The fact that the standard ECG is readily available for more than 7 decades in the clinical practice makes it attractive for big data analysis algorithms. Per year hundreds of millions of ECGs are recorded worldwide. These enormous amounts of ECG data are also more and more available in digital format. Based on this vast amount of digital ECG data artificial algorithms are able to detect diseased or potentially diseased hearts just from the ECG, of which Attia et al. gives a concrete example . The fact that this group showed a clear relationship between the dysfunction of the contractile myocardium and the ECG is a strong argument of the (potential) diagnostic power of the standard 12 lead ECG. The limited positive predictive of the proposed algorithm was partly attributed by the authors to the selection of the ejection fraction cutoff point at 35%. However, another major limitation, not mentioned by Attia et al, is the fact that ECG recordings are prone to several errors and therefore contain a large variation in the ECG measurements.
Strangely enough the standard 12 lead ECG is thus not delivering standardized output. This hinders the full exploitation of the ECG merits. A major limitation of the ECG are the (potential) human error, which cannot be quantified with the current ECG technology. Contrary to AI based quantitative image processing algorithms, with excellent performance in the detection of aberrations in the medical images, ECGs lack the check on data consistency, especially in the morphology based ECG waveform assessment. Factors that influence the morphology of the ECG waveforms are chest circumference, male versus female, and consequently heart size, position and orientation (figure 1). But the most influencing factor of all are electrode positions, prone to human errors and the factors mentioned above . The reason the electrode positions are so important is because each electrode functions as a virtual camera on the heart. Changing its position automatically means a different view on the heart, and thus a different waveform. For electrodes far away this influence is limited, but for the precordial electrodes (close to the heart) the influence is major and significantly hampers the diagnostic ability of the standard 12 lead ECG .
The traditional way to limit these errors in lead placement is to train the staff in positioning the precordial electrodes at standard electrode positions, which means V1-V2 in the 4th intercostal space, and in a line V4-6 just a few cm under the nipple. Obviously this guideline has been developed for male (figure 1). Consequently, female ECG electrodes will frequently be placed below the left breast, thus with a different view on the heart. This is one of the major reasons the ECG waveform morphology is only of limited use in clinical practice. The diagnosis of the right or left bundle branch blocks, acute coronary syndromes (ACS) all are frequently problematic which can result in a delayed treatment. Consequently alternatives have been introduced, like echo or for instance troponin levels for ACS. On the other hand studies like the above mentioned study of Attia et al. show that the information is somehow present in the ECG, otherwise their big data algorithms would not have found a relation between contractile function and the ECG.
To improve the standardized output and thus the diagnostic value of the ECG, both in clinical practice as well as to increase the positive predictive value of artificial intelligence algorithms, the ECG electrode positions needs to be known and used in deriving the standard and correct ECG output. For the localization of the electrodes on the chest a 3D camera can be used  (figure 2). The use of such a camera has already been proven to be of importance for the localization of cardiac arrhythmias .
The use of the camera enables a reliable and patient specific diagnosis, because it not only enables the localization of the electrodes, but also the body build (figure 2). The combination of these different modalities can very well be used to ensure standard ECG output which supports using artificial intelligence algorithms in the analysis of ECG signals. Not only is evaluating a single standard ECG output unavailable, but also comparing a person’s ECG over time. More than 115 year after the first clinical ECG recording by Einthoven it gets time to push the ECG forward into the artificial intelligence age!