New Hyfe AI Study Validates the Accuracy of AI-Powered Cough Detection Human Annotators

Wilmington, Del., December 12, 2023 -- Hyfe, Inc., the global leader in AI-powered cough detection, tracking, and classification, today announces breakthrough results from a performance evaluation study comparing human cough annotators to emerging automated technologies using AI and machine learning. The BMJ Open Respiratory Research study demonstrates Hyfe’s cough detection AI technology is nearly as accurate as the human ear in analyzing the amount and duration of cough in a real-world environment. 

The study analyzed 40 hours of audio randomly collected from participants wearing audio recording smartwatches throughout an average day. The audio samples were manually reviewed and annotated twice by one expert human annotator and some samples of the 40 hours were annotated a third time by six of Hyfe’s expert annotators. Key insights from the study include:

  • After evaluating several ways to track cough, researchers found that tracking cough by cough seconds, which are defined as any second of time that contains at least one cough, rather than counting individual coughs, decreased annotator discrepancy by 50 percent. 
  • Compared to counting individual coughs, the study proposes that using cough rate over time is a more clinically relevant and reproducible performance metric for evaluating automatic systems.
  • Hyfe’s labeling software was reported by labelers to be easy to use, an improvement over Audacity, and likely to lead to fewer errors in data management.
  • The study provides guidance for researchers and developers working on these technologies and has the potential to lead to more reliable and consistent automatic cough monitoring tools.

Notably, the study is also the first to observe sex differences in cough, observing for the first time that the duration of cough sounds and epoch size differed between male and female participants. The study found that women tend to have shorter cough sounds but more coughs in each episode compared to men. These observed differences in cough characteristics could have important implications for the development of cough monitoring tools. They will be key in better understanding how diseases spread, diagnosing illnesses, and how people seek medical help.

“We’re thrilled that the results of this study confirm the accuracy of our robust automated cough monitoring technology,” said Joe Brew, CEO of Hyfe AI. “These findings pave the way for more accurate, reliable, and clinically relevant methods for cough tracking. They provide valuable insights that could significantly impact the much-needed development for monitoring coughs in healthcare and clinical trial settings. These are breakthrough discoveries in the chronic cough space and we look forward to taking an even deeper look into cough in future studies like this.” 

Hyfe is the leader in longitudinal cough monitoring with over 700 million sounds in its cough database. It can track and detect cough with 90%+ accuracy on any mic-enabled device with no need for patient intervention while preserving patient privacy.

For more information, visit hyfe.com

About Hyfe

Hyfe, Inc. is the global leader in AI-powered cough detection and classification that provides insight into cough patterns and correlations and is being widely used to help patients gain a better understanding of their cough and have more informed conversations with their providers. With more than 700 million samples, Hyfe maintains the largest cough dataset in the world enabling the building of powerful models to track, manage and diagnose respiratory illnesses. Hyfe provides platforms and data for pharmaceutical companies, medical researchers, government agencies, health care providers and patients and has partnerships with leading academic institutions including Johns Hopkins University and the University of California at San  Francisco. The company was founded in 2020 and is headquartered in Wilmington, Delaware. More information is available at hyfe.com on social media @hyfeapp and LinkedIn at /hyfe.

###

Contact: 

[email protected]

You might also like: