Daybreak - Is AI in cancer care just hype or the real deal?

Episode Date: June 18, 2024

AI algorithms for cancer screening are being developed around the world. Most medical professionals will agree that there is tremendous potential here. If developed properly, AI can potentially detect various cancers at very early stages – which would make it easier to treat cancer and possibly even increase chances of survival. But all of that is great in theory. In reality, the general consensus amongst the medical community is that AI-led cancer screening just isn’t there yet. When it comes to screening, accuracy is everything. And There’s a long way for this technology to go before it is able to detect cases of cancer with close to perfect accuracy. 

Hi, this is Rohan Dharma Kumar. If you've heard any of the Ken's podcasts, you've probably heard me, my interruptions, my analogies, and my contrarian takes on most topics. And you might rightly be wondering why am I interrupting this episode too? It's for a special announcement. For the last few months, I and Sita Raman Ganesh, my colleague and the Ken's deputy editor, have been working on an ambitious new podcast. It's called Intermission.

We want to tell the secret sauce stories of India's greatest companies. Stories of how they were born, how they fought to survive, how they build their organizations and culture, how they manage to innovate and thrive over decades, and most importantly, how they're poised today. To do that, Sita and I have been reading books, poring over reports, going through financial statements, digging up archives, and talking to dozens of people. And if that wasn't enough, we also decided to throw in video into the mix. Yes, you heard that right. Intermission has also had to find its footing in the world of multi-camera shoots in professional studios, laborious editing, and extensive post-production.

Sita and I are still reeling from the intensity of our first studio recording. Intermission launches on March 23rd. To get an alert as soon as we release our first episode, please follow intermission on Spotify and Apple Podcasts or subscribe to the Ken's YouTube channel. You can find all of the links at the ken.com slash I am. With that, back to your episode. A couple months ago, the country's biggest cancer hospital,

the Tata Memorial Hospital in Mumbai, did something pretty incredible. It set up something called a bioimaging bank. What they did was they collected data from 60,000 cancer patients from across the country. So this was literally a bank of information. Information about everything to do with specific cases of cancer. So radiology and pathology images, specifics of their different treatment plans, outcome data, the works. And here's where it gets interesting. The reason this bank of data exists is so it can train and test artificial

intelligence algorithms. It is training these algorithms to be able to do things like screen for cancer and also predict how a patient may respond to different therapies. So this is some cutting-edge groundbreaking stuff. It also isn't the first time AI algorithms have been developed or used by medical professionals to detect cancer cases. In fact, AI algorithms like these are being developed around the world. Most medical professionals will agree that.

there's a tremendous amount of potential here. If developed properly, AI can potentially detect various cancers at really early stages, which in turn would make it easier to treat cancer and possibly even increase chances of survival. But all of that is great in theory. In reality, the general consensus amongst the medical community is that AI-led cancer screening just isn't there yet. When it comes to screening, accuracy is everything. and there's a long way to go for this technology before it's able to detect cases of cancer with close to perfect accuracy.

So take for instance Bangal-based deep tech startup, Nira Mai Health Analytics. It came up with a thermal imaging-based AI tool for breast cancer screening. And its accuracy at the moment is about 80 to 85%. Now, this may seem okay at first,

but to be clinically relevant and algorithm has to be 90% to 95% accurate. So a lot of medical professionals, particularly in hospital settings, think of it as a maturing technology, which is why they end up being hesitant to introduce this sort of thing in their hospitals. But having said that, in the last few years, pretty much every major hospital you may have heard of has jumped on the artificial intelligence bandwagon.

You may notice that very often many of these hospitals announce that they are integrating AI, but are pretty vague about how exactly they're doing that. So in this episode, we delve into whether AI in cancer is hype or the real deal. Welcome to Daybreak, a business podcast from the Ken. I'm your host, Rahil Philippos, and I'll be joining Snigda every week to bring you one business story that is worth understanding and worth your time. Today is Tuesday, the 18th of June. Since 2020, a Mumbai-based startup called Cure AI has been using an AI algorithm to carry out tuberculosis screening. And so far, they've been able to screen well over 1,000 people at about 139 health facilities across the country. Now the algorithm works is it's able to scan chest X-stays to tell you if someone has TB or not.

And in a lot of remote parts of the country, where radiologists aren't available, this has been a real game changer. So after it figured out how to do TB screening, Cure AI decided to develop an algorithm for lung cancer screening. The great thing here is that it could use the same chest x-rays it was using for screening TB to develop the algorithm. So this algorithm was literally built on this huge repository of x-rays it had collected over the years. My colleague Seema Singh, the Ken's founder editor,

looked at another Chennai-based diagnostics chain called Arti Scansans. They are in the process of carrying out a study using curei's cancer algorithm. It basically ran CT scans of about 45,000 people who came for routine checkups through the algorithm. It detected lung nodules, which are tiny spots on the lung, in about 1,300 of those scans. And 87 of these turned out to be cancer cases. Now, the reason this is so important is because most people who are diagnosed with lung cancer, end up succumbing to the disease.

