SCH: Explainable Learning of Heart Actions from Pulse to Broaden Cardiovascular Healthcare Access

Cardiovascular disease is the most prevalent cause of death. Early treatment can effectively reduce the risk of sudden cardiac death, but a many cardiac issues show no obvious symptoms in the early stage and would benefit from long-term continuous cardiac monitoring to capture the intermittent and asymptomatic abnormalities of the heart. This disproportionately affects low-income and disadvantaged populations, who have limited access to affordable preventive care.

An electrocardiogram (ECG) is a non-invasive gold standard for diagnosing cardiovascular diseases. Although it is currently possible to obtain an instant ECG test through a special smartwatch or special attachment to a smartphone, these current options require continuous user participation and are impractical to meet the needs of long-term continuous monitoring.

This project investigates a new Artificial Intelligence (AI) powered health solution to automated and continuous cardiac monitoring by inferring ECG from the readily available continuous measurements, such as those sharing the same principles as in many wearable devices. The research from this project will provide insights on how to transfer the ECG-based rich knowledge base to the diagnosis of cardiovascular diseases from wearable sensors.

In order to broaden participation and impact, the project will integrate research and educational activities. These include supporting the workforce development in such in-demand technical areas as machine learning and smart health, and actively engaging students in hands-on and exploratory interdisciplinary research, especially those from the under-represented groups. The project will contribute to promoting national health, welfare, and prosperity.

The key research issues of inferring ECG from photoplethysmogram (PPG), which can be monitored continuously without constant user attention, include: (1) how to apply biomedical insights to model the relations between ECG and PPG;  (2) how to carry out explainable learning for inferring ECG from PPG;  (3) how to make a transformative expansion of public health knowledge based on the newly developed bridge between ECG and PPG; and (4) how to address a variety of diverse and practical conditions, including population diversity, disease progression, and noise/distortions in real-world PPG sensing sources. The investigator team plans to carry out the core inference from PPG to ECG in several stages, starting with modeling the biophysical relation between ECG and PPG and representing both waveform families through the well-understood basis in the Fourier family as a proof-of-concept.

The team plans to utilize data next to refine the representation using dictionary learning, and incorporate a deep model when extensive data can be leveraged to provide a refined inference. The bridge between ECG and PPG enabled by explainable AI can bring unprecedented opportunities to expand smart health knowledge to benefit public health. The investigator team will work closely with a medical expert to explore AI-enabled understanding and promotion of cardiovascular health in exercise physiology, and transferring rich ECG medical knowledge base to the more user-friendly PPG domain.

The team plans to embrace the opportunity of cross-disciplinary collaboration to evaluate the new capabilities in practical settings as well as promote participation and feedback from a diverse population.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.