Google's AI model to help researchers identify TB, other diseases from the sound of cough

Google introduced HeAR bioacoustic foundational AI model in March. It helps researchers build models that can listen to various human sounds and flag early signs of diseases.

Our bodies produce various sounds from coughs to breathing. These sounds carry subtle clues that can serve as valuable health indicators for detecting diseases such as tuberculosis.

Google now wants to harness the power of artificial intelligence (AI) to help researchers extract health insights from these sounds, aiding in the early detection of such diseases.

On August 20, Google announced a collaboration with Salcit Technologies, an India-based respiratory healthcare firm, which has come up with Swaasa that uses AI to analyse cough sounds and assess lung health.

Cough alert

As part of this collaboration, Swaasa will utilise HeAR (Health Acoustic Representations), a bioacoustic foundational AI model released by Google in March, to help improve the accuracy and effectiveness of its tuberculosis AI models.

Shravya Shetty, director and engineering lead, Google Health, said HeAR has been trained on 300 million pieces of audio data curated from a diverse and de-identified dataset. The cough model, in particular, was trained on roughly 100 million cough sounds.

"HeAR AI model is able to detect subtle differences in cough patterns, which can help with diagnostic accuracy and speed for detecting tuberculosis. By listening to patients' coughs, it could potentially help and triage patients much earlier in their healthcare journey. This means we can identify those at higher risk and determine who actually needs follow-up testing to confirm whether it is actually tuberculosis or not," Shetty said in a select media roundtable.

Shetty said HeAR’s goal is to power various screening tools that can help physicians and community health workers screen for respiratory diseases on the fly, by simply using the microphones on their smartphones.

HeAR is not built just for a single disease and is intended as a general acoustic foundation model for health. It is designed to help researchers build models that can listen to various sounds and flag early signs of diseases.

"(HeAR) can be fine-tuned for various health related sounds and use cases. Hence, it can be a powerful foundation for medical and audio analysis in the future" she said.

Researchers can request access to the HeAR API that will be made available via a Google Cloud API.

"We hope to advance the development of future diagnostic tools and monitoring solutions in tuberculosis, chest, lung and other disease areas, and help improve health outcomes for communities around the globe through our research" Shetty said.

In March, Google announced a partnership with Apollo Radiology International (ARI) for early disease detection in India.

Apollo Radiology International is deploying Google's AI models to improve health outcomes for people with tuberculosis, lung cancer and breast cancer.

Over the next 10 years, ARI will use these models to provide three million free AI-powered screenings for the three diseases, the company had said.

Detecting breast cancer via ultrasound

Google is also partnering with Taiwan's Chang Gung Memorial Hospital (CGMH) to develop AI models that can detect early signs for breast cancer through ultrasound images.

"Traditional mammography-based screening, while effective, faces cost and accessibility barriers in low-resource environments. Ultrasound is more accessible with portable, affordable ultrasound devices readily available," said Preeti Singh, Lead Technical Program Manager, Google Health.

"It is a potentially effective primary screening tool, especially for younger women and those with dense breast tissues, which are more prevalent among Asian demographics," Singh said.

Google plans to start with Taiwan and then eventually expand to rest of the world, she said.

The tech giant is also collaborating with organisations such as hearing solutions provider Cochlear, and Australia government's National Acoustics Laboratory for an initiative called Australian Future Hearing Initiative, which aims to make hearing devices more personalised and more effective in noisy environments.

This project will explore new applications of AI to better identify, categorise and segregate sound sources, thereby offering a more natural and comfortable listening experience to people.