Commercially available heart sensors have been reported to last nearly 84 hours without needing battery replacements. These lifetimes are clearly not enough since long term monitoring for congenital heart diseases are typically prescribed for months if not years. Generative Model-based Resource Efficient ECG Monitoring (GeM-REM) leverages the morphologic predictability of ECG signals to considerably reduce wireless communication from a wearable ECG sensor to a desktop, laptop or a smart-phone and increases their lifetime by 40 fold. In principle, the GeMREM technique considers the periodicity of ECG signals and develops a generative model. The generative model if supplied with the correct parameters can generate synthetic physiological signals that are equivalent to the original signal in diagnostic content. Theoretical study with MIT BIH data shows that this technique can result in huge increase (40 fold) in lifetime and reduction in storage requirements while maintaining required accuracy. Relying on predictability, GeM-REM immediately recognizes changes in the ECG and thus it can promptly generate alerts based on pre-defined thresholds. Additionally, it can identify segments of ECG trace which are normal and abnormal. Such a classification can be extremely useful for cardiologists to quickly browse through lengthy ECG traces. GeM-REM drastically reduces wireless data transmission allowing several more wireless sensors to be in close proximity, e.g., in an ICU. Additionally, reduced wireless communication saves considerable energy and consequently extends the time between battery recharges or changes. In an ongoing clinical study with St. Luke's hospital we have deployed GeMREM enabled ECG sensors on 25 patients each monitored for 20 hours to analyze the feasibility of the technology in a hospital environment. However, GeMREM is intended for long term use and hence in this project we plan to deploy sensors at a hospital and home setting with frequent motion artifacts and distortions of ECG signals due to pathological conditions. In this project we aim to deploy sensors on 100 ICU patients at St. Luke's hospital and also require them to take home and monitor for a week. The main study aims are: a) to analyze the feasibility of using GeMREM sensors at home and hospital, and b) to obtain a statistically stable lifetime and storage needs.
Long term monitoring (possibly for months) of vital signs are essential for mitigating chronic diseases such as congenital heart disease, which costs the US $2 billion hospital expenditures a year. Current technology used in hospitals or home based cardiac monitoring can only last for 84 hours after which the user needs to download the data for analysis and replace batteries. In this proposal, we are testing the efficiency of GeMREM, a resource efficient monitoring technique that promises a 40 fold increase in the lifetime of ECG sensors and can revolutionize long term monitoring technology. This project builds upon initial technological clinical study at St. Luke's to establishing clear medical benefits of GEM-REM arising from long-term ECG collection and better presentation of information to cardiologist.
