Voice and breath data provided good diagnostic accuracy in the prediction of FVC and FEV1 and may be able to detect and monitor lung function, according to a study presented at virtual Chest 2020.
Voice is an important biomarker of many medical conditions, in particular, diseases of the respiratory system that leave signatures in breath and speech. Researchers hypothesized that voice and breath sounds in human speech are strongly correlated to lung function. Using voice as a biomarker enables noninvasive and accurate detection of lung function.
Patients were recruited into the study during their regularly scheduled pulmonary function tests at a large suburban general hospital in Western Pennsylvania. The prospective cohort study collected cross-sectional data from participants where they provide pre- and post-pulmonary function testing (PFT) voice and breath sound samples, which correspond to pre- and post-bronchodilator samples. Researchers collected pre- and post-PFT FEV1 and FVC results from all patients.
A total of 133 participants aged 18 to 85 years took part. Voice and breath audio samples were recorded on a smart tablet using a proprietary software application developed by the researchers. Data was analyzed offline on the secure Telling.ai cloud. Researchers found that automated voice and breath analysis delivers good diagnostic accuracy in the prediction of FVC (R squared of 0.785, MSE of 0.192) and FEV1 (R squared of 0.745 and MSE of 0.170). The classification of obstructed versus non-obstructed participants with 82% accuracy and F-score of 0.80.
The technology used requires no additional custom-built hardware, is cost-effective, noninvasive and practical for abundant and frequent use. The data can be collected passively, such as during conversation, using widespread, existing infrastructure such as mobile phones. The solution is ideally suited for personalized, frequent monitoring and analysis of an individual’s lung function to enable the development of clinical decision support systems and advance personalized medicine for chronic and acute respiratory care.
“The technique offers the promise of widespread, affordable medical screening and real-time monitoring of respiratory disease. Voice and breath data have proven an ideal medium to detect and monitor lung functioning, especially respiration impairment at scale,” said Obaid Ashraf, MD, lead researcher and Chest 2020 presenter.