Ilman Shazhaev, is the Co-Founder & CEO of Acoustery, a health-tech company that develops AI technology for the early recognition of respiratory diseases.
What initially attracted you to computer science and engineering?
The amount of data available today is more extensive than ever, and AI technology — which is very data-dependent — has made tremendous progress in the past few years. This is why doing research in this field is so exciting.
Right now, I am focused on Big Data projects. During COVID-19, I co-founded Acoustery: a fully automated AI-powered solution for monitoring one’s health based on the analysis of their voice, cough, and breath.
The next step was to combine health research and gaming. Why? The amount of data this industry generates is unique; what's more, gamers are early adopters ready to share their data and contribute to scientific progress. At the same time, the number of ongoing clinical trials is low, the progress is slow, and the gaming sector allows for much more dynamic data processing.
Could you elaborate on the genesis story behind Acoustery?
As I mentioned before, Acoustery was started during the pandemic. Even though business opportunities in 2020 were relatively limited, I was staying in Dubai, one of the few locations where a project could operate without super strict limitations.
My co-founder Dr.Dmitry Mikhaylov, a professor at the National University of Singapore, and I started on a new challenge: early-stage detection of COVID-19. At the time, UAE was massively exploring early diagnosis technologies and largely supported AI projects.
Thanks to this, we got access to one of the best testing facilities in the UAE: Sheikh Zayed military hospital, where we had data from hundreds of COVID-19 patients to train our AI engine on.
At the next stage, tests showed our technology was very accurate and had great potential. Researchers published their results in the top rate journals in Japan and USA, and our testing method was used in several Asian countries during pandemics as an emergency tool.
When COVID-19 was over, we focused on detecting asthma using the same approach. Sharjah University, which is currently leading in UAE's research, ground-approved these tests.
For COVID-19 how accurate is this system compared to PCR, LFT, and antibody tests?
The positive predictive value of Acoustery in the context of community-wide screening for COVID-19 is relatively high (81%) compared to Xpert MTB/RIF, a new test that is revolutionizing tuberculosis detection and control by contributing to the rapid diagnosis of the disease (61%) and PCR throat swabs (71%).
Our findings have shown that the software developed by Acoustery can be used as a primary non-laboratory screening tool to detect cases of COVID-19 and route patients to laboratories for PCR testing.
Could you tell us more about the machine learning used to train the AI?
We assumed that to get an accurate detection rate of COVID-19, we could train convolutional and recurrent networks to diagnose the disease by analyzing the spectrograms of cough and breath of patients. A spectrogram is a visual way of representing the signal strength at various frequencies. A number of medical studies showed significant differences between the cough of patients who had COVID and those who did not, so we trained our AI engine to recognize such differences.
Acoustery’s developments can be used to diagnose Alzheimer's, which is commonly perceived as a neurological disorder. How exactly does it work?
Our study explores how speech measures may be linked to language profiles in participants with Alzheimer's disease (AD) and how these profiles could distinguish AD from changes associated with normal aging. To achieve this, our AI analyzes simple sentences pronounced by older adults with and without AD, from the percentage and number of voice breaks to shimmer (amplitude perturbation quotient) and noise-to-harmonics ratio. The accuracy of this analysis reaches 90%.
Later on, we used the same approach in Farcana Labs – a venture focused on collecting Big Data generated by gamers to research disease progression, especially with mental disorders.
What other diseases can be diagnosed using this method?
Asthma is our key priority now. Tuberculosis is another focus, as well as chronic obstructive pulmonary disease (COPD), pulmonary fibrosis, pneumonia, and lung cancer.
How large are the training data sets for these use cases?
We have thousands of cough recordings in our database collected during the last four years.
What is your vision for the future of medical diagnosis across the board?
The data collected by personal devices will play a significant role in diagnosing diseases at an early stage and preventing pandemics. Even our mobile phones have multiple sensors: a microphone is just one of those. Accelerometers that can analyze motor skills and detect numerous diseases are another.
Even though these technologies shouldn’t be the only source for diagnosing, they can significantly help predict and prevent the spread of highly infectious respiratory diseases — and, consequently, new pandemics. Acoustery can also be used in developing countries where access to PCR testing is limited.
You seem to have multiple projects on the go; what are some other exciting use cases that you see for AI?
The AI space is unique. As AI researchers, we focus on niches that generate big data, which is necessary for any AI research. We need a lot of patients to compile quality datasets, so we have a few pieces of research going in parallel and are exploring several business verticals.
We see gaming as an area where a massive amount of data is generated. Today, people play a lot of video games, which is a valuable source of data for health research. Collecting data from personal devices and wearables is another vector with significant potential.
All in all, it’s exciting to be exploring this technology now, and I believe it has much more potential still to be harnessed across other sectors.
Thank you for the great interview, readers who wish to learn more should visit Acoustery.