P onset and P termination.
Individual QRS and T fiducial points are derived for
all leads and a method of selecting the earliest QRS onset
for example is utilised in order to determine a global QRS
onset. A similar approach is adopted for QRS
termination and the difference between the two global
measurements is taken as the overall QRS duration. It
was found optimum to utilise a common P onset and P
termination in view of the unreliability of P wave
detection in many ECGs.
The wave measurement section of the program meets
all the requirements of the relevant IEC test procedures as
shown in Table 1.
Table 1. This table shows the mean and standard
deviation of the difference between the measurements
made by the Glasgow program and by 5 referees in the
100 ECGs in the CSE measurement set. The values in 
are the IEC acceptable differences and standard
deviations for global durations and intervals for biological
ECGS. It can be seen that the program results are well
within the recommended tolerances.
Difference Mean Standard Deviation
P Duration 1.348 
QRS Duration 1.609 
PR Interval 1.043 
QT Interval 0.602 
2.2. Rhythm analysis
The approach to rhythm analysis remains as before (8)
in that three leads are used. These are II, V1 and a third
lead selected from limb leads, usually the one with the
largest P wave amplitude in the case of sinus rhythm.
The basic rhythm strategy is to determine a dominant
rhythm such as sinus rhythm or atrial fibrillation and
thereafter determine any supplementary abnormalities
such as first degree AV block or ventricular extrasystoles.
A significant amount of work was done on the use of
neural networks to attempt to improve the accuracy of
determining atrial fibrillation (9) but ultimately it was
found that deterministic methods were equally acceptable.
Differentiation of atrial fibrillation with rapid ventricular
response from sinus tachycardia with frequent supra VES
still remains a difficult problem for automated techniques.
Relatively recently, newer methods for enhancement
of reporting atrial flutter were reported by the group (10).
While logic for detection of saw tooth waves has always
been present, the more recent logic adopted a threshold
crossing technique combined with regularity of intervals
between peaks resulting in an improvement in the
sensitivity of reporting atrial flutter from 27% to 79%,
with a specificity exceeding 98% in both cases.
2.3. Diagnostic interpretation
The diagnostic component of the software is capable of
using age, sex, race, clinical classification and drug
therapy within its logic. Experience has shown,
however, that many staff, particularly nursing staff, will
simply not take the time to input the appropriate measures
to the software, even the age and sex of a patient which it
is known are fundamental to accurate interpretation.
The basic approach to interpretation is through the use
of rule based criteria, but relatively recently this approach
has been enhanced in several ways. First of all,
smoothing techniques were introduced (11) to try to
minimise repeat variation in interpretations by avoiding
the use of strict thresholds between abnormal and normal.
In short, instead of a step function separating normal from
abnormal an exponential or even a linear function
between the normal and abnormal threshold value can be
used as illustrated. This is usually associated with a
scoring technique whereby it can be seen that a small
change in voltage for example results in a small change in
score. In the case of multiple parameters, more complex
combination rules apply as discussed elsewhere (12).
Neural networks have also been introduced for
detection of abnormal Q waves. However, it was found
in practice that these perform best in combination with
deterministic criteria (13).
Electrocardiography has not stood still in recent years
and new terminology such as ST elevation myocardial
infarction (STEMI) has been introduced. The software
acknowledges the newer diagnoses and a significant
amount of work has been done to adapt the output
appropriately (14). Another example of newer
terminology is that of the Brugada pattern of which
account has to be taken (Figure 1).
The software makes extensive use of age and sex of
Figure 1: Example showing Brugada pattern
Figure 2: Example showing two styles of report presentation for illustrative purposes only.
The brief format is on the right while the long format with reasons is shown on the left.
patients in reaching an interpretation. Continuous limits Download full-text
of normality have been introduced particularly for
children and younger males while different equations for
normal limits of amplitudes are used for males and
females especially in the younger adult age ranges. To a
certain extent, the race of a patient is acknowledged
through lower limits of normal voltage for Chinese
individuals, for example.
