Early Detection To Faster Treatment & Care, How AI Is Helping Us Fight Against Deadly Diseases

Imagine these two objects -- a smartphone and an MRI scan bed. Think about all the technological advancements smartphones have witnessed over the last decade. Contrast it with the image of an MRI scan bed. If you're thinking the latter hasn't changed since forever, you are more or less right.

Just recently, I remember waiting for some 90 minutes while my mother was getting an MRI scan of her knee. Add all the follow-up time on interpreting the scanned images and face-to-face with her consulting doctor, it was about the runtime of an average Bollywood flick complete with the song-and-dance routine.

Reuters

ALSO READ: I Saw How AI Is Changing Our World, And Here's How It Influences What Products You See Online

Why can't MRI or CT scans be speedier? If there's just one government doctor available for 11,082 people in India, ten times worse than the recommended ratio of 1:1000 as per the National Health Profile (2018), shouldn't there be faster machines that help doctors treat more patients in lesser time? Fight diseases better and save lives?

This problem of excruciatingly slow care in the medical industry has been there for a while, and Dr Shreyans Vasanawala, Professor of Radiology (Pediatric) at Stanford University is well aware of it. But he also strongly feels that the status quo won't be there for long, and a major change is coming in how tech enables healthcare. Positive change that will actually be felt by both patients and doctors, and everyone involved in receiving or administering medical care.

Improving MRI scans through machine learning

Over a telephone call, Dr Vasanawala explained how machine learning implementation at Stanford's Children's Hospital is helping enhance the quality of MRI scans while also cutting short the entire scanning process.

Stanford // B is more clearer than A

Training image recognition algorithms on years of medical data -- on how healthy internal organs differ from unhealthy ones -- has led to highly effective machine learning models with a very high rate of positive disease detection. Not only has the quality of image outputs also gone up, Dr Vasanawala explains, but speedier diagnosis has led to dramatically reduced time spent by patients.

Not only this, but machine learning models are now being implemented right at the scanning stage when a patient lies down for an MRI, according to Mark Burby from Intel's Health & Life Sciences division for Asia Pacific & Japan.

Building more intelligent machines

Mark told us how companies like General Electric -- which manufactures CT scanners -- are using Intel¡¯s AI features to not only allow scanning device from faster image processing but also determining unhealthy traits, thereby freeing up a specialist doctor's or radiologists time from doing non-specialized work. It can process a whopping 600 images per second and accurately identify anomalies in each one of them.

reuters

Thanks to image recognition AI, in particular, machines can recognize images (and by extension internal anomalies of the human body) almost as well as a human (or a doctor). In fact the accuracy rate is a fairly awesome 97%, according to Mr Burby and Dr Vasanawala. Deep learning models trained on 2D and 3D medical imaging data can now identify a brain tumour in patients in a few minutes compared to hours in scenarios where assistive AI isn't present.

All this identification and analysis at the machine level, thanks to artificial intelligence, ensures doctors get more time to do specialty work, and clinic assistants spend less time on mundane analysis. This will have a direct positive impact on the number of patients treated by a clinic or hospital and the quality of their treatment -- which is great news for India's healthcare sector.

In fact, in Max Healthcare in Delhi, AI is already being used to help monitor critical patients to free up intensive care unit (or ICU) beds, and cut down the cost of critical care by as much as 30 percent.

From the doctor to the drug

If that's all that AI can do in the medical industry, you couldn't be more wrong. Believe it or not, but machine learning techniques can re-energize the pharmaceutical sector, not just in India but all over the world, reducing the cost of new drug development by as much as 50 per cent. This should make medicine more affordable in the near future -- at least, one can hope.

reuters

Take this example: Teva Pharmaceuticals, an Israeli drug manufacturer, teamed up with Intel and deployed wearable fitness bands with sensors that monitored the progression of Huntington's Disease, a deadly degenerative nerve disease which currently has no cure. Paired to a smartphone, the gyro and accelerometer inside of which tracked patient's movements, as the wearable fitness band measured other vital signs.

The vital signs of patients were tracked in real-time on a cloud software platform, running advanced machine learning models to assess the severity of degeneration of motor functions of patients.

In Phase I, this has successfully helped in predicting the advancement of Huntington's Disease in 90 patients, and Phase 2 will soon kick in for manufacturing a drug which effectively fights the degenerative symptoms of the disease. Because the drug manufacturer has access to exact symptoms and other vitals of the patients in this trial, they will now be able to aim to design a drug that much more quickly.

Risk vs reward

Obviously, all of this sounds very promising. But isn't there a fundamental risk when a machine tries to intervene before the doctor, and process the data in a certain way? Machine learning is almost as good as humans, but in terms of legal and practical terms how risky is it to completely place trust in a smart algorithm versus trusting the doctor?

reuters

"It's definitely one of those questions that's being continuously asked by both the tech and healthcare industry," said Mark Burby. "While the introduction of meaningful technology can make a big impact on healthcare, the challenge of course is that AI is still in a developmental phase, and certainly not perfect at this stage." There's certainly a role for regulation and other controls to be in place when applying technology in healthcare, Burby emphasized.

"AI today is still in an assisted mode, rather than actually replace the need for humans in a medical function."

Dr Shreyans Vasanawala also agrees that humans need to be in control, at least for now. "The use of AI techniques for image reconstruction can develop problems, for instance, where you may hallucinate new structures and images," he said. "So there has to be a lot of care how these types of algorithms are implemented, to make sure that the reconstructed images are fully consistent with raw imaging data acquired to prevent hallucinations."

ALSO READ: MIT Researchers Have Built An AI To Automate Drug Design So We Can Get Newer Medicines Faster