Though breast cancer treatment can be highly effective, women across the globe face drastically different outcomes depending on where they live.
According to research compiled by the World Health Organization, survival for at least five years after diagnosis ranges from more than 90% in high-income countries to only 66% in India and 40% in South Africa.
Geetha Manjunath, founder and CEO of Bengaluru, India-based Niramai Health Analytix, set out to improve access to screening when a close family member died of breast cancer in her early 40s not long after receiving a diagnosis. The company recently participated in the M2D2 Impact accelerator at the University of Massachusetts Lowell and received FDA 510(k) clearance earlier this year.
Manjunath sat down with MobiHealthNews to discuss how Niramai’s artificial intelligence-enabled screening system works, the importance of explainability when using AI in healthcare and what’s next for the company.
MobiHealthNews: Can you tell me a little bit about how the Thermalytix system works for breast cancer screening?
Geetha Manjunath: I’ll set a little bit of context. If you look at the mortality rates across different countries, there is a huge variation in the number of people who survive breast cancer. In order to stop these deaths, we need regular screening, but that is not feasible today. One, because of the economic constraints. Such a huge initiative is usually restricted to women around 45 years and older, because there is a relationship with age. Also, mammography, which is the standard for breast cancer detection, does not work as well on younger women below 45 years old, because they have what is called dense breasts. In fact, in almost 50% of the ladies above 40 there is a density issue again.
In countries like India, China, the Philippines, the affordability of the machine itself is a big issue for the government as well as small diagnostic centers or private hospitals. So with all this, what Niramai has developed is an affordable, accessible method of detecting breast cancer in women of all age groups and all breast densities. In addition, the machine is actually very portable. You can do the test in the hospital. You can also take it out to do the test in remote areas, rural villages as well as corporate offices. We also have a home screening for breast cancer screening.
The lady enters a small room, like a small booth. She goes in, she closes the door and then she removes her clothes in front of this device. Nobody is inside, it’s like a changing room. Nobody sees her or touches her during the test, which is unlike the experience of doing a mammogram, for example.
It uses an imaging technique called thermal imaging, which can be controversial. Traditionally, thermal imaging has been used for abnormality detection. However, it has never been accurate enough to be used or recommended in hospitals, because we are measuring, let’s say, 400,000 temperature points per person. It’s very hard for the human eye to differentiate between different shades of yellow, different shades of oranges, and so on.
We have developed our artificial intelligence-enabled smart software, which analyzes this temperature distribution on the chest area, and converts that into a cancer report. That is completely done automatically with scoring indicating the level of abnormality. That is our main value proposition, AI algorithms to convert temperature distribution into a cancer report.
MHN: So the cancer report is not saying, you 100% have breast cancer. Is the idea that it highlights potential concerns and you get further tests?
Manjunath: Absolutely. It’s a screening test, which means that out of 100 women screened, we identify those nine or 10 women who need to go for a follow-up diagnostic workup – maybe another mammogram, or 3D mammogram, or more sophisticated breast MRI, or a breast ultrasound.
MHN: AI is becoming a lot more prevalent in healthcare, especially for imaging. How do you balance concerns about introducing bias or not understanding how the AI is making its recommendations?
Manjunath: AI is a machine, and a machine behaves the way you train it. So the training phase is very, very important. What kind of samples you use for training, making sure that the training set is addressing multiple abnormal aspects. For example, in breast cancer, we looked at pregnant women, we looked at people who are menstruating, we looked at people who had fibroadenomas. All of the different categories and subcategories of potential abnormalities have to be included. You definitely need to work with a medical expert to actually ensure that your training is unbiased. It’s really multidisciplinary, because the domain experts and the technology experts have to come together.
And the explainability part is also hugely important. So for example, initially, we just said it would look at a patient and say, cancer or no cancer. But the doctor said, “What do I do with this? I can’t take any action with this. You just say cancer, but which breast and what happened?” So we now have a three page PDF report that is automatically generated, which gives scores for the left breast and the right breast. We do markings on the breast automatically, saying this is where you want to check again.
MHN: You recently received FDA 510(k) clearance here in the U.S. What are the next steps for the company?
Manjunath: We recently received the U.S. FDA clearance, we’re just finishing device registration, though we launched in a beta mode last month. So I’m already looking for partners. To start with, we will be working with thermographers, people who are already using thermal imaging. Our current clearance from FDA is to use this as an adjunct to mammogram, so we would love to work with these imaging centers to provide this facility as well.
In parallel, we are working on the next device, which is a little more sophisticated than our current device, for clearance by the FDA. We need a multisite clinical study in the U.S., so we have identified hospitals in New Jersey and Arizona, and probably Florida as well.
Meanwhile, we have received a huge response from low and middle income countries because of the affordability and accessibility part of it. So, in countries like the Philippines, the UAE, India, Indonesia, we are working with distributors in the local domestic market to take the solution to the developing world. And also we are cleared for use in Europe.
So I’m very excited. I tried to solve a very, very local problem of trying to get Indian women detected with cancer. We’ve now screened 60,000 women in India alone, which is a considerable number, given it’s a new medical device. We have already launched in Kenya. So, I’m very excited to have an opportunity to make a difference in the lives of women, hopefully, around the world.