A method to extract realistic artifacts from electrocardiogram recordings for robust algorithm testing.

OBJECTIVE: Recordings of signal noise and artifacts can be added to clean electrocardiogram (ECG) records to assess the performance of ECG and arrhythmia analysis algorithms in the presence of noise. We present a method to estimate device-specific signal noise and artifacts from ECG records. This method can be applied to obtain noise estimates from healthy subjects on any ECG lead, allowing a simple device-specific recording. The proposed approach is assessed using the MIT-BIH Noise Stress Test Database recordings combined with simulated ECGs. METHODS: The proposed noise-estimation method is based on the subtraction of a time-aligned median beat from a noisy ECG recording. To test our method, electrode motion and muscle artifact noise from MIT-BIH Noise Stress Test database were added to simulated ECG signals at signal-noise ratios (SNR) from -6 to 20 dB. A comparison between noise and estimated noise signal statistical characteristics was made including root-mean squared error and assessment of the power content in three frequency bands (cardiac [0.5-5 Hz], mid [5-25 Hz], and high [25-40 Hz]). RESULTS: Visual assessment and frequency analysis demonstrate the good quality of noise estimation. Root-mean squared error between noise and estimated noise signals was <0.5 Normalized Units across all SNR levels. Band power error was stable across SNR levels with median percentage error between noise and estimate noise signals of <10% for cardiac and mid frequency bands. CONCLUSION: Estimating noise from ECG records is a viable approach to generate noise and artifacts-only signals. These signals are device-specific and easy to collect from healthy subjects without requiring special electrode set-ups. Therefore, they may be suitable for use with annotated ECG databases to assess the robustness of ECG analysis algorithms in the presence of noise.

A novel ECG detector performance metric and its relationship with missing and false heart rate limit alarms.

PURPOSE: Performance of ECG beat detectors is traditionally assessed on long intervals (e.g.: 30min), but only incorrect detections within a short interval (e.g.: 10s) may cause incorrect (i.e., missed+false) heart rate limit alarms (tachycardia and bradycardia). We propose a novel performance metric based on distribution of incorrect beat detection over a short interval and assess its relationship with incorrect heart rate limit alarm rates. BASIC PROCEDURES: Six ECG beat detectors were assessed using performance metrics over long interval (sensitivity and positive predictive value over 30min) and short interval (Area Under empirical cumulative distribution function (AUecdf) for short interval (i.e., 10s) sensitivity and positive predictive value) on two ECG databases. False heart rate limit and asystole alarm rates calculated using a third ECG database were then correlated (Spearman's rank correlation) with each calculated performance metric. MAIN FINDINGS: False alarm rates correlated with sensitivity calculated on long interval (i.e., 30min) (ρ=-0.8 and p<0.05) and AUecdf for sensitivity (ρ=0.9 and p<0.05) in all assessed ECG databases. Sensitivity over 30min grouped the two detectors with lowest false alarm rates while AUecdf for sensitivity provided further information to identify the two beat detectors with highest false alarm rates as well, which was inseparable with sensitivity over 30min. PRINCIPAL CONCLUSIONS: Short interval performance metrics can provide insights on the potential of a beat detector to generate incorrect heart rate limit alarms.

Computing the spatial QRS-T angle using reduced electrocardiographic lead sets.

The 'spatial QRS-T angle' (SA) is frequently determined using linear lead transformation matrices that require the entire 12-lead electrocardiogram (ECG). While this approach is adequate when using 12-lead ECG data that is recorded in the resting supine position, it is not optimal in monitoring applications. This is because maintaining a good quality recording of the complete 12-lead ECG in monitoring applications is difficult. In this research, we assessed the differences between the 'gold standard' SA as determined using the Frank VGG and the SA as determined using different reduced lead systems (RLSs). The random error component (span of the Bland-Altman 95% limits of agreement) of the differences between the 'gold standard' SA and the SA values based upon the different RLSs was quantified. This was performed for all 62 RLSs that can be constructed from Mason-Likar (ML) limb leads I, II and all possible precordial lead subsets that contain between one and five of the precordial leads V1 to V6. The RLS with the smallest lead set size that produced SA estimates of a quality similar to what is achieved using the ML 12-lead ECG was based upon ML limb leads I, II and precordial leads V1, V3 and V6. The random error component (mean [95% confidence interval]) associated with this RLS and the ML 12-lead ECG were found to be 40.74° [35.56°-49.29°] and 39.57° [33.78°-45.70°], respectively. Our findings suggest that a RLS that is based upon the ML limb leads I and II and the three best precordial leads can yield SA estimates of a quality similar to what is achieved when using the complete ML 12-lead ECG.

