Noninvasive reconstruction of cardiac electrical activity: update on current methods, applications and challenges

The 12-lead electrocardiogram (ECG) is a well-established, patient-friendly, quick, reproducible and cheap tool to determine normal cardiac activation and repolarisation to diagnose cardiac arrhythmias, altered activation, ischaemia, infarction, primary electrical abnormalities of the heart, structural disease, metabolic disorders, electrolyte imbalance and other conditions. It reflects the attenuated and dispersed result of propagated electrical activity and recovery in the heart on the body surface. However, it lacks the capacity to directly assess electrical activity at the level of the myocardium at high resolution. A modality that noninvasively images electrical activation and recovery at the heart surface could fill the gap between the noninvasive (low-resolution) 12-lead ECG and invasive (high-resolution) electrophysiology studies (EPs). Among other applications, this could facilitate the characterisation and imaging of electrical gradients under physiological and pathological conditions to localise the origins and circuits of ventricular tachycardias (VTs), to determine the substrate complexity underlying atrial fibrillation (AF) and to assess the size and location of an infarct scar, which are important for risk stratification.

Durrer et al. [1] were the first to thoroughly investigate the electrical activity of the intact human heart at the organ level. In their studies, they used explanted human hearts from patients who had died in the absence of heart disease. Years later, attempts to achieve this noninvasively in analytical and computer models [2‐11], experimental dogs [12], torso-tank experiments with isolated canine hearts [13, 14] and humans [15, 16] were published.

During the past decades, much progress has been made in solving the inverse problem of electrocardiography [17‐21] and applications in humans appear with increasing frequency [15, 16, 22]. Notably, Yoram Rudy’s group at Washington University in St. Louis, USA has published a diverse series of clinical studies in the past years (e.g. [23‐29]). In previous investigations in the Netherlands, noninvasive reconstruction of electrical heart activity was studied as well [7, 30‐34], and clinical collaborations will be pursued in the upcoming years [33, 35]. These methods carry different names, such as noninvasive electrocardiographic imaging (ECGI or ECG-imaging, coined by Yoram Rudy’s team) [15], myocardial activation imaging [22], electrocardiographic mapping [36], inverse solution mapping [16], noninvasive imaging of cardiac electrophysiology [37], three-dimensional cardiac activation imaging (3-DCAI) [38], or similar terminology. We will refer to this full range of techniques as ‘electrocardiographic imaging’.

In the present paper, we will discuss the different techniques available to noninvasively reconstruct electrical heart activity based on body-surface ECG recordings and anatomical imaging. A patient case will illustrate the application of ECGI in our centre in Maastricht (Fig. 1). This patient, a 63-year-old man, was known with third-degree atrioventricular block, for which he received a pacemaker. After developing cardiomyopathy, his device was upgraded to a biventricular (BiV) pacemaker and implantable cardioverter-defibrillator. His heart failure worsened, and he developed a high count (20 %) of ventricular extrasystoles (VES). No genetic or ischaemic causes could be found and treatment with drugs proved insufficient. Ultrasound repeatedly confirmed a low left ventricular ejection fraction of 20 % and moderate mitral regurgitation. When readmitted with recurrent monomorphic sustained VT (VT morphology similar to his VES beats, see Fig. 1), an origin at the aortomitral continuity was suspected (based on the 12-lead ECG), but during the EP study, an epicardial origin close to the distal coronary sinus seemed more likely. The patient underwent ECGI before the EP study. His case will demonstrate how noninvasively reconstructed epicardial electrograms can help localise the origin of VT, and how reconstructed activation isochrones are superior to QRS duration and morphology on the regular 12-lead ECG for determining the electrical synchrony in cardiac resynchronisation therapy (CRT). In this paper, we will also point out current shortcomings and basic questions that need to be addressed before noninvasive ECGI can be fully accepted in the clinic. Understanding the methods and pitfalls of ECGI is essential for the correct interpretation of its results.

