Predicting Electrocardiogram and Arterial Blood Pressure Waveforms with Different Echo State Network Architectures

Alarm fatigue caused by false alarms and alerts is an extremely important issue for medical staff in Intensive Care Units. The ability to predict and classify ECG and ABP patient waveforms can potentially help the staff and hospital systems better classify a patient’s waveforms and subsequent alarms. This project approaches this problem by evaluating the effectiveness of different Echo State Network (ESN) architectures at predicting ECG and ABP waveforms. Several architectures and metrics are evaluated showing similar performance between parallel ESN architectures and integrated architectures. The results also suggest potentially greater benefit of larger integrated reservoirs at predicting ECG waveforms and the adaptability of such models across individuals. Although there are limitations to this analysis, the work presented here offers a unique way of understanding and predicting a patient’s waveforms and provides suggestions for further extensions of this research.