Advancing regulatory science to bring novel medical devices for use in emergency care to market: the role of the Food and Drug Administration.

The Food and Drug Administration (FDA) performs regulatory science to provide science-based medical product regulatory decisions. This article describes the types of scientific research the FDA's Center for Devices and Radiological Health performs and highlights specific projects related to medical devices for emergency medicine. In addition, this article discusses how results from regulatory science are used by the FDA to support the regulatory process as well as how the results are communicated to the public. Regulatory science supports the FDA's mission to assure safe, effective, and high-quality medical products are available to patients.

Computer simulations to investigate the causes of T-wave notching.

Drugs that cause strong hERG potassium channel block (e.g., dofetilide, quinidine) cause T-wave notching. It has been suggested that this is due to prolongation of mid-myocardial (M) cells' action potential duration relative to endocardial and epicardial cells. However, the role of M cells in intact human hearts is debated. We simulated 2025 electrocardiograms representing changes in ventricular action potentials using the equivalent double layer mode that does not include M-cells. Action potential changes included prolongation, triangularization, squaring, and bumps in late repolarization, which have been observed experimentally and in single cell models with block of the hERG potassium channel. Changes were applied globally and spatially dispersed. Action potential bumps (slowing in late repolarization) produced T-wave notching similar to that observed clinically in healthy subjects receiving dofetilide or quinidine. Conversely, all other action potential changes (i.e., prolongation, triangularization, squaring), either global or spatially dispersed, resulted in T-wave changes, but did not cause T-wave notching. This study demonstrates that M-cells are not required to simulate T-wave notching.

The derivation of the spatial QRS-T angle and the spatial ventricular gradient using the Mason-Likar 12-lead electrocardiogram.

Research has shown that the 'spatial QRS-T angle' (SA) and the 'spatial ventricular gradient' (SVG) have clinical value in a number of different applications. The determination of the SA and the SVG requires vectorcardiographic data. Such data is seldom recorded in clinical practice. The SA and the SVG are therefore frequently derived from 12-lead electrocardiogram (ECG) data using linear lead transformation matrices. This research compares the performance of two previously published linear lead transformation matrices (Kors and ML2VCG) in deriving the SA and the SVG from Mason-Likar (ML) 12-lead ECG data. This comparison was performed through an analysis of the estimation errors that are made when deriving the SA and the SVG for all 181 subjects in the study population. The estimation errors were quantified as the systematic error (mean difference) and the random error (span of the Bland-Altman 95% limits of agreement). The random error was found to be the dominating error component for both the Kors and the ML2VCG matrix. The random error [ML2VCG; Kors; result of the paired, two-sided Pitman-Morgan test for statistical significance of differences in the error variance between ML2VCG and Kors] for the vectorcardiographic parameters SA, magnitude of the SVG, elevation of the SVG and azimuth of the SVG were found to be [37.33°; 50.52°; p<0.001], [30.17mVms; 39.09mVms; p<0.001], [36.77°; 47.62°; p=0.001] and [63.45°; 80.32°; p<0.001] respectively. The findings of this research indicate that in comparison to the Kors matrix the ML2VCG provides greater precision for estimating the SA and SVG from ML 12-lead ECG data.

Recent advances in medical device triage technologies for chemical, biological, radiological, and nuclear events.

In 2010, the US Food and Drug Administration (Silver Spring, Maryland USA) created the Medical Countermeasures Initiative with the mission of development and promoting medical countermeasures that would be needed to protect the nation from identified, high-priority chemical, biological, radiological, or nuclear (CBRN) threats and emerging infectious diseases. The aim of this review was to promote regulatory science research of medical devices and to analyze how the devices can be employed in different CBRN scenarios. Triage in CBRN scenarios presents unique challenges for first responders because the effects of CBRN agents and the clinical presentations of casualties at each triage stage can vary. The uniqueness of a CBRN event can render standard patient monitoring medical device and conventional triage algorithms ineffective. Despite the challenges, there have been recent advances in CBRN triage technology that include: novel technologies; mobile medical applications ("medical apps") for CBRN disasters; electronic triage tags, such as eTriage; diagnostic field devices, such as the Joint Biological Agent Identification System; and decision support systems, such as the Chemical Hazards Emergency Medical Management Intelligent Syndromes Tool (CHEMM-IST). Further research and medical device validation can help to advance prehospital triage technology for CBRN events.