Applications of noninvasive ECGI are summarised in Table 1. One of these is the ability to detect the origin of VT [26, 39‐42]. Although catheter ablation of sustained monomorphic VTs has been very successful in the past years, recurrence of VTs is still an issue, requiring repeated ablation procedures. There is not only a substantial recurrence rate in the large group of myocardial infarction-related VTs, but especially also in (inherited) cardiomyopathies [43]. Noninvasive ECGI may expedite EP studies and improve outcome, particularly in cases where the myocardium is very sensitive to catheter manipulation, or (such as the case presented in this paper) when only sporadic VES are present during the ablation procedure. These issues make intracardiac point-by-point mapping a strenuous task, whereas ECGI can be performed noninvasively for several hours and requires only one extrasystole to reconstruct its origin. Localising the origin of VT noninvasively can also reduce radiation burden and procedural complications and improve ablation success rate. For similar reasons, the management of atrial tachycardia may be facilitated by ECGI as well [44‐46]. Furthermore, the suggested ability of ECGI to differentiate between an epicardial versus endocardial VT origin [40] would allow to select the appropriate ablation setup beforehand.

ECGI may also help to reduce the currently large group of CRT non-responders, as it has been shown that noninvasive reconstruction of local activation timing predicts clinical CRT response better than QRS duration or the presence of left bundle branch block (LBBB) [36, 37, 47, 48]. Despite the low voltage amplitudes, AF has also been characterised by noninvasive reconstruction, assessing AF activation patterns and complexity [25], and ECGI significantly reduces invasive procedural time during AF ablation [49].

Another important public health problem is sudden cardiac death (SCD). Individual risk stratification for SCD is notably difficult [50]. ECGI may prove useful for the risk prediction of SCD by its ability to image and quantify conduction and repolarisation abnormalities, which are an important substrate for arrhythmias [24, 28, 51‐55]. More specifically, the electro-anatomical characterisation of scar tissue after myocardial infarction, myocarditis or cardiomyopathies by assessing local low-amplitude and fractionated electrograms could contribute to risk stratification for VT and SCD [27, 56, 57].

Although all these examples illustrate the potential value of inverse electrocardiography, no specific application has been developed to such an extent yet that it has been included in a clinical guideline.

The setup necessary for inverse reconstruction consists of specialised hardware and mathematical algorithms. So-called ‘forward models’ describe the propagation of electromagnetic activity from heart to body surface (Fig. 2). An inverse model is based on a forward model with known ‘output’ and unknown ‘source’; it essentially reverses the natural electromagnetic relationship between heart and body surface. Forward models consist of three parts: the cardiac source (representing the electrical activity of the heart), the model output (the body-surface potentials) and the electromagnetic source–output relation (capturing patient-specific propagation by an anatomical and conductivity reference). Reversing this electromagnetic relationship requires a fourth element, namely regularisation methods, to deal with the reconstruction’s sensitivity to noise. We will discuss these four elements in the next sections. Some technique characteristics are summarised in Table 2.

In the selected case from our patient studies, we obtained body-surface potentials with 256 electrodes and chose a potential-based representation of the cardiac source. The most important reason for doing so is to be able to reconstruct local electrograms and not only activation/repolarisation timing. Moreover, it has been validated more thoroughly in clinical settings. The patient-specific geometry, based on a CT scan during diastole, was chosen to be homogeneous. As a regularisation method, we applied Tikhonov zeroth order regularisation. Reconstructed electrograms are shown in Fig. 5, for three different beats in this patient. The insets in panel a and b show pseudo-unipolar electrograms recorded at the pacemaker lead tips in the right and left ventricle that correspond to the location of the shown reconstructed electrograms. Correlation coefficients between measured and reconstructed electrograms range from 0.79 to 0.85. Although the correlation coefficients are reasonably high, some important features are clearly not reconstructed correctly. This might be due to a combined effect of the pseudo-unipolar character of the pacemaker recordings, cardiac changes in the months between recording of body-surface potentials and pacemaker electrograms, the location mismatch in case of the right ventricle (endocardial recording versus epicardial reconstruction) and shortcomings in the inverse algorithm.

