What sparked your interest in creating this device, especially for SSA patients?
So the reason why I started this is that, obviously, SSA is a genetic predisposition, which means it runs in the family. It actually does run in my family: my dad’s side, my grandmother, and two of my uncles suffer from ankylosing spondylitis, and I witnessed their suffering first hand because they are part of my family. I thought that even if this disease were detected earlier, even 6 months to 1 year earlier, their lives would have been so much easier, and they wouldn’t have to go through as much stress as they do now. So I thought, why don’t I work on this using the skills that I have acquired and my previous knowledge of robotics and biomedical engineering.
Did someone mentor you, or did you research it all alone and build it?
For the technical side, like the robotics side, I have a lot of prior knowledge and I could easily use that. However, on the medical side, I didn’t have as much knowledge about the disease, so I was mentored by Dr Amit Joshi, professor of orthopedics & Head of orthopedic department , at Balasaheb Hospital, Jogeshwari. He is an orthopedic surgeon, and with the medical side of things, I could easily consult him if I needed to verify my findings or just make sure I was doing things correctly.
Your innovation targets a significant gap in healthcare by providing an affordable alternative for diagnosing and monitoring this disease. So, can you walk us through the design process of it and how you ensured it meets the needs of patients as well as healthcare providers?
The way this initially started is that two years back, I had made a slightly different prototype, not dedicated to ankylosing spondyloarthritis, but more for diagnosing stress using how deeply a person is breathing. So that was a very similar concept where it used belts on the chest and the stomach to measure how many deep breaths and shallow breaths a patient took. From that, I got the idea. In my previous prototype, I used something called FSRs, which are force sensors, and I found that these are not very accurate. So I changed to something called IMUs, which are inertial measurement units. This form of electronics is extremely accurate in terms of distance; it provides an accuracy of even less than 0.1 mm. So it’s extremely precise in that sense. In the initial stages, I didn’t have much of the beautification or any of that; it was just raw electronics. I was working with that and testing. Slowly, for example, earlier the sensors were all put on a belt, but then I made them free sensors, not on a belt because the belt was restricting movements. After that, I realised that this is not really a one-size-fits-all type of prototype, so I added bungee cables that stretch according to whatever shape and size the patient is. I also added casings for all the sensors to ensure safety and hygiene.
Tell us how the device works
So basically, as I already mentioned, we have five IMUs. All of these five IMUs send data to a microcontroller. The microcontroller I have used is an ESP8266. All of this data is stored here, and it all runs on battery, so it’s completely wireless. The ESP8266 has Wi-Fi capability, so all that data is then sent to a computer. While this happens on my computer, it is possible with any computer. It sends the data to the computer, and then the computer performs the rest of the processing. There, it does a bunch of mathematics, like integration, quaternion rotation, and all of these things. After that, we use an AI model because, in each second, we have 70 data points-that’s the speed of the Wi-Fi sending. Each data point has around 15 different data points, and if you consider a five-minute test, that becomes such a large amount of data that if you give it to a doctor, it’s nearly impossible for them to read all of that and make sense of it. That’s why I’ve used an AI model, a machine learning model, and I’ve chosen to use an LSTM model, which stands for long short-term memory. This model can pick up on really complex patterns in both long-term and short-term breathing patterns so that it can accurately differentiate between a healthy person and a diseased patient.
What other unique features differentiate it from traditional or preexisting treatments? Are there any more features that set this device apart from existing testing methods?
The current testing method, the primary one used, is the HLA-B27 blood test. This is problematic for several reasons. Firstly, it has a low accuracy of around only 50 per cent according to several studies. It is also extremely expensive for the general population in India, costing 4000 rupees, and not everyone in India can afford that. Another issue is that this test can only be done in very specific labs. Only metropolitan cities like Mumbai, Delhi, and other big cities can perform this test, whereas in more rural setups, it’s not possible. In rural areas, they measure breathing using a measuring tape around the chest, but that’s extremely inaccurate because it’s not a 100 per cent guaranteed way of diagnosing something. My prototype solves many of these problems because it’s cost-effective; there’s no onetime use. Once you have it, you can test as many patients as you like, and the cost is only 2000 to 3000 rupees. It can be used on as many patients as you want, completely free of cost
What challenges do you anticipate in terms of adoption, training, and scalability within government hospitals?
I feel that in training, there won’t be much of a problem because there isn’t much technical knowledge required to operate this device. You don’t need a doctor or a nurse; any healthcare professional without that kind of expertise can easily operate it because I’ve made it very easy to use. In that sense, it’s easily operable. One issue that might arise is that while a lot of patients are very accustomed to blood tests, they may find this device strange. They’re not used to it, and that might be an issue during scalability.
How do you plan to address that challenge?
To address that, it’s important to educate the patient about what this is doing, what kind of test this is, and obviously that this is non-invasive. So that’s another great advantage.
Are there any ongoing developments, collaborations, or potential partnerships that you are exploring for further enhancement of this device in the market?
I’ve applied for a utility patent on this, which will allow me to scale it much more. I’m also looking to improve the AI model to make it more accurate and generate better reports.
How are you using AI in this?
As I mentioned, the amount of data is so large that it’s really hard for doctors to perform a diagnosis. That’s why we use an AI model that processes all that data and gives an accurate output. The AI model not only removes the burden from the doctor to require that technical knowledge, but it’s also a very straightforward process. An AI model is obviously more accurate than a human in terms of performing a diagnosis.
However, sometimes we see that AI also makes mistakes in its calculations. How will you ensure that patients receive accurate information?
Right. This has been tested. I’ve tested this in public hospitals, and I’ve done my own testing on many different patients. From all of the testing, the accuracy is around 95 per cent. Obviously, you can’t guarantee that accuracy will be 100 per cent. But compared to the 50 to 55 per cent accuracy provided by the traditional test, this is a much bigger improvement. This is more time-saving and does not require much effort to get the data.
How does this cost compare to similar devices or diagnostic tools currently available in the market? How much does this vary? Right now, what is the cost in the market?
In terms of seronegative spondyloarthropathies, there is no other such device that performs this kind of diagnosis. HLA-B27 is the most widely used. There is no kind of electronic device or any kind of device that uses sensors to create a diagnosis.
neha.aathavale75@gmail.com