A scoping review of 79 studies identified three main categories for generating synthetic ECG data: mathematical modelling, computer vision inherited methods, and deep generative models.
This scoping review categorizes synthetic ECG generation methods into three main types and highlights the critical need for standardized evaluation metrics.
The scientific community has recently shown increasing interest in generating synthetic ECG data. In particular, synthetic ECG signals can be beneficial for understanding cardiac electrical activity, developing large and heterogeneous unbiased datasets, and anonymizing data to favour knowledge sharing and open science. In the present scoping review, various methodologies to generate synthetic ECG data have been thoroughly analysed, highlighting their limitations and possibilities. A total of 79 studies have been included and classified, depending on the methodology employed, the number of leads, the number of heartbeats, and the purpose of data synthesis. Three main categories have been identified: mathematical modelling, computer vision inherited methods, and deep generative models. This thorough analysis can assist in the choice of the most suitable technique for a specific application. The biggest challenge is identifying standardized metrics that can comprehensively and quantitatively assess the fidelity and variability of generated synthetic ECG data.
Zanchi et al. (2024) conducted a review in Synthetic ECG signals generation (n=79). Synthetic ECG generation methodologies was evaluated on Methodologies to generate synthetic ECG data. A scoping review of 79 studies identified three main categories for generating synthetic ECG data: mathematical modelling, computer vision inherited methods, and deep generative models.