Activation isochrones can be created by determining the activation time at each epicardial location, as shown in the last two columns of Fig. 5. Activation of a sinus beat (panel a) starts at the right ventricle (red) and spreads to the left ventricle (blue), as was expected based on the LBBB morphology on the 12-lead ECG. Panel b shows activation isochrones for a paced beat (BiV paced). On the right ventricle, the reconstructed location with earliest activation (indicated with symbol #) is 23 mm from the known pacing location (indicated with symbol *) that was determined from the anatomical reference. For the left ventricle, the reconstructed and known locations of first activation are 28 mm apart. In agreement with pacemaker settings, the right ventricle was paced and activated before the left ventricle. The activation isochrones of a VES beat (panel c) suggest an origin of extrasystolic activity on the superior part of the left ventricle, although spatial resolution was too low to give a very precise location. From the EP study, an epicardial location at the mid-coronary sinus was suspected to be the origin, consistent with these reconstructions.

Figure 5 also demonstrates the advantage of inverse electrocardiography in CRT optimisation, by showing that electrical synchrony, i.e. measuring the time difference of activation of the complete left versus the complete right ventricle based on inversely reconstructed electrograms, gives better insight in ventricular (dys)synchrony than the often used 12-lead QRS duration.

Although reconstructed electrograms and activation times are consistent with expectations based on 12-lead ECGs, pacemaker recordings and the EP study, the level of accuracy and detail is suboptimal, with correlation coefficients being approximately 0.80 and a reconstructed pacing-origin mismatch of 23–28 mm. Other research groups have performed more extensive validation studies, claiming 10 mm accuracy and cross correlations of 0.70. [52] Other studies have found a significant decrease of accuracy in diseased hearts, with localisation accuracy going from 13 ± 9 mm (mean ± standard deviation) in healthy hearts up to 28 ± 27 mm in infarcted hearts and even 43 ± 11 mm near scarred tissue. [16] However, no thorough in vivo validation has been performed with potential recordings simultaneously at extensive locations at the heart and body surfaces in intact organisms.

The potential applications of noninvasive reconstruction of electrical heart activity are promising. Localisation of extrasystolic activity in a noninvasive, per-beat and precise manner combined with an anatomical reference can expedite diagnosis, guide therapy and reduce procedural radiation burden. Electrical (dys)synchrony assessment allows patient-specific CRT and might help in improving therapy for current non-responders. Future applications could also include noninvasive risk stratification for conduction abnormalities (conduction slowing, electrical chaos in AF, low-voltage amplitudes), direct relation of structural abnormalities to electrical abnormalities (fibrosis, fibrofatty replacement) and easier visual interpretation of abnormalities that are already diagnosed on the powerful 12-lead ECG. Commercial, easy-to-use setups such as those developed by CardioInsight (Cleveland, OH, USA) will be essential to fully exploit the potential of these applications in daily clinical routine.

Whereas noninvasive reconstruction of electrical heart activity currently has a clear research potential, its added value in clinical practice remains to be established. It is still unclear when, from a patient and socioeconomic perspective, it is worth the extra effort of extensive mapping of body-surface potentials, performing a CT/MRI scan for anatomical reference, and carefully avoiding the regularisation and reconstruction pitfalls. More importantly, cardiologists should be aware of the existence and influence of those pitfalls and the assumptions underlying the reconstruction algorithms. Debate remains on topics such as the influence of inhomogeneities in the torso (notably the lungs) on the reconstructions, the optimal regularisation methods and their parameters, the maximum achievable resolution of reconstruction, the number and positioning of body-surface electrodes and the assumption of a static geometry. We feel that more extensive validation should be performed, besides additional clinical studies investigating inverse ECGI for various purposes. Extensive validation by in vivo studies, and careful interpretation of clinical results with knowledge about the underlying methods and assumptions, will allow full exploitation of the potential of noninvasive ECGI, improving diagnosis, therapy and risk stratification.