Reservoir Computing Machine Learning for Real-Time Health Surveillance

­An energy efficient artificial intelligence (AI) platform for real-time patient monitoring and disease prediction using mixed signals. Background: Though the utility and value of wearable devices for continuous health monitoring is becoming incr…

An energy-efficient artificial intelligence (AI) platform for real-time patient monitoring and disease prediction using mixed signals. Background: Though the utility and value of wearable devices for continuous health monitoring are becoming increasingly evident, a number of challenges and limitations prevent the full exploitation of their potential. For example, existing devices lack the ability to integrate the electronic medical record (EMR) in real-time. These devices also tend to lack automated inference capability, instead of depending on telemetry and medical professionals for actionable inference. Further, attempts to address the latter through the utilization of AI analysis either increase the risk for patient data breaches during network transmission to the cloud or consume significant energy due to von-Neumann architecture, which separates mathematical operations and memory into two separate units, and the computationally intensive requirements of the algorithms involved. Technology Overview: This invention provides a reservoir computing platform possessing a simple architecture and reduced training requirements that make it a highly attractive candidate for resource-constrained wearable devices. The design of the platform breaks the von-Neumann bottleneck, thereby providing improved energy efficiencies. The output of this platform can be directly used as a classification result and displayed on a mobile device, or combined with EMR data through a cloud-based fusion AI model for monitoring and prediction of personalized health outcomes. To date, the platform has been tested for its ability to be applied to the prediction and/or detection of heart disease, cardiac stress, and sepsis. In each case, the platform has demonstrated superior accuracy and energy efficiency when compared to state-of-the-art digital classifiers. Advantages: The high-energy efficiency demonstrated by this platform enables integration with resource-constrained wearable devices.

Applications:

  • Real-time, continuous patient health monitoring and disease prediction, including onset
    • Cardiac stress
    • Coronary heart disease
    • Sepsis
    • Other

Intellectual Property Summary:

US Provisional Patent Application 63/234,529 filed August 18, 2021 Stage of Development: Prototype. Licensing Status: Available for licensing or collaboration. Additional Information: Publication link(s): IEEE Solid-State Circuits Letters, Vol. 3, 2020 https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9617636 https://pubmed.ncbi.nlm.nih.gov/3538323

Website

https://suny.technologypublisher.com/tech/Reservoir_Computing_Machine_Learning_for_Real-Time_Health_Surveillance

Contact Information

TTO Home Page: https://suny.technologypublisher.com

Name: Timothy Dee

Title: Associate Director

Email: tpdee@buffalo.edu

Phone: 716-645-8139