An electrocardiographic sign of ischemic preconditioning.

Ischemic preconditioning is a form of intrinsic cardioprotection where an episode of sublethal ischemia protects against subsequent episodes of ischemia. Identifying a clinical biomarker of preconditioning could have important clinical implications, and prior work has focused on the electrocardiographic ST segment. However, the electrophysiology biomarker of preconditioning is increased action potential duration (APD) shortening with subsequent ischemic episodes, and APD shortening should primarily alter the T wave, not the ST segment. We translated findings from simulations to canine to patient models of preconditioning to test the hypothesis that the combination of increased [delta (Δ)] T wave amplitude with decreased ST segment elevation characterizes preconditioning. In simulations, decreased APD caused increased T wave amplitude with minimal ST segment elevation. In contrast, decreased action potential amplitude increased ST segment elevation significantly. In a canine model of preconditioning (9 mongrel dogs undergoing 4 ischemia-reperfusion episodes), ST segment amplitude increased more than T wave amplitude during the first ischemic episode [ΔT/ΔST slope = 0.81, 95% confidence interval (CI) 0.46-1.15]; however, during subsequent ischemic episodes the T wave increased significantly more than the ST segment (ΔT/ΔST slope = 2.43, CI 2.07-2.80) (P < 0.001 for interaction of occlusions 2 vs. 1). A similar result was observed in patients (9 patients undergoing 2 consecutive prolonged occlusions during elective percutaneous coronary intervention), with an increase in slope of ΔT/ΔST of 0.13 (CI -0.15 to 0.42) in the first occlusion to 1.02 (CI 0.31-1.73) in the second occlusion (P = 0.02). This integrated analysis of the T wave and ST segment goes beyond the standard approach to only analyze ST elevation, and detects cellular electrophysiology changes of preconditioning.

Mechanisms of sex and age differences in ventricular repolarization in humans.

INTRODUCTION: Corrected QT interval (QTc) is shorter in postpubertal men than in women; however, QTc lengthens as men age and testosterone levels decrease. Animal studies have demonstrated that testosterone decreases L-type calcium current and increases slow delayed rectifier potassium current; however, it is not known how these contribute to QTc differences by sex and age in humans. We separately analyzed early versus late repolarization duration and performed simulations of the effect of testosterone on the electrocardiogram (ECG) to examine the mechanism of sex and age differences in QTc in humans. METHODS: Twelve-lead ECGs from 2,235 healthy subjects (41% women) in Thorough QT studies were analyzed to characterize sex- and age-dependent differences in depolarization (QRS), early repolarization (J-T(peak)), and late repolarization (T(peak)-T(end)). In addition, we simulated the effects of testosterone on calcium current, slow delayed rectifier potassium current, and surface ECG intervals. RESULTS: QTc was shorter in men than in women (394 ± 16 vs 408 ± 15 milliseconds, P < .001), which was due to shorter early repolarization (213 ± 16 vs 242 ± 16 milliseconds, P < .001), as men had longer depolarization (94 ± 7 vs 89 ± 7 milliseconds, P < .001) and longer late repolarization (87 ± 10 vs 78 ± 9 milliseconds, P < .001). Sex difference in QTc decreased with age and was due to changes in early repolarization. Simulations showed that the early repolarization changes were most influenced by testosterone's effect on calcium current. CONCLUSION: Shorter QTc in men compared to women is explained by shorter early repolarization, and this difference decreases with age. These sex and age differences in repolarization appear to be caused by testosterone effects on calcium current.

Clinical Evaluation of the ECG Noise Extraction Tool as a Component of ECG Analysis Algorithms Evaluation.

OBJECTIVE: The aim of this work is to demonstrate the performance of the ECG noise extraction tool (ECGNExT) which provides estimates of ECG noise that are not significantly different from the inherent noise in an ECG generated by motion artifacts and other sources. In addition, this paper elaborates on use of ECGNExT in an algorithm evaluation context comparing two QRS detection algorithms. METHODS: 140 simultaneous pairs of clean ECGs and ECGs corrupted with motion-induced noise from 29 participants under five different and separate motion conditions were collected and analyzed. Estimates of the noise component of the ECGs recorded with noise were obtained using ECGNExT and were then added to the clean ECGs yielding estimated ECGs with noise. Root mean squared error (RMSE) between the recorded and estimated ECGs with noise was calculated for temporal comparison, and band powers of the signals were calculated for spectral comparison. RESULTS: A t-test revealed that the mean RMSE < 150-microvolts with p-value < 0.001 and, and equivalence tests showed that the band powers of the two ECGs were statistically equivalent with . CONCLUSION: ECGNExT can reliably estimate the underlying ECG noise while preserving temporal and spectral features. SIGNIFICANCE: We previously proposed ECGNExT as a component of ECG analysis algorithm testing during noise conditions and reported its performance based on simulated ECG data. This work provides additional support of the performance and functionality of the ECGNExT algorithm from a study with pairs of simultaneously recorded ECGs with and without noise from human subjects.

