A Transformer-Based Diffusion Probabilistic Model for Vital Signs Forecasting

Case ID:
UA24-130
Invention:

The model proposed in this technology, Transformer-based Diffusion Probabilistic Model for Sparse Time Series Forecasting (TDSTF), is a novel deep-learning algorithm that can be used for vital signs forecasting, data synthesis, simulation, and digital twins. Trained and tested on data from over 46,000 patients, the TDSTF technology merges transformer and diffusion models to forecast vital signs. The TDSTF model showed state-of-the-art performance in predicting vital signs in the intensive care unit, outperforming other models’ ability to predict distributions of vital signs and being more computationally efficient.

Background: 
Intensive care units (ICUs) are vital for enhancing the survival of critically ill patients through the continuous monitoring and maintenance of their vital functions. Vital signs indicate the status of the patient’s life-threatening functions in the ICU. Continuous monitoring of patients’ vital signs and physiological functions aids in ensuring patient safety through awareness of critical changes in the patient’s health status, and it guides daily therapeutic interventions. This underscores the need for an accurate predictive system, as early recognition of patient deterioration and timely intervention are critical in saving patients’ lives. Vital sign monitoring in the ICU is crucial for enabling prompt interventions for patients. 

Applications: 

  • ICU vital sign monitoring 
  • General vital sign monitoring 
  • Clinical decision support


Advantages: 

  • Novel algorithm
  • Improved efficiency and efficacy over baseline models
Patent Information:
Contact For More Information:
Jay Martin
Licensing Associate, Software and Copyright
The University of Arizona
jaymartin@arizona.edu
Lead Inventor(s):
Ao Li
Ping Chang
Huayu Li
Janet Roveda
Keywords: