Summary:
UCLA researchers in the Department of Electrical and Computer Engineering have developed a method to avoid expensive clinical trials by generating synthetic face videos.
Background:
Healthcare providers have increased their use of virtual diagnostic tools during the COVID-19 pandemic. These tools have allowed patients to undergo basic health assessments from the comfort and safety of their own home. However, some diagnostics, such as heart rate measurements via photoplethysmography (PPG), still require in-person visits or specific tools that the patient must have physical access to. PPG is a widely used optical and noninvasive diagnostic tool that measures blood volume pulse (BVP) by measuring blood volume changes at the skin surface. Currently, there are no remote photoplethysmography (rPPG) methods that a physician can utilize to measure BVP without a physical tool touching the patient. Recent advances have enabled novel deep-learning rPPG methods but these methods need copious amounts of data to ensure accuracy. Therefore, there is a need for a rPPG model that can generate synthetic rPPG videos with high fidelity to improve deep-learning rPPG methods.
Innovation:
UCLA researchers in the Department of Electrical and Computer Engineering have developed a model that can render realistic photoplethysmography videos to assist the development of rPPG methods. The invention can generate synthetic videos that can enable improved performance in state-of-the-art deep photoplethysmography methods. Furthermore, this method has been used by the research group to generate datasets for underrepresented groups to mitigate biases in rPPG associated with diverse skin tones. It can also produce videos with large variations, such as facial appearance, expression, head motion, and environmental lighting. This improvement would ensure that rPPG measurements can maintain their accuracy across edge cases.
Potential Applications:
• Cardiac Diagnostics
• Remote photoplethysmography r(PPG)
• Heart rate measurements
• Virtual diagnostics
• Telemedicine
Advantages:
• High fidelity and realism
• Can generate data with wide demographic ranges
• Adaptable to video and environmental variations
Development to Date:
First description of complete invention.