The relation of 12 lead ECG to the cardiac anatomy: The normal CineECG


The interpretation of the 12‑lead ECG is notoriously difficult and requires experts to distinguish normal from abnormal ECG waveforms. ECG waveforms depend on body build and electrode positions, both often different in males and females. To relate the ECG waveforms to cardiac anatomical structures is even more difficult. The novel CineECG algorithm enables a direct projection of the 12‑lead ECG to the cardiac anatomy by computing the mean location of cardiac activity over time. The aim of this study is to investigate the cardiac locations of the CineECG derived from standard 12‑lead ECGs of normal subjects.


In this study we used 6525 12‑lead ECG tracings labelled as normal obtained from the certified Physionet PTB XL Diagnostic ECG Database to construct the CineECG. All 12 lead ECGs were analyzed, and then divided by age groups (18–29,30-39,40-49,50-59,60-69,70–100 years) and by gender (male/female). For each ECG, we computed the CineECG within a generic 3D heart/torso model. Based on these CineECG‘s, the average normal cardiac location and direction for QRS, STpeak, and TpeakTend segments were determined.


The CineECG direction for the QRS segment showed large variation towards the left free wall, whereas the STT segments were homogeneously directed towards the septal/apical region. The differences in the CineECG location for the QRS, STpeak, and TpeakTend between the age and gender groups were relatively small (maximally 10 mm at end T-wave), although between the gender groups minor differences were found in the 4 chamber direction angles (QRS 4°, STpeak 5°, and TpeakTend 8°) and LAO (QRS 1°, STpeak 13°, and TpeakTend 30°).


CineECG demonstrated to be a feasible and pragmatic solution for ECG waveform interpretation, relating the ECG directly to the cardiac anatomy. The variations in depolarization and repolarization CineECG were small within this group of normal healthy controls, both in cardiac location as well as in direction. CineECG may enable an easier discrimination between normal and abnormal QRS and T-wave morphologies, reducing the amount of expert training. Further studies are needed to prove whether novel CineECG can significantly contribute to the discrimination of normal versus abnormal ECG tracings.


Since its first introduction one hundred years ago, the 12‑lead electrocardiogram (ECG) has become the cornerstone tool for the diagnosis of cardiac rhythm disorders [1]. Even if today ECGs are recorded in digital format, the clinical interpretation is still largely based on the expert visual examination of ECG waveforms, such as QRS and T wave amplitude and morphology.

The interpretation of the electrocardiographic (ECG) waveforms represents a challenging task for all physician at all training levels [1], and the inability to interpret the ECG with adequate accuracy in clinical practice often persists despite extra training [2]. One of the major difficulties in the ECG interpretation is due to the inter-individual variability in the normal ECG waveforms [[3][4][5]], influenced by gender, electrode positions during the ECG recording, and specific electrophysiological properties of the atria and ventricles. The current standard ECG analysis is essentially an unconscious expert pattern recognition process, by which specific ECG morphologic characteristics allow the diagnosis of distinct cardiac diseases [2]. Thus, correct diagnoses maybe performed only by highly trained and experienced professionals, and less common and more subtle abnormalities may go overlooked, missing crucial clinical diagnosis. On the other hand, less experienced personnel are more likely to forward non-pathological ECG deviations to the cardiology department, thereby unnecessarily increasing the expert ECG reader workload. Often, even expert cardiologists cannot agree on identifying some subtle properties of ECG tracings, providing inconsistent and discordant diagnostic results; moreover, the access to experts may be limited and costly, highlighting the need for less expensive and more affordable and objective methods for ECG analysis [6,7].

We propose a new approach to the ECG analysis, aimed to correlate the cardiac electric activity to its source, amplifying even the low amplitude ECG components, and utilizing the direction of the cardiac electrical activity to estimate the mean trajectory of cardiac activation and recovery. In this study, the movement of the CineECG1 is computed for the three major cardiac axes: X) posterior-anterior, Y) left-right, and Z) base-apex [8,9]. Ultimately, with these three CineECG graphs we aim to facilitate the detection of pathologic QRS and T-wave waveform, even by limited trained personnel.

