Speech recognition tech diagnoses respiratory disorders from coughs

App nearly as accurate as a panel of paediatricians

A smartphone app that records children's coughs has been found to be almost as effective at diagnosing common childhood respiratory disorders as a panel of paediatricians who had met with the child and reviewed X-rays, lab results and hospital charts.

In a paper published today, researchers from Curtin University and The University of Queensland demonstrate how they developed algorithms to detect coughs from a recording and produce a high accuracy diagnosis for asthma, croup, pneumonia, lower respiratory tract disease and bronchiolitis.

The work, published in open access journal Respiratory Research, was in two parts. The researchers first developed an automatic cough detector to identify and extract cough sounds in a continuous audio stream. This was done using a Time Delay Neural Network operating and identifying Mel Frequency Cepstral Coefficients (MFCC), a common technique used in speech recognition systems to, for example, distinguish spoken words from background noise.

A diagnostic algorithm was then developed to classify and group alike cough sounds, using a neural network trained to diagnose target diseases working on an initial dataset of 852 coughs. The algorithms were optimised in combination with answers to five questions put to parents about their child's symptoms, for example how many days they'd been coughing and if they had a fever.

The algorithms were built into an app – developed by ResApp Health in which a number of the paper's authors have a stake or role – which was then used to categorise the coughs of 585 children aged between 29 days and 12 years who were being cared for at two hospitals in Western Australia.

The app was found to have a high accuracy of between 81 per cent and 97 per cent when compared to paediatricians' diagnosis. The app proved more accurate in diagnosing some disorders than others, with the results for asthma and pneumonia being the most accurate.

"The results indicate that this technology has a role as a high-level diagnostic aid in the assessment of common childhood respiratory disorders," said author Dr Paul Porter.

“It can be difficult to differentiate between respiratory disorders in children, even for experienced doctors. This study demonstrates how new technology, mathematical concepts, machine learning and clinical medicine can be successfully combined to produce completely new diagnostic tests utilising the expertise of several disciplines,” he added.

Diagnosis of respiratory issues typically requires examinations, medical imaging, bronchodilator-response testing, spirometry and body fluid analyses. Undertaking all the necessary tests is not always possible, particularly in remote areas where access to expertise and clinical testing is limited. Even in well-equipped hospitals, the testing can be time consuming and expensive.

"The technology can be installed onto ubiquitous devices, agrees with existing standard-of-care clinical diagnosis and provides a point-of-care diagnosis without the need for clinical examination, supplemental investigations, or bronchodilator testing…Its use in different settings such as hospitals, ambulatory care, community and telehealth deserves evaluation," Porter said.

ResApp Health, which was founded in 2014, is now commercialising the technology under license from the University of Queensland. The ASX-listed company's share price has gone up and down over the last year, chiefly in response to accuracy results achieved in clinical trials.

In April, the company published results of a similar trial but with adults, achieving greater than 86 per cent accuracy for lower respiratory tract disease and pneumonia when compared with clinical diagnosis. The release of the results saw ResApp's share price rocket by 36 per cent.

Last month, ResApp announced it was working with UK-based medical device consultants and manufacturers to develop custom hardware and wearable devices capable of running its machine learning algorithms.

"The Android-based, ruggedised handheld device will be a low-cost option, complementary to off-the-shelf smartphones, for using ResApp’s mobile software diagnostic app in specific in-person clinical environments where certain hardware characteristics are desirable. The wearable monitor will provide ResApp with an easily worn, unobtrusive platform for monitoring patients with chronic disease 24/7 over extended periods," the company said.

Functional prototypes are expected within eight months.

Join the newsletter!


Sign up to gain exclusive access to email subscriptions, event invitations, competitions, giveaways, and much more.

Membership is free, and your security and privacy remain protected. View our privacy policy before signing up.

Error: Please check your email address.

Tags smartphonehealthuniversity of queenslandCurtin Universityappmachine learningNeural NetworksUQResApp

More about 24/7AustraliaCurtin UniversityThe University of QueenslandUniversity of Queensland

Show Comments