Salcit Technologies, an Indian respiratory healthcare startup, is pioneering the use of Google’s Health Acoustic Representations (HeAR) bioacoustic foundation model to revolutionise tuberculosis (TB) detection in India.
The startup aims to enhance early TB detection through the analysis of cough sounds, leveraging the HeAR model’s vast training on approximately 300 million audio samples, including 100 million cough sounds.
AI TB Screening
Salcit Technologies plans to integrate the HeAR model with its existing product, Swaasa, which has a track record of employing machine learning for early disease detection. Swaasa has been instrumental in offering location-independent, equipment-free respiratory health assessments, bridging significant gaps in healthcare delivery across India.
By harnessing HeAR’s capabilities, Salcit intends to make TB screening more widespread and accessible, especially in areas with limited access to advanced medical resources.
HeAR, which was publicly introduced in March 2024, is designed to assist researchers in building models capable of listening to human sounds and identifying early signs of disease.
Sujay Kakarmath, a product manager at Google Research, emphasised the model’s accessibility, stating, “Compared to blood tests and imaging, sound is by far the most accessible piece of information that we can get about a person. HeAR can pick up chest x-ray findings, tuberculosis and even detect COVID from cough sounds.”
HeAR’s Potential in Global Health
Shravya Shetty, Director and Engineering Lead at Google Research, highlighted HeAR’s ability to discern patterns within health-related sounds, making it a powerful tool for medical audio analysis. Shetty noted, “We found that, on average, HeAR ranks higher than other models on a wide range of tasks and for generalising across microphones, demonstrating its superior ability to capture meaningful patterns in health-related acoustic data.”
“In places where access to Advanced Medical resources is scarce, we can imagine a future where a healthcare professional with a machine learning model and a phone can collect a sample of your sound and inform clinical care,” said Kakarmath.
The Stop TB Partnership, a United Nations-hosted organisation committed to ending TB by 2030, also supports this innovative approach.
Google’s invitation for researchers to explore HeAR’s potential through its API signifies a major advancement in acoustic health research. Salcit Technologies’ application of HeAR in TB detection underscores the model’s promise in addressing global health challenges, particularly in resource-limited settings. Through continued research and development, both Google and Salcit aim to improve health outcomes for communities worldwide.