Normal ventricular activation is initiated by the Purkinje system, as this specialized ventricular conduction tissue rapidly distributes the electrical activation through the heart [10,11]. The dense and widely branched endocardial system of Purkinje fibers originates from the His-bundle from which bifurcates in several major branches to innervate the left and right ventricle [12,13]. The ECG signals are the result of multiple activation waves initiated at multiple endocardial sites of the left and right ventricles. Consequently, the ECG activation waveforms, constituting the QRS complex, depend on the timing and anatomical locations of these early breakthroughs. The resulting normal mean QRS axis, however, shows limited correlation with the anatomical axis of the heart, indicating the high interindividual variability in the anatomical locations of the initiation sites [14]. Ventricular repolarization represented by the T-wave following the QRS complex, is a much slower process than the activation process, and consequently the local electrical gradients within the myocardium are smaller. The direction of both the depolarization and repolarization gradients imaged by the 12‑lead ECG are utilized to compute the CineECG trajectory using a standard 3D heart/torso model. Aim of this study is to investigate the ability of the CineECG to relate the standard 12‑lead depolarization and repolarization ECG waveforms to a standard cardiac anatomy, in order to improve the understanding of the standard 12‑lead ECG waveforms.


CineECG method

The CineECG describes the average location of all cardiac electric activity during the activation and recovery phase, as previously described [8,9,15,16], by means of mean temporal-spatial isochrones (mTSI). In summary, as described in Fig. 1:

A) The input for CineECG is the standard digital 12‑lead ECG and an anatomical heart/torso model with standard positions of the 12 lead ECG electrodes on the chest, utilizing the heart/torso of a 57-year male, with average body build relative to the study population;
B) The 12‑lead ECG is converted into the vectorcardiogram (VCG) as described by Boonstra el al. [8]. The computed VCG is a 3D- representation of the direction of cardiac activity per time interval.
C) In the next step, the CineECG trajectory is computed, depicting the localization of the average electrical activity of the ventricles at each given time interval. The CineECG uses only the VCG direction, indicating the direction of electrical activity per time step. The VCG direction is obtained by normalizing the VCG signal to a unit vector (VCGtVCGt). To localize the CineECG trajectory within the cardiac anatomy, the mid QRS location is set to the center of mass of the ventricular model. This site represents the center of electrical ventricular activation, when approximately half of the ventricular mass is activated. From this site, the subsequent CineECG cardiac locations are computed according to the following equations:CineECGt=CineECGt−1−v∙dt∙VCGtVCGtt<midQRS

CineECGt=center of ventricular masst=midQRSCineECGt=CineECGt+1+v∙dt∙VCGtVCGtt>midQRS

Fig. 1


Fig. 1. The workflow of the computation of the CineECG. A) the CineECG input is the standard 12 lead ECG assuming standard 12 lead ECG electrode positions. B) The ECG is converted into the vectorcardiogram, representing the direction of cardiac activity through the heart beat. Notice the differences in amplitude between QRS (in white-red) and ST-T-wave (yellow-blue). C) The VCG direction is used to estimate the mean temporo-spatial isochrone position for the QRST sequence. D) The X,Y,Z components of the CineECG are plotted relative to the first CineECG position in the heart. For this normal example the CineECG position moves initially trans-septal, i.e. to the right, subsequently back to the left as the LV has more mass, and finally the T-wave CineECG position moves towards the apex. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

in which v is the velocity with which the CineECG travels through the myocardial anatomy, and dt is the time step, in this case 1 ms. For the QRS the velocity is set to 0.7 ms−1, and for the STT segment the velocity is set 0.7xQRSdurationSTTduration ms−1, to ensure the length of QRS and STT segment is the same. The used velocity of 0.7 m/s is chosen to be in the physiological range of the myocardial propagation velocity [17,18].

D) In the final step the CineECG location per X, Y, and Z axis is plotted relative to the start of the CineECG (Fig. 1D), where the X-axis represents the posterior-to-anterior axis, Y-axis the right-to-left axis, and Z-axis the base-to-apex axis. Plotting the movement of the CineECG location relative to the initial cardiac location enables the comparison between ECGs of different subjects.

