It’s not news. The use of artificial intelligence in the medical sector is not new. And yet, Google’s intention to use sound signals to predict the first signs of a disease is surprising. How? The internet giant has trained its AI model base with 300 million audio pieces which included coughing, mucus and difficulty breathing, to identify, for example, someone who is fighting tuberculosis.
And all this through bioacoustics, the combination of biology and acoustics, which helps us obtain information from the sounds produced by animals and humans. The AI model created by Google uses Sound signals to predict early signs of disease. The technology can travel on a smartphone and track high-risk populations in difficult geographies. Where expensive diagnostic hardware, such as X-ray machines or technical expertise, is not within reach, AI coupled with a mobile phone’s microphone could come to the rescue.
That AI system is already helping fight the world’s leading infectious disease killer. Nearly 4,500 people die every day and 30,000 fall ill from tuberculosis, according to the World Health Organization. It’s treatable, but Millions of people are not diagnosedIn India, the disease kills nearly a quarter of a million people a year, and early detection is key to stopping its spread.
Google trained its base AI model with 300 million audio clips from around the world, including coughs, sneezes, and breaths. Two-second audio clips were collected from publicly viewable, copyright-free content, such as YouTube videos and even coughing sounds recorded in a Zambian hospitalwhere patients came for tuberculosis testing. Body sounds are packed with information about our well-being and contain barely perceptible clues that can help detect, diagnose and manage health conditions.
The data that feeds Google’s HeAR (short for Health Acoustic Representations) AI model included 100 million cough sounds that now help detect tuberculosis. The smartphone-loaded AI tool can be easily taken to the most remote populations to detect the disease. The AI detects early signs based on subtle differences in coughing patterns, helping to triage patients and align them for further investigation and treatment. The goal is to enhance tools to help even people with limited training to detect respiratory diseases.
The tech giant’s Indian collaborator, Hyderabad-based Salcit Technologies, is an AI startup for respiratory healthcare. Salcit is using the AI model to improve the accuracy of TB diagnosis and lung health assessments by combining it with its own machine learning AI called Swaasa, which is the word for breath in the ancient Indian language of Sanskrit.
Swaasa technology is being used to screen patients even in remote areas and has been approved by India’s medical device regulator, a first for a software tool being deployed as a medical device. On its mobile app, Swaasa allows users to upload a 10-second cough sample and perform disease testing with 94% accuracy.
The sound of coughing is the equivalent of giving a blood sample, only this sample is processed in the cloud instead of in a laboratory. The screening test is sold for 200 rupees (just over 2 euros), in compared to a spirometry test which could cost almost 15 times more.
But there are challenges. It is necessary to ensure that Audio samples do not come with a lot of background noiseRural users, who are not familiar with technology, may not be able to record coughs on the app.
But this is not the only bioacoustics initiative, Google is investigating a model ultrasound-based screening for early detection of breast cancer and Montreal-based company Ubenwa has created a baseline model for infant crying and interprets the needs and health of babies.ands analyzing biomarkers in their crying sounds.