Evaluating strict and conventional left bundle branch block criteria using electrocardiographic simulations.

AIMS: Left bundle branch block (LBBB) is a critical predictor of patient benefit from cardiac resynchronization therapy (CRT), but recent studies suggest that one-third of patients diagnosed with LBBB by conventional electrocardiographic (ECG) criteria may have a false-positive diagnosis. In this study, we tested the hypothesis that recently proposed strict LBBB ECG criteria improve specificity in cases of left ventricular hypertrophy (LVH) /dilatation and incomplete LBBB. METHODS AND RESULTS: We developed five heart models based on a healthy male with increasing degrees of LV hypertrophy and/or dilation. With each model, we simulated six conduction types: normal conduction, four increments of delayed initiation of LV activation (incomplete LBBB), and complete LBBB. Simulated ECGs were evaluated for the presence of LBBB by conventional (LV conduction delay and QRSd ≥120 ms) and strict ECG criteria (LV conduction delay, QRSd ≥140 ms men or ≥130 ms women, and mid-QRS notching in at least two of the leads I, aVL, V1, V2, V5, and/or V6). Both conventional and strict LBBB criteria had 100% sensitivity. However, conventional criteria falsely diagnosed LBBB in cases with LVH + LV dilated 10 mm, LVH or LV dilated 10 mm combined with LV initiation ≥6 ms after the right ventricle (RV), and with LV dilated 5 mm combined with LV initiation ≥12 ms after RV (48% specificity). Strict LBBB criteria resulted in no false positives (100% specificity). CONCLUSIONS: New strict LBBB criteria increase the specificity of complete LBBB diagnosis in the presence of LV hypertrophy/dilatation and incomplete LBBB, which is critical for selecting CRT patients.

Development of an automated method for display of ischemic myocardium from simulated electrocardiograms.

BACKGROUND: Knowledge of the size and location of ischemic myocardium during acute coronary occlusion could provide decision support before reperfusion therapy. Electrocardiogram (ECG) scores based on the number of leads and the sum of ST-segment elevation have been unreliable in quantifying ischemia. We aimed to develop a new method to graphically display ischemic myocardium from simulated ECGs (DIMS-ECG) associated with known ischemic regions. METHODS: Twenty-one patterns of ischemia based on normal coronary anatomy were programmed into the freely available program ECGSIM (www.ecgsim.org). Minor variations of these patterns and 5 levels of ischemia severity produced 45 455 ECGs; 1000 normal ECGs were also added. Given a de novo ECG (an ECG from a patient), ST-segment and T-wave measurements are compared with ECG measurements in the database. The closest 200 matches are selected, and the corresponding ischemic areas are "averaged" to create a graphical display of the ischemic myocardium. RESULTS: Three patients are presented who underwent elective coronary angioplasty with continuous ECG recording and scintigraphically defined ischemic myocardium. Based on ECG analysis, the program graphically displays the ischemic myocardium with close agreement to the scintigraphic images. The program's source code and the ECG database will be made freely available. CONCLUSIONS: The DIMS-ECG method graphically displays ischemic myocardium from information contained in the 12-lead ECG based on a novel approach to use a large simulated database instead of rule- or score-based method. After further development and testing, the DIMS-ECG method could be used to risk stratify patients with acute myocardial infarction.

Comprehensive Translational Assessment of Human-Induced Pluripotent Stem Cell Derived Cardiomyocytes for Evaluating Drug-Induced Arrhythmias.

