Salcit and Google partner to enhance AI-based TB detection

This is the second installment of a two-part series based on an exclusive interview with Venkat Yechuri, CEO of Salcit Technologies. In part one, Yechuri detailed how his company is transforming TB screening in underserved communities through AI-powered cough analysis, making diagnosis possible with just a smartphone. He also shared his vision for expanding the technology’s capabilities to other respiratory conditions and remote patient monitoring. Here, Yechuri discusses how partnerships with major tech companies and leading healthcare institutions are helping to enhance and scale the technology globally.

Why did Salcit decide to partner with Google and incorporate its Health Acoustic Representations (HeAR) technology into Swaasa?

Yechuri: Salcit is always looking to collaborate with the best partners around the world, and though Google has no interest in manufacturing a medical device, they’re interested in the work we are doing at a higher level, and its applications.

HeAR is a bioacoustic foundation model trained on audio data. To develop it, Google took the “Deep Learning” approach, which essentially requires that you get a lot of data, hundreds of millions of data sets, and then you kind of let loose your algorithms, and let them come up with some patterns.

Google trained their algorithms on every cough sound that was ever recorded on YouTube, hundreds of millions of them. And then they ran one or two smaller trials to see if there was validity to what they were doing.

Salcit took a different approach when we developed Swaasa. We have a carefully curated set of information, a proprietary database, which includes physician annotation. Our data is high quality, specialized, and we know exactly what we’re looking at with it, but it only numbers in the tens of thousands, not the hundreds of millions. So we asked the question—is there a way to marry Google’s HeAR model with our dataset and come up with even better patient outcomes?

For example, when Swaasa hears a sound, it checks to see if it’s a cough, and if the cough is “real” or not. We have created this checking mechanism through our own research, but Google has done the same thing with HeAR, because they had to look through every sound in their dataset and label them as either “a cough” or “not a cough.” Going further, we tested the algorithms powering Swaasa on a variety of smartphones, ranging from a $30 model all the way up to an iPhone, and we know how Swaasa performs on these devices—we’ve fine-tuned its performance on each one. 

For us, working with Google makes sense because when Swaasa is used in the field, it will inevitably encounter a lot of sounds and situations our own algorithms have not seen in our dataset. By leveraging the HeAR model, Swaasa now can perform better in the field.

Do you currently have any other partnerships in addition to the one with Google?

Yechuri: After I joined Salcit, we took the company global, not just in terms of market, but in terms of our partners, too. For example, we have an ongoing collaboration with the University of California, San Francisco that deploys Swaasa in five countries for TB diagnosis, and we are also working with the University of California, Irvine.

Our mission is, broadly, about improving accessibility and affordability in healthcare, and there are accessibility and affordability problems all over the world, even in the United States and health systems like UC Irvine’s. These problems simply manifest differently than they do in India.

UC Irvine has a main hospital center and a number of satellite locations, and they do employ spirometry to check in with patients on their respiratory function at these satellite locations. However, these locations are not properly staffed, meaning accessibility is still an issue, though not in the same way as many rural communities—it’s not because UC Irving can’t afford the testing equipment needed, but rather because it can’t provide the trained staff needed to run the equipment. 

How do you evaluate potential new partners for Salcit?

Yechuri: We’ve had potential investors propose to us working to help with the issues of childhood asthma and other respiratory conditions. These are big problems with no easy solutions. Childhood pneumonia is the leading cause of death worldwide for kids ages zero to five. It’s very treatable, but only if you catch it early, which can be difficult when you circle back to the same problems of needing the proper equipment and the expertise to run it.

When asked to work on childhood pneumonia, we had to think it through carefully. It’s a different disease than TB, and the pediatric population is very different to the adult population. Even working on pediatric TB is very different.

We were also contacted recently by a different group that asked us if we were interested in taking the diagnosis of other biomarkers from concept to product as we’ve done with cough. My first question was, “Can you guys support us in taking Swaasa to a hundred countries around the world?” Ultimately, that aspect is really important to us, knowing we can scale our product globally. 

This interview has been edited for brevity. Read part one here.