More than four million Americans experience cardiac arrhythmia. Arrhythmia can lead to morbidity and mortality and is a substantial economic burden. Electrocardiogram(ECG) monitoring is widely used to detect arrhythmia. In 1996, it was estimated that 300 million ECGs will be recorded annually and this number is increasing due to aging populations and availability of easy-to-use wearable devices. Due to the volume of recorded ECGs, manual interpretation for all people under monitoring might not be a feasible and scalable solution. Therefore, machine learning (ML) algorithms are widely used for automatic ECG interpretation. ECG monitoring using ML algorithms is advancing as automatic diagnosis of arrhythmia is challenging due factors such as inter-subject variability and poor signal to noise.
In this session, the challenges and differences between ML techniques will be reviewed, from traditional ML classifiers to deep learning (DL) and their combination. The requirement for all supervised ML methods is the availability of annotated data that is typically provided by a domain expert such as cardiologists. A challenge of traditional ML is the pre-processing and feature engineering that is required for quantification of arrhythmia. However, traditional ML methods typically allow interpretation of developed models based on physiology. On the other hand, the reduced feature engineering efforts of DL-based approaches may allow novel feature discovery and learning hidden patterns that might not be clear using traditional ML methods. However, interpretation of DL models is challenging. Combination of traditional ML and DL has been applied in recent research and may allow the benefits of both approaches when applied to ECG monitoring. In conclusion, it is not trivial to pick the best ML method for ECG monitoring and interpretation. Promising results achieved by combining traditional ML with novel DL approaches and improvement in interpretability of these combined models may increase the popularity to these techniques.