Researchers at Monash University have developed software that, used in conjunction with a digital stethoscope, improves screening and monitoring capability and more accurate diagnosis of respiratory issues in vulnerable newborn babies. Their findings were published last week in IEEE Access.
The software removed surrounding noise from chest recordings. Such noise may come from the external environment, internal body sounds or the device itself, and can affect the quality of chest sound obtained with stethoscopes. Low-quality chest sound can make monitoring and diagnosis challenging, or lead to misdiagnosis.
“Respiratory issues are common in preterm babies,” says Dr Faezah Marzbanrad from the Monash University Department of Electrical and Computer Systems Engineering.
“The software we’ve created removes all of the surrounding noise from chest recordings so the heart and lung sounds are separated and very clean. This enables doctors and nurses to listen to them very clearly without interference and better diagnose any potential issues.”
The team collected 207 chest sounds from 119 preterm babies, each 10 seconds long. They used a deep learning model called YAMNet, pre-trained on sound classification to automatically detect heart and respiratory rate.
They fine-tuned YAMNet on the 207 chest sounds and found that the model could predict heart and respiratory rates with about 57% and 51% accuracy. They also found that increasing sound quality reduces vital sign error, prompting the development of the new software that improves chest sound quality.
“Chest sounds in newborn babies are very difficult to assess and interpret, especially in preterm or sick babies,” says Associate Professor Atul Malhotra, Senior Neonatologist and Head of Early Neurodevelopment Clinic at Monash Children’s Hospital.
He says small chest size, fast breathing and heart rate, and additional noise from neonatal intensive care unit (NICU) equipment can affect chest sound quality. “We rely a lot on chest X-rays and invasive blood gas monitoring to indicate and monitor cardio-respiratory illness in babies,” he adds. “This software gives us a much better resolution to interpret, assess and monitor newborn’s illness.”
The neonatal period is the most vulnerable time for a baby, with 1.7% of live births resulting in deaths. Stethoscope-recorded chest sounds contain crucial cardiac and respiratory information that helps clinicians timely assess for signs of severe health risks.
Marzbanrad says the software is easy to use for hospital staff and parents and would be precious in rural and remote regions and low- and middle-income countries where health resources may be limited. A baby’s chest sound can be recorded and sent to a specialised doctor for real-time analysis.
“It’s not always practical to get to a doctor, and on many occasions, breathing problems happen overnight when you can’t get to a doctor,” she says. “This ensures that you can record the sound in real-time, and it’s something useful for the doctor to assess.”
The team will trial the software in conjunction with new digital stethoscope hardware at the Monash Children’s Hospital and expect it to be available worldwide in the following months.