Study Population for CineECG analysis

As reference for normal CineECG locations, 6525 (3681 female, 2834 male) ECGs labelled as normal from the certified Physionet PTB XL Diagnostic ECG Database were used [19]. Each ECG was assigned to an age and gender groups by decades (Table 1). For each ECG a median beat was automatically constructed. For each median beat, the QRS onset, QRS end, and T wave end fiducial points were automatically determined and then visually checked using dedicated software to adapt the timing of the fiducials. For this process, also the root mean square curve signal of the median beat was used. Based on the median beat fiducials, the T-peak time was automatically determined from the root-mean-square signal as the time where the peak of the T-wave occurred. To compare ECG signals and CineECG locations for every individual QRST beat, every QRST durations have to be the same. Obviously the QRST durations were different for every ECG, which required a resampling process to create QRST sequences with the same length. To that end, each ECG was split in to three segments: a) QRS complex (up to 115 ms), b) STpeak segment from 115 ms to Tpeak, and c) from Tpeak to Tend. As the QRS complex is relatively homogeneous in duration, this segment was not resampled, whereas the recovery phase (STpeak-TpeakTend) depends significantly on the heart rate. To correct for these heart rate related differences in STT segment duration this segment is resampled. The STpeak segment was resampled to 160 ms, while the TpeakTend segment was resampled to 125 ms. Thus, the resulting QRST had a 400 ms duration for every ECG beat.

Table 1. The number of different age/gender groups of all normal labelled PTB-XL ECGs for 3 segments of the ECG (QRS, STpeak, and TpeakTend) for each of the cardiac views: 4 Chamber (4CH) RAO and LAO, see also Fig. 3) the angles is given in that plane. All angles are given as average ± standard deviation relative to the positive x-axis.

30 ± 34
38 ± 44
105 ± 93
102 ± 83
16 ± 34
18 ± 47
100 ± 15
105 ± 12
72 ± 10
72 ± 7
−117 ± 36
−131 ± 27
84 ± 11
88 ± 11
83 ± 7
80 ± 7
−45 ± 48
−79 ± 50
26 ± 36
33 ± 38
93 ± 93
98 ± 91
13 ± 34
14 ± 37
103 ± 17
105 ± 12
72 ± 13
72 ± 7
−123 ± 40
−129 ± 32
83 ± 11
89 ± 11
83 ± 7
78 ± 7
−44 ± 49
−84 ± 43
25 ± 36
30 ± 39
94 ± 98
95 ± 94
12 ± 33
13 ± 38
103 ± 17
106 ± 10
72 ± 13
71 ± 7
−126 ± 39
−131 ± 31
81 ± 10
90 ± 10
82 ± 9
77 ± 7
−44 ± 42
−90 ± 42
24 ± 35
29 ± 41
95 ± 92
89 ± 95
10 ± 35
11 ± 40
106 ± 19
107 ± 13
73 ± 16
71 ± 8
−130 ± 55
−132 ± 30
80 ± 10
90 ± 11
81 ± 7
77 ± 7
−42 ± 38
−88 ± 43
20 ± 40
21 ± 42
74 ± 101
77 ± 105
7 ± 36
8 ± 40
106 ± 24
109 ± 14
74 ± 18
71 ± 9
−134 ± 57
−135 ± 40
80 ± 12
88 ± 11
81 ± 9
78 ± 7
−43 ± 43
−80 ± 44
24 ± 34
26 ± 37
82 ± 89
85 ± 95
3 ± 31
9 ± 34
108 ± 21
111 ± 20
75 ± 17
72 ± 10
−140 ± 72
−139 ± 51
83 ± 13
88 ± 11
81 ± 10
79 ± 7
−50 ± 46
−80 ± 49
21 ± 36
25 ± 40
23 ± 38
55 ± 95
57 ± 97
56 ± 96
3 ± 34
4 ± 39
4 ± 36
104 ± 24
109 ± 39
106 ± 21
70 ± 17
69 ± 9
70 ± 14
−113 ± 51
−126 ± 34
−119 ± 45
81 ± 11
89 ± 11
85 ± 12
81 ± 8
78 ± 7
80 ± 8
−49 ± 44
−79 ± 45
−62 ± 47

Based on the CineECG locations, the directions per segment can be computed by the difference between the last minus the first cardiac location of the respective segment.


