Advancing Early Detection of Sepsis With Temporal Convolutional Networks Using ECG Signals

In the hours leading up to the onset time of sepsis, the autonomic nervous system presents sub-clinical indicators of disease that may not be observable to providers. The objective of this research is to create an interpretable sepsis prediction algorithm utilizing continuous electrocardiography (ECG) signals, with the aim of implementing it in patient monitoring systems for individuals in intensive care units (ICU). We develop an early sepsis detection algorithm utilizing two datasets; in particular, the Medical Information Mart for Intensive Care (MIMIC-III) Clinical Dataset and MIMIC-Waveform Database. We carry out a systematic approach to selecting ECG segments of superior quality that are recorded in highly dynamic intensive care unit environments. Later, we use the single-lead ECG waveform to investigate the potential of heart rate variability (HRV) for continuous monitoring. In this study, 715 patients were included, of whom 65 are sepsis patients labeled with recent sepsis definition on an hourly basis. The predictive potential of the critical features is visualized to assist the interpretation of the model in a clinical practice. Moreover, since we are framing the early sepsis prediction as a supervised time series classification task, we evaluate the model performance by implementing Temporal Convolutional Networks (TCN). Performance analysis reported with varying prediction windows preceding sepsis onset time using area under the receiver-operating-characteristic curve (AUROC) and area under the precision-recall curve (AUPRC). The deep learning model delivers promising results by leveraging time series with the use of temporal convolutions. Our findings reveal that the HRV characteristics of adults can be a valuable indicator for continuous sepsis monitoring in an ICU. Finally, this research work adds to the field of early sepsis detection by providing an annotated continuous waveform dataset from the MIMIC-Waveform Database, which is made accessible to the public.