Automated Detection of Ventricular Heartbeats from Electrocardiogram (ECG) Acquired During Magnetic Resonance Imaging (MRI)

ECGs are highly distorted by the MRI environment, making automated ECG analysis highly difficult. This study aimed at implementing a machine-learning (ML) based heartbeat classifier, using hand-crafted features, for the automatic detection of ventricular heartbeats during MRI. A model was trained on the MIT-BIH Arrhythmia Database and assessed on an in-house database of ECG acquired inside a 1.5T MRI (ECG-MRI). Features were extracted for each heartbeat from single-lead ECG signals including QRS morphological features based on Hermite functions, and RR interval-based features. A support vector machine was trained to classify normal (N) and ventricular ectopic beats (V‘). The classifier achieved F1 scores of 0.85 on the V' class on the validation fold on the MIT-BIH database, while it only achieved F1 scores of 0.15 on the ECG-MRI database. The proposed heartbeat classifier was developed on the MIT-BIH arrhythmia database using temporal features and QRS morphological features based on the assumption they would be less distorted by the MRI environment. However, even if performance on MIT-BIH were acceptable (although slightly lower than state-of-the-art approaches), results were poor on the ECG-MRI database. The results highlight the need for further developments by suppressing MRI-related artifacts, and by retraining on MRI specific datasets.