For each of the age groups, the average CineECG location is plotted along one of the three cardiac axes (Fig. 2). The variation among the age groups is limited (Fig. 2 bottom panel) with a maximal difference of about 10 mm between the different gender groups. The CineECG direction for the QRS, STpeak segment, and TpeakTend segment are used to compute the angle in one of the three standard heart views, i.e., 4-chamber view, right and left anterior oblique (Fig. 3). The angle distribution (histograms) in each of these three heart views and its spatial distribution for the 5–95% angle range is shown in Fig. 3. The variation for the QRS direction is large, spread over the whole left ventricle free wall, whereas the variations for the STpeak and TpeakTend direction are rather limited, both directed towards the apex of the heart among all age/gender groups (Table 1). The CineECG directions found for male and female were similar, both in the average values as well as in standard deviation. The largest differences in average direction were found in the 4-chamber view (QRS 4°, STpeak 5°, and TpeakTend 8°) and LAO view (QRS 1°, STpeak 13°, and TpeakTend 30°), see Table 1. The relatively large spread in the TpeakTend segment of the LAO angle only applies to the apical region as identified in the 4 chamber view and RAO view, making the distance variation in the apical region still small.

Fig. 2

Fig. 2. The mean of the CineECG per heart axis for each age/gender group (red liens female, blue lines male). The mean and the 5%, 95% confidence interval CineECG trajectories are very similar over the age group (see bottom panels), although a small difference can be seen between male and females in the right to left axis (X-axis) and the apex to base axis (Z-axis).

(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)


Fig. 3


Fig. 3. The distribution of the angle in the 3 standard heart view for QRS, STpeak segment and for the TpeakTend. The zeros degree axis is in the spatial distribution view is the positive x axis, from the middle of the heart to the right. The spatial distribution of the QRS, STpeak segment and for the TpeakTend is shown on the right. Notice that the variation of the TpeakTend is much smaller than that of the QRS, with the ST segment taking an intermediate behaviour.

The distribution of these CineECG waveforms for normal ECG tracings is plotted as a background for a specific CineECG in Fig. 4a. The orange lines represent the 90% range (5–95%) of the CineECG as derived from all normal ECG tracings, providing a normal range for CineECG tracings. Additionally, the CineECG range class is shown, i.e., the CineECG trajectory shown in the three heart planes with three different colors: green CineECG within the 5–95% normal range, orange for the 0–5% and 95–100% normal range and outside the normal range as red. For the normal CineECG example in Fig. 4a, the trajectory is completely green, indicating that it is within the normal range for the complete QRST sequence. Additionally, three examples are shown in Fig. 4b) Left bundle branch block (LBBB), with the CineECG oriented to the left and outside the normal range, indicated as an orange/red trajectory; Fig. 4c) Right bundle branch block (RBBB), with the CineECG oriented to the right base and outside the normal range, also indicated as an orange/red trajectory; Fig. 4d) the CineECG of a COVID-19 patient with T-wave inversion in some of the precordial leads, also indicated in green for the QRS, and in red for the T-wave trajectory.

Fig. 4


Fig. 4. The 90% distribution range of the normal CineECG positions along the 3 heart axis (orange lines) in combination with the CineECG positions as shown in Fig. 1 (dark green line) for a) normal ECG, b) LBBB ECG, c) RBBB ECG, and d) an COVID patient ECG. For each ECG a CineECG range class view is added in which the colors (green/orange/red) indicate the if the CineECG is within the normal range. For the normal ECG all samples of the CineECG wall within the normal distribution and thus are indicated in green in the range class view. For the LBBB and RBBB the CineECG is clearly outside the normal range and thus indicated in orange (borderzone) and red (outside the normal range). The last example in panels d) show an example of a COVID patient with clear deviating T-wave, wherase the QRS complex is still within the normal /border zone normal range.xy. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)


This study shows that the cardiac axis view of the CineECG (Fig. 2 and Fig. 4a) displays uniform QRS, STpeak and TpeakTend locations relative to the cardiac anatomy, providing a standardized and easy to interpret method to discriminate normal from abnormal activation and recovery. Then, the CineECG provides a novel view of the 12‑lead ECG with a direct link to the cardiac anatomy, which may enable easy detection of abnormalities of the conduction or recovery processes, even by less experienced ECG readers (Fig. 4) [8,16].