Induced pluripotent stem cell-derived cardiomyocytes (iPSC-CM) hold promise for assessment of drug-induced arrhythmias and are being considered for use under the comprehensive in vitro proarrhythmia assay (CiPA). We studied the effects of 26 drugs and 3 drug combinations on 2 commercially available iPSC-CM types using high-throughput voltage-sensitive dye and microelectrode-array assays being studied for the CiPA initiative and compared the results with clinical QT prolongation and torsade de pointes (TdP) risk. Concentration-dependent analysis comparing iPSC-CMs to clinical trial results demonstrated good correlation between drug-induced rate-corrected action potential duration and field potential duration (APDc and FPDc) prolongation and clinical trial QTc prolongation. Of 20 drugs studied that exhibit clinical QTc prolongation, 17 caused APDc prolongation (16 in Cor.4U and 13 in iCell cardiomyocytes) and 16 caused FPDc prolongation (16 in Cor.4U and 10 in iCell cardiomyocytes). Of 14 drugs that cause TdP, arrhythmias occurred with 10 drugs. Lack of arrhythmic beating in iPSC-CMs for the four remaining drugs could be due to differences in relative levels of expression of individual ion channels. iPSC-CMs responded consistently to human ether-a-go-go potassium channel blocking drugs (APD prolongation and arrhythmias) and calcium channel blocking drugs (APD shortening and prevention of arrhythmias), with a more variable response to late sodium current blocking drugs. Current results confirm the potential of iPSC-CMs for proarrhythmia prediction under CiPA, where iPSC-CM results would serve as a check to ion channel and in silico modeling prediction of proarrhythmic risk. A multi-site validation study is warranted.

Robust algorithm to locate heart beats from multiple physiological waveforms by individual signal detector voting.

Alarm fatigue is a top medical device hazard in patient monitoring that could be reduced by merging physiological information from multiple sensors, minimizing the impact of a single sensor failing. We developed a heart beat detection algorithm that utilizes multi-modal physiological signals (e.g. electrocardiogram, blood pressure, stroke volume, photoplethysmogram and electro-encephalogram) by merging the heart beats obtained from signal-specific detectors. We used the PhysioNet/Computing in Cardiology Challenge 2014 training set to develop the algorithm, and we refined it with a mix of signals from the multiparameter intelligent monitoring in intensive care (MIMIC II) database and artificially disrupted waveforms. The algorithm had an average sensitivity of 95.67% and positive predictive value (PPV) of 92.28% when applied to the PhysioNet/Computing in Cardiology Challenge 2014 200 record training set. On a refined dataset obtained by removing 5 records with arrhythmias and inconsistent reference annotations we obtained an average sensitivity of 97.43% and PPV of 94.17%. Algorithm performance was assessed with the Physionet Challenge 2014 test set that consisted of 200 records (each up to 10 min length) containing multiple physiological signals and reference annotations verified by the PhysioNet/Computing in Cardiology Challenge 2014 organizers. Our algorithm had a sensitivity of 92.74% and PPV of 87.37% computed over all annotated beats, and a record average sensitivity of 91.08%, PPV of 86.96% and an overall score (average of all 4 measures) of 89.53%. Our algorithm is an example of a data fusion approach that can improve patient monitoring and reduce false alarms by reducing the effect of individual signal failures.

The mixing rate of the arterial blood pressure waveform Markov chain is correlated with shock index during hemorrhage in anesthetized swine.

Identifying the need for interventions during hemorrhage is complicated due to physiological compensation mechanisms that can stabilize vital signs until a significant amount of blood loss. Physiological systems providing compensation during hemorrhage affect the arterial blood pressure waveform through changes in dynamics and waveform morphology. We investigated the use of Markov chain analysis of the arterial blood pressure waveform to monitor physiological systems changes during hemorrhage. Continuous arterial blood pressure recordings were made on anesthetized swine (N=7) during a 5 min baseline period and during a slow hemorrhage (10 ml/kg over 30 min). Markov chain analysis was applied to 20 sec arterial blood pressure waveform segments with a sliding window. 20 ranges of arterial blood pressure were defined as states and empirical transition probability matrices were determined for each 20 sec segment. The mixing rate (2(nd) largest eigenvalue of the transition probability matrix) was determined for all segments. A change in the mixing rate from baseline estimates was identified during hemorrhage for each animal (median time of 13 min, ~10% estimated blood volume, with minimum and maximum times of 2 and 33 min, respectively). The mixing rate was found to have an inverse correlation with shock index for all 7 animals (median correlation coefficient of -0.95 with minimum and maximum of -0.98 and -0.58, respectively). The Markov chain mixing rate of arterial blood pressure recordings is a novel potential biomarker for monitoring and understanding physiological systems during hemorrhage.

Comparison of the relation between left ventricular anatomy and QRS duration in patients with cardiomyopathy with versus without left bundle branch block.

