(A) Study protocol for audio data collection at the Kenya Medical Research Institute (KEMRI), Nairobi and subsequent cough annotation at the University of Washington, Seattle. (B) Bar graphs represent total passive and voluntary coughs (including all recording devices) in the Nairobi cough dataset. The lighter shade in the bar graphs indicates a rejected cough due to ambient noise or audio distortion, and the darker shade represents selected coughs by group. Credit: Scientists progress (2024). DOI: 10.1126/sciadv.adi0282
What telltale features, most of which are inaudible to the human ear, distinguish one type of cough from another? Scientists are about to find out thanks to a new machine learning tool aimed at identifying the characteristic sounds of tuberculosis.
Cough is one of the main symptoms of respiratory infections. And because the pattern and frequency of coughing episodes differ between diseases, efforts are underway to develop a smartphone app sensitive enough to accurately discern coughs associated with tuberculosis.
For years, researchers have been searching for a high-tech, low-cost TB screening tool, particularly for use in resource-limited parts of the world where health care infrastructure is lacking and diagnostic tools are rare.
Tuberculosis incidence and mortality are rising again after years of decline, intensifying the need for accurate screening tools. Current gold standards for TB diagnosis include sputum culture or GeneXpert molecular testing. But even if these diagnoses are very accurate, their cost remains a concern in the regions of the world hardest hit by tuberculosis.
An international team of researchers is testing the hypothesis that tuberculosis’s unique profile and cough frequency can provide enough data to screen for this highly infectious bacterial disease using technology integrated into a smartphone app.
Currently in the investigation phase, the application is not yet ready for distribution. It’s currently a machine learning tool called TBscreen, but given the increasing number of TB cases worldwide, its development couldn’t have come at a more opportune time.
Write in Scientists progress, a team of collaborators from the University of Washington in Seattle and the Kenya Respiratory Disease Research Center in Nairobi has published data on their experimental application. The research team includes engineers and computer scientists as well as doctors and infectious disease experts.
When they captured audio of coughs through various microphones in TBscreen, the team found that TBscreen – the experimental app – and a smartphone microphone identified active TB more accurately than when audio from the cough was transmitted via expensive microphones.
“To investigate the characteristics of cough as an accurate classifier of TB versus non-TB related cough, we recruited adults with cough due to pulmonary TB and non-TB related etiologies to Nairobi, Kenya,” writes Manuja Sharma, an engineer at the University of Washington in Seattle.
The machine learning tool is being “trained” to recognize the pattern and frequency of coughs caused by TB. The experimental app is also being trained to distinguish coughs related to tuberculosis from those caused by other respiratory disorders.
Researchers found that there are many factors affecting the basic patterns of coughing, nuances (some inaudible to the human ear) that the tool must discern in order to accurately screen for TB.
“The mechanism of cough production varies depending on the properties of mucus, strength of respiratory muscles, mechanosensitivity, chemosensitivity of airways and other factors leading to various cough sounds,” added Sharma, author main point of the new analysis.
“We constructed a study design with minimal background noise and minimal environmental variability between control and TB groups to ensure that the model relies on differences in cough characteristics rather than ambient noise,” Sharma explained, referring to the app, a machine learning tool. .
The new cough classification technology was tested by analyzing 33,000 passive coughs and 1,200 forced coughs from 149 patients with pulmonary tuberculosis and 46 patients with other respiratory conditions. TBscreen was able to distinguish active TB coughs from non-TB coughs with an overall accuracy of approximately 82%. The team found that the app and a smartphone microphone more accurately predicted which coughs meant active TB than using more expensive microphones.
TBscreen performed better when using audio from a Pixel smartphone to assess passive cough and identify coughs in patients with higher bacterial loads. The last item indicates the potential of TBscreen as a triage tool, as patients with high bacterial load tend to be sicker.
All participants were enrolled in the study at the Kenya Medical Research Institute in Nairobi, where Dr Videlis Nduba led this part of the analysis. Each participant was asked to sit in a quiet room and allow the cough reflex to occur naturally. Their coughs were recorded for two hours. The group of 46 controls whose coughs were also recorded for two hours followed the same instructions.
This is not the first time that medical scientists have considered cough as a possible audio marker for the diagnosis, screening or monitoring of tuberculosis. Last year, scientists at the University of California, San Francisco reported the results of a study using a cell phone app to monitor cough frequency in patients being treated for tuberculosis.
The UC San Francisco research involved collaborators in Uganda, South Africa, India, the Philippines and Vietnam, all regions of the world where tuberculosis incidence remains significantly high. This application made it possible to collect data on patients’ coughs 24 hours a day.
Tuberculosis is caused by inhaling the bacteria Mycobacterium tuberculosis, and it is the second leading infectious cause of death after COVID, Sharma and his team reported.
In 2022, the most recent year for comprehensive World Health Organization statistics, tuberculosis infected 10.6 million people worldwide and killed about 1.4 million. Tuberculosis, which is spread largely through coughing and sneezing, has infected humans for at least 9,000 years and was for centuries the leading cause of death from infectious diseases. The COVID pandemic has pushed tuberculosis to second place.
“Our results support the feasibility of using a widely available recording device – smartphones – for point-of-care cough-based TB screening,” Sharma concluded, noting that the basic framework behind this screening strategy could also advance research in the diagnosis of tuberculosis. other lung diseases.
More information:
Manuja Sharma et al, TBscreen: A Passive Cough Classifier for TB Screening with a Controlled Dataset, Scientists progress (2024). DOI: 10.1126/sciadv.adi0282
© 2024 Science X Network
Quote: Can an experimental cell phone app detect cough for tuberculosis? Scientists say “yes” (February 7, 2024) retrieved February 7, 2024 from
This document is subject to copyright. Apart from fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for information only.