Finally, the software contains methods for comparison
of serial tracings. Two approaches are utilised the first
of which involves integrating criteria within the main
logic leading to statements such as “serial changes of
myocardial infarction” (15). A newer approach has been
to add on separate logic for serial comparison, which then
functions as a secondary program that is run following the
main diagnostic logic. In this case, there are advantages
of having almost all serial comparison logic in the same
section of code although it is perhaps an approach more
favoured in North America than elsewhere. This,
therefore, highlights the question of user choice which
also applies to the style of output presentation. Two
different styles are offered, one whereby explanatory
reasons are printed along with a diagnostic statement and
the other where a much more brief diagnostic comment is
produced. The different styles can be compared in
Figure 2 where a research style output is produced to
illustrate the different approaches.
Finally, it should be remarked that the program has a
capability of handling 15 leads and the user is at liberty to
select for example V3R, V4R and V7 even although the
diagnostic logic at present does not incorporate criteria
from these leads. If the additional leads happen to be X,
Y, Z leads computed from the 12-lead ECG using an
inverse Dower transformation for example (16), then
additional vectorcardiographic measurements can be
made and vectorcardiographic loops output.
3. Discussion and conclusions
The Uni-G program has continued to evolve over a
long period of time and could still be said to be under
development, given the changing fashions in medicine
and the underlying fact that the 12-lead ECG still remains
the most commonly used diagnostic test in clinical
medicine despite the availability of much more complex
procedures. The ECG still provides unique information
which, in many ways is complementary to the newer
techniques but is obtained in a much more simple and
rapid fashion as demanded in many clinical situations.
 Macfarlane PW. ECG waveform identification by digital
computer. Cardiovasc Res 1971;5:141-6.
 Macfarlane PW, Cawood HT, Taylor TP, Lawrie TDV.
Biomed Eng 1972;7:176-80.
 Watts MP, Shoat DB. Trends in electrocardiographic
design. J IERE 1987;57:140.
 Macfarlane PW, Coleman EN, Pomphrey EO, McLaughlin
S, Houston A, Aitchison TC. Normal limits of the high-
fidelity pediatric ECG. Preliminary observations. J
 Macfarlane PW, Lawrie
electrocardiogram and vectorcardiogram. In: Macfarlane
PW, Lawrie TDV, eds. Comprehensive Electrocardiology.
Oxford: Pergamon Press, 1989;Vol 1:407-458.
 Macfarlane PW, van Oosterom A, Janse M, Camm J,
Kligfield P, Pahlm O. Comprehensive Electrocardiology
(2nd Edition) London:Springer 2007. In preparation.
 Macfarlane PW, Devine B, Latif S, McLaughlin S, Shoat
DB, Watts MP. Methodology of ECG interpretation in the
Glasgow program. Meth Inf Med 1990;29:354-61.
 Macfarlane PW. Computer interpretation of cardiac
rhythm. In: JL Willems, JH van Bemmel, C Zywietz, eds.
Computer ECG Analysis:
Amsterdam: North Holland, 1986:279-84.
 Yang TF, Devine B, Macfarlane PW. Artificial neural
networks for the diagnosis of atrial fibrillation. Med Biol
Eng Computing 1994;32:615-9.
 Morrison S, Macfarlane PW. Computer detection of atrial
flutter. Annals of Non Invasive Electrocardiology
 McLaughlin SC, Aitchison TC, Macfarlane PW. Improved
repeatability of 12-lead ECG analysis using continuous
scoring techniques. J Electrocardiol, 1993;26(suppl):101-7.
 McLaughlin SC, Aitchison TC, Macfarlane PW. Methods
for improving the repeatability of automated ECG analysis.
Meth Inform Med 1995;34:272-82.
 Yang TF, Devine B, Macfarlane PW. Use of artificial
neural networks within deterministic logic for the computer
ECG diagnosis of inferior myocardial infarction. J
 Macfarlane PW, Browne D, Devine B, Clark E, Miller E,
Seyal J, Hampton D. Modification of ACC/ESC criteria
for acute myocardial
 Hedstrom K, Macfarlane PW. Development of a new
approach to serial analysis. The manufacturer’s
viewpoint. J Electrocardiol 1996;29(suppl):35-40.
 Macfarlane PW. Lead systems. In: PW Macfarlane, TDV
Lawrie (eds) Comprehensive Electrocardiology Oxford
Pergamon Press 1989:315-352.
Address for correspondence
Peter W Macfarlane
Section of Cardiology and Exercise Medicine,
10 Alexandra Parade,
Glasgow G31 2ER,
automated electrocardiogram interpretation.
TDV. The normal
infarction. J Electrocardiol