QRS duration (QRSd) is used to diagnose left bundle branch block (LBBB) and is important to determine cardiac resynchronization therapy eligibility. The same QRSd thresholds established decades ago are used for all patients. However, significant interpatient variability of normal QRSd exists, and individualized QRSd thresholds might improve diagnosis and intervention strategies. Previous work reported left ventricular (LV) mass and papillary muscle location predicted QRSd in healthy subjects, but the relation in diseased ventricles is unknown. The aim of the present study was to determine the association between LV anatomy and QRSd in patients with cardiomyopathy. Patients referred for primary prevention implantable defibrillators (n = 166) received cardiac magnetic resonance imaging, and those with normal conduction (without bundle branch or fascicular block) and LBBB were studied. The LV mass, length, internal diameter, LV end-diastolic volume, septal and lateral wall thicknesses, and papillary muscle location were measured. In patients with normal conduction, LV length (r = 0.35, p <0.001), mass (r = 0.32, p <0.001), diameter (r = 0.20, p = 0.03), and septal wall thickness (r = 0.20, p = 0.03) had positive correlations with QRSd. In patients with LBBB, LV length (r = 0.32, p = 0.03), mass (r = 0.39, p = 0.01), diameter (r = 0.34, p = 0.02), and LV end-diastolic volume (r = 0.32, p = 0.04) had positive correlations with QRSd. Contrary to previous studies in healthy subjects, papillary muscle angle (location) was not associated with QRSd in cardiomyopathy patients with normal conduction or LBBB. In conclusion, increasing LV anatomical measurements were associated with increasing QRSd in patients with cardiomyopathy. Future work should investigate the use of LV anatomical measurements in developing individualized QRSd thresholds for diagnosing conduction abnormalities such as LBBB and identifying candidates for cardiac resynchronization therapy.

Automatic ECG quality scoring methodology: mimicking human annotators.

An algorithm to determine the quality of electrocardiograms (ECGs) can enable inexperienced nurses and paramedics to record ECGs of sufficient diagnostic quality. Previously, we proposed an algorithm for determining if ECG recordings are of acceptable quality, which was entered in the PhysioNet Challenge 2011. In the present work, we propose an improved two-step algorithm, which first rejects ECGs with macroscopic errors (signal absent, large voltage shifts or saturation) and subsequently quantifies the noise (baseline, powerline or muscular noise) on a continuous scale. The performance of the improved algorithm was evaluated using the PhysioNet Challenge database (1500 ECGs rated by humans for signal quality). We achieved a classification accuracy of 92.3% on the training set and 90.0% on the test set. The improved algorithm is capable of detecting ECGs with macroscopic errors and giving the user a score of the overall quality. This allows the user to assess the degree of noise and decide if it is acceptable depending on the purpose of the recording.

Transformation of the Mason-Likar 12-lead electrocardiogram to the Frank vectorcardiogram.

Vectorcardiograpic (VCG) parameters can supplement the diagnostic information of the 12-lead electrocardiogram (ECG). Nevertheless, the VCG is seldom recorded in modern-day practice. A common approach today is to derive the Frank VCG from the standard 12-lead ECG (distal limb electrode positions). There is, to date no direct method that allows for a transformation from 12-lead ECGs with proximal limb electrode positions (Mason-Likar (ML) 12-lead ECG), to Frank VCGs. In this research, we develop such a transformation (ML2VCG) by means of multivariate linear regression on a training data set of 545 ML 12-lead ECGs and corresponding Frank VCGs that were both extracted surface potential maps (BSPMs). We compare the performance of the ML2VCG method against an alternative approach (2step method) that utilizes two existing transformations that are applied consecutively (ML 12-lead ECG to standard 12-lead ECG and subsequently to Frank VCG). We quantify the performance of ML2VCG and 2step on an unseen test dataset (181 ML 12-lead ECGs and corresponding Frank VCGs again extracted from BSPMs) through root mean squared error (RMSE) values, calculated over the QRST, between actual and transformed Frank leads. The ML2VCG transformation achieved a reduction of the median RMSE values for leads X (13.9µV; p<.001), Y (15.1µV; p<.001) and Z (2.6µV; p=.001) when compared to the 2step transformation. Our results show that the 2step method may not be optimal when transforming ML 12-lead ECGs to Frank VCGs. The utilization of the herein developed ML2VCG transformation should thus be considered when transforming ML 12-lead ECGs to Frank VCGs.