This usually happens because of a late diagnosis, usually in the advanced stages of the disease. Now, with AI algorithms like cures, cases of lung cancer can be detected early. From the Arty Labs example, it's apparent that lung cancer screenings should ideally take place alongside your standard TB scan. Except that isn't generally what happens here in India.

This largely has to do with cost. screening for cancer is not a high priority of the government because if you're found to have lung cancer, the government won't be able to foot the bill for your treatment. And this doesn't just apply to lung cancer screening. It applies to all cancer screening algorithms. Basically, even if technologies for cancer screening in India are getting there, the business models aren't.

Because the question is, who pays for it? And that's not it. In the Rati scans example we just discussed, the general population was scanned, parts of rural India or other parts of the country where healthcare systems aren't in the best shape. But in hospitals, the algorithm for screening isn't the same. That's because the goal is different. With general population screening, algorithms are designed not to leave out any cancer undetected.

The unfortunate byproduct of that is there are more false positives. While in a hospital setting, the aim is to identify more symptomatic patients and spared the healthy ones from follow-up tests. Basically, the hospitals demand more accuracy. And according to a lot of them, just AI screening doesn't offer that level of accuracy. So a lot of hospitals are seeing it more as a secondary tool. Take for instance the case of breast cancer. Some hospitals are making use of NIRAMIA analytics thermal imaging-based AI tool that we discussed a little while earlier.

But they're making use of it along with the three other wasted issues. detect breast cancer, hand examination, mammography and ultrasound, which until now was the industry standard. But the other issue with AI algorithms like these are biases. More on that in the next segment. Most people believe that if you're smart, work hard and meet your goals, a promotion is guaranteed. But the truth is, a lot of talented people fail to get ahead while seemingly ordinary peers blow right past them. So how do organising? decide who gets promoted over whom. If it's not entirely based on performance,

does it mean that you have to suck up to your higher-ups? Kind of play office politics? Be everything everywhere all at once? These were the questions I was exploring and assumptions I was challenging in the latest episode of the first two years. An early careers podcast from the Ken. If you're starting out,

and you're probably in the 18 to 25 age group, this episode is a great place to start. I am Aksha Chandrashakran, the host. You can't have favourites as a podcast host, but this one about how to make a case for your promotion is definitely my favourite. Click the link in the show notes to listen to the episode or just look up the first two years

on Spotify or Apple Podcasts or wherever you get your podcast. Thank you. Now back to Rahal. Okay, here's a scary statistic. One in 10 Indians will be diagnosed with cancer in their lifetime. But each cancer is a distinct entity entirely, which is why the medical world is constantly trying to find the ideal treatment plan for each patient, using drugs that will kill cancer without wrecking the patient's body.

Developing an AI algorithm to screen cancer is even more complicated. Like we discussed earlier, many of these algorithms are trained on images like x-rays, PET scans or MRIs. Now, all imaging in cancer is largely done on imported machines. And what happens is that an alga trained primarily on one type of machine, like a Phillips one for instance, will behave slightly differently when it's tested on, say, a GE or a general electric machine. Most of these machines are imported.

But like Swapnil Rane, associate professor at Tata Memorial Center pointed out, AI is not a machine that one can simply import. And what happens more often than not is that companies from the West have started selling their machines with some form of integrated AI. The catch is that their devices are primarily trained on Western data sets, meaning imaging, outcome data, specific treatment plans for thousands of cancer patients from Western countries. But this doesn't necessarily work in the Indian context. In fact, doctors at Tata Memorial Center have tested.

tested this out extensively. They tested both an AI algorithm trained on Western data sets and one trained on Indian data sets and found huge variations. The model trained on Western data sets performed with 85% accuracy, but its accuracy when trained on the Indian data set drop to 50%. That's basically as good as tossing a coin. Also, like I mentioned earlier, to be clinically relevant and algorithm has to be 90 to 95% accurate. What this proves is that there are huge variations across geographies. So for an AI algorithm to be accurate in India, it has to be trained on Indian data sets. Remember the bioimaging bank I told you about at the beginning of this episode?

Rane is developing it along with his team at the Tata Memorial Center for exactly that reason. Once it goes live, it'll be an invaluable resource for anyone trying to develop AI algorithms for cancer screening. But again, that brings us. me back to the point I made earlier on in this episode. All of this is great in theory. But all these algorithms, including the one that Rani is developing, aren't where they need to be yet. Private hospitals don't have enough incentive to adopt them because accuracy rates aren't what they want them to be, and public hospitals don't have enough resources to add data storage overheads. But like Rane told my colleague Seema, there is still hope.

AI can bridge the gap in Indian health systems. It just needs to check all biases at the door. Daybreak is produced from the newsroom of the Ken India's first subscriber-focused business news platform. What you're listening to is just a small sample of our subscriber-only offerings. A full subscription unlocks daily long-form feature stories, newsletters and podcast extras. Head to the Ken.com and click on the red subscribe button on the top of the website.

Today's episode was hosted by Rahil Filippos, produced by me Snigda Sharma, and edited by Rajiv Sien.