Objective Assessment of Pathological ECG Waveforms Using Gender-specific Normal Waveform Distributions.

Introduction: The objective definition of the normal ECG waveform is difficult, even 121 years after Einthoven's invention of the ECG. Consequently, ECG evaluation is primarily focused on interval measurements. For the waveforms, many rules and exceptions apply, making the interpretation of the ECG waveforms an art. In this study, a novel method is proposed that compares ECG waveforms with a distribution of normal ECG waveforms (WaveECG) and vectorcardiographic-derived positions of the electrical activity (PathECG) in order to determine the pathological status of the ECG. Methods: Human-annotated ECGs from the PhysioNet PTB-XL Diagnostic ECG database were used. ECGs classified as normal were used to create gender-specific distributions of the normal WaveECG and PathECG. The similarity between the evaluated ECG and the distributions was then quantified. Logistic regression was used to examine the potential of PathECG and WaveECG in the prediction of an abnormal ECG waveform. The independent China Physiological Signal Challenge database was used for validation. Results: A total of 14,381 ECGs were used to construct the distribution of the normal WaveECG and PathECG and to fit the model, and 6,320 independent ECGs were used for validation. The QRS model performance had an area under the receiver operating characteristic curve of 82.1% (95% CI [80.3-84%]) for male patients and 86.1% (95% CI [84.7-87.6%]) for female patients. The area under the receiver operating characteristic curve of the P wave model was 65.4% (95% CI [62.8-68.0%]) for female patients and 66.8% (95% CI [63.9-69.6%]) for male patients. Conclusion: An abnormal ECG can be detected in an explainable and visually understandable way by comparing the ECG to the distribution of normal ECGs. This method might significantly support the less experienced ECG reader in identifying patients requiring cardiac care.