The normal activation of the ventricles

CineECG highlighted a much larger variation in the QRS direction compared to the terminal T wave direction (Fig. 3), indicating a rather limited relationship between the QRS direction and the orientation of the ventricles in the chest, whereas the terminal T-wave direction is more likely to be uniquely correlated to the heart orientation in the chest. The limited agreement of QRS axis and heart orientation is in agreement with the study by Engblom et al. [14], in which they could not find a correlation between anatomical cardiac axis and the electrical QRS axis. Even the variation of the position of the papillary muscles in the left ventricle is not enough to explain the relation between the cardiac anatomy and the ventricular activation, indicating that also the left and right septal activation sites show a significant interindividual variability contributing to the wide normal range of the electrical QRS axis [20]. This variability was also shown in studies conducted in normal hearts by Durrer et al. [21] and Opthof et al. [22], in whom the high left septal activation was not found in every patient, and the position of the initial activation of the right free wall differed among patients as well. This variability is also found in our results, showing a large variation of the QRS CineECG direction (Table 1). Of note, the differences in variations between QRS and T wave directions within this population of normal subjects using the same heart/torso model, suggest that those differences were intrinsic of the depolarization and repolarization process. The CineECG QRS directions also support the large range found in the normal frontal QRS axis range, thereby indicating the limited value of the frontal QRS axis. Our preliminary results indicate that the CineECG may capture the mean cardiac activity location over time, allowing for the detection of conduction disorders [8], or late QRS electrical forces [16]. Examples of three different abnormal CineECG’s (Fig. 4b-d) clearly show that these deviate from the normal CineECG (Fig. 4a). The CineECG thus might reduce the workload of the experienced ECG readers as seemingly abnormal ECGs are currently forwarded to the cardiology department by less experienced ECG readers. With the clear CineECG indicators, this may be prevented.

The normal recovery of the ventricles

The repolarization process, depicted as the T-wave, is a much slower process than the activation process, and consequently the local electrical gradients within the myocardium are much smaller. Therefore, the electrotonic interaction, due to the cell-to-cell coupling, plays a significant role in the cardiac recovery processes, i.e., areas activated earlier recover later, and vice versa [[23][24][25]]. Such repolarization sequence is reflected in the mean path of recovery, as represented by the CineECG. Our results show that repolarization gradients are moving from base to apex, with a smaller variation in the STpeak and TpeakTend directions compared to the QRS direction (Table 1Fig. 3). A potential explanation of these results might be that the epi-endo and left right recovery gradients effectively cancel each other, leaving only the base-to apex gradient as the driving forces for CineECG direction, especially for the terminal part of the recovery (TpeakTend). According to this hypothesis, small gradients in the recovery sequence, for instance due to myocardial ischemia or other cardiac disorders, may result in a significant change of the CineECG direction in the STpeak and TpeakTend directions and location. Such changes in the CineECG direction during the recovery phase can be visually detected as deviations from the CineECG normal distribution, as shown in Fig. 4.

VCG versus CineECG

The VCG amplitude reflects the volume from which the electrical activity arises at any time interval. Consequently, the mean QRS axis direction is both dominated by the left ventricle, as this encompasses the majority of cardiac tissue and significantly influenced by precordial electrodes positioned close to a local maximum or minimum on the body surface [26]. Obviously, the latter argument is of limited influence for the frontal plane QRS axis. The CineECG, however, only uses the direction of the VCG to compute subsequent positions of the average cardiac activity in the heart. By taking only the direction and leaving out the VCG amplitude, the CineECG is less sensitive to electrode positions, not taking into account the size of the area being active. Additionally, the CineECG computation provides the opportunity to correct for electrode placement. Thus, the CineECG can be viewed as an upgraded view of the VCG.

