Overview of the proposed machine learning-based acoustic model for hypertension screening. Abbreviations: BP – blood pressure; SBP – systolic blood pressure; DBP – diastolic blood pressure; LLD – low-level descriptor; LASSO – minimum absolute shrinkage and selection operator; SMOTE – synthetic minority oversampling technique. The subscripts I and F refer to the initial and final measurements, respectively. Credit: IEEE Access (2024). DOI: 10.1109/ACCESS.2024.3443688
Researchers at Klick Labs have unveiled a cutting-edge, non-invasive technique that can predict chronic high blood pressure with a high degree of accuracy using just a person’s voice. Just published in the journal IEEE AccessThese findings hold enormous potential for advancing the early detection of chronic high blood pressure and present another novel way to leverage vocal biomarkers for better health outcomes.
The 245 study participants were asked to record their voices up to six times a day for two weeks while speaking into a proprietary mobile app, developed by Klick scientists, which detected high blood pressure with up to 84% accuracy for women and 77% for men.
The app uses machine learning to analyze hundreds of speech biomarkers that are indistinguishable to the human ear, including pitch variability (fundamental frequency), speech energy distribution patterns (Mel frequency cepstral coefficients), and the sharpness of sound changes (spectral contrast).
“By leveraging diverse classifiers and building predictive models based on gender, we have discovered a more accessible way to detect hypertension, which we hope will lead to earlier intervention for this widespread global health problem. Hypertension can lead to a number of complications, from heart attacks and kidney problems to dementia,” said Yan Fossat, senior vice president of Klick Labs and principal investigator of the study.
More accessible screenings for the “silent killer”
The World Health Organization (WHO) calls hypertension a “silent killer” and a global public health problem that affects more than 25% of the world’s population. Half of them are unaware of their condition and more than 75% of those diagnosed live in low- and middle-income countries.
Conventional methods of measuring blood pressure (and therefore detecting hypertension) include the use of a cuff (sphygmomanometry) or an automatic blood pressure measuring device. However, these methods may require technical expertise, specialized equipment, and may not be readily available to people living in underserved areas.
This study marks Klick Labs’ first foray into using voice technology to identify conditions beyond diabetes, as the company expands its research to evaluate the effectiveness of its AI algorithms in detecting and managing a broader range of conditions.
Klick Labs has been collaborating with hospitals, academic institutions and public health authorities around the world since its research revealed that voice analysis combined with AI can accurately screen people for type 2 diabetes. Mayo Clinic Proceedings: Digital Health in October 2023. Last week, Scientific reports published another study from Klick Labs confirming the link between blood glucose levels and voice pitch.
“Voice technology has the potential to exponentially transform healthcare, making it more accessible and affordable, especially for large and underserved populations,” said Jaycee Kaufman, a research scientist at Klick Labs and co-author of the study.
“Our ongoing research increasingly demonstrates the considerable value of vocal biomarkers for detecting hypertension, diabetes and a growing list of other health problems.”
More information:
Behrad Taghibeyglou et al., Hypertension screening by machine learning using acoustic speech analysis: model development and validation, IEEE Access (2024). DOI: 10.1109/ACCESS.2024.3443688
Provided by Klick Applied Sciences
Quote: Scientists use AI to detect chronic high blood pressure in people’s voice recordings (2024, September 10) retrieved September 10, 2024 from
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.