查看文章

RESULTS AND DISCUSSION: Our approach based on both reconstruction and the discriminators performs better than the baseline, which is informed only by MSE. We strongly believe that what makes this method particularly effective is implementing an architecture of both style and content as opposed to being restricted to feature extraction algorithms. This approach outperforms the baseline by reproducing signals that are 3.6 X closer to the original signals for wearable asymptomatic ECG signals, and 1.3 X closer to the original signals for clinical symptomatic ECG data by producing more accurate samples of unseen classes, with about a 1.6 X improvement in MSE. Our immediate next steps are to provide a comprehensive evaluation framework and metrics as well as compare the validation results by both an AI system and human specialised cardiologists.

ACKNOWLEDGEMENTS: The authors would like to thank NASA, CSA and the SETI Institute for their continuous support during this work. The authors are also grateful for the computational resources provided by IBM.