Anatomical relation

In this study, the CineECG was computed using a standard heart/torso model, in order to provide an anatomical localization of the electrical activity on the heart model. The CineECG thus offers a quantifiable insight in the relation between the ECG waveforms and cardiac anatomy. Patient specific models of the heart and torso including the localization of the electrode positions might display a more uniform distribution of the CineECG waveforms, resulting in smaller standard deviations than those reported in this study (Table 1). Using a 3D-camera to localize the electrode position or to guide electrode positioning on the chest might enable this information to be provided without modifying the clinical practice of ECG recording [27].

A remarkable result was the direction of the terminal T-wave (TpeakTend), which was rather homogeneously directed towards the left apex of the standard heart in all age and gender groups. The limited angle with a relatively small standard deviation might indicate that the normal terminal T-wave direction might correlate better with the anatomical heart axis than the QRS direction [14]. These results also suggests that the electrical heart axis only changes very little with increasing age. Moreover, the terminal T-wave direction differed between males and females on average by 8° in 4-chamber angle and 30° in the LAO angle. This difference might be due to systematic gender differences in the electrode position on the chest, since in women they are generally positioned below the breast. This aspect will need a specific analysis, preferably by using a 3D camera setup to specifically localize the ECG electrode position with respect to the heart using a patient specific heart/torso model [27].

CineECG in cardiac diseases

In previous studies we have already shown that the CineECG can be used to identify specific CineECG patterns related to the underlying disease. In Brugada syndrome patients the terminal part of the QRS moves in the anterior direction, i.e. the RVOT, the location of the arrhythmogenic substrate [16]. In this study CineECG was even able to identify the Brugada patients in ECG waveforms with limited signs of Brugada patterns (type 2 and 3). In another study the ECG of several conduction disorders showed specific CineECG patterns [8]. The typical left or right bundle CineECG pattern shows that the CineECG moves outside the normal distribution of CineECG locations (Fig. 4 b,c) at the terminal part of the QRS, either to the left or right in the Left-Right CineECG axis view. In a preliminary study we also showed that the deviating T-wave directions might be usable for the identification of COVID patients with underlying heart problems [15], which is supported by the results shown in Fig. 4d.


The main limitation of this study is the fact that a standard male heart/torso model is used to compute the VCG and then the CineECG from the 12‑lead ECG. This may influence the results, especially in females. However, the potential error is the same for all ECGs included in the study and the difference is likely to be small, particularly among subjects with normal hearts and body size [19]. The advantage of using a standard anatomical model is that the CineECG technology can be directly implemented to any ECG recorder. However, to better study specific disease conditions in which the cardiac anatomy is structurally affected, the CineECG might require a patient or disease specific heart/torso anatomical model, to obtain a reliable anatomical representation of the localization of the ECG waveforms.

In this study, we explored the temporo-spatial localization of CineECG for the QRs, STpeak and TpeakTend segments. However, a more specific tempo-spatial localization needs to be implemented, with a precise characterization of the duration of the CineECG trajectories will be necessary to provide a precise discrimination between normal and abnormal ECG tracings.

In this study we provided a quantitative CineECG interpretation, but a major advancement may come by the implementation of deep learning methods to CineECG analysis. The combination of these two approaches may be synergically provide an accurate discrimination between normal and abnormal ECG tracings, especially because CineECG allows an anatomical localization of the ECG waveforms, enabling a possible interpretation of the findings provided by deep learning algorithms.


CineECG computes an immediate visual and pragmatic representation of the ECG waveforms, relating the ECG waveforms directly to the cardiac anatomy. This study shows that CineECG can provide the standard tempo-spatial location of the activation and recovery process in normal subjects, observing a larger variation in the QRS directions than in the T wave direction, consistent with previous studies. CineECG may be used to discriminate between normal and abnormal QRS and T-wave morphologies, even without a major expert training. Further studies need to explore the tempo-spatial characteristics of the CineECG in different cardiac conditions, to verify the ability of CineECG to diagnose and stratify patients with different cardiac diseases.



Leave A Comment