AI Changing the World of Medicine
By Becca Friedman
Getting sick is the worst. But it’s even worse when a doctor misdiagnoses your symptoms and you get put on the wrong antibiotics for a month — only to get a new prescription after the first one didn’t work.
This often happens due to the doctor’s fatigue or failure to spot a certain detail. We accept this because, after all, doctors are only human.
Scientists now are working on ways to take this human error out of the equation by implementing artificial intelligence into various aspects of medicine. The AI runs a program of “neural networks” resembling the human brain to learn and identify certain things based off of the data it’s been given.
A relatively recent study and project by Google focused on using AI to detect cancerous tissue remaining in lymph nodes after the tumor is removed from a breast cancer patient. The Google scientists taught the AI to be able to discern between healthy and cancerous cells by showing it large data sets of what both looked like, and eventually, the AI was able to tell the difference with astounding accuracy.
“We perform inference over patches in a sliding window across the slide, generating a tumor probability heatmap. For each slide, we report the maximum value in the heatmap as the slide-level tumor prediction,” said the Google AI team of how the detection process works.
However, misdiagnosing the content of these slides presents a great risk for the future of the patient’s health.
When the tissue is found to be cancerous, the next step for the patient is most often to begin chemotherapy. If a patient is misdiagnosed, then she would have to suffer through the harrowing treatment for virtually no reason. Similarly, a cancer patient who is falsely cleared would find the cancer later in her life, often at a higher and more untreatable stage.
The AI detection technology is still imperfect which could be dangerous when being used for such important decisions.
“We discovered tumors in two ‘normal’ slides… Fortunately, the challenge organizers confirmed that both were data processing errors, and the patients were unaffected,” said the Google team.
As it stands currently, this AI program is one best used in pairing with an attentive doctor who can double-check the program’s work and interact with the patient in a comfortingly human way.
Alternatively, AI can be used to complete tedious tasks without the exhaustion of a human.
“Artificial intelligence is also being used to analyze vast amounts of molecular information looking for potential new drug candidates — a process that would take humans too long to be worth doing. Indeed, machine learning could soon be indispensable to healthcare,” wrote science journalist, Bianca Nogrady, for BBC.
The data analyzing algorithms that AI uses could be utilized to potentially find cures for diseases and allow doctors and researchers to focus on other tasks.
“At the same time, we need to accept a difficult truth: If technology is going to improve quality and lower costs in healthcare, some healthcare jobs will disappear. According to one study, Artificial Intelligence is set to take over 47% of the U.S. employment market within 20 years,” said Dr. Robert Pearl for Forbes.
AI taking over human jobs is a concern across all fields of employment. The reality of the situation though, is that human supervision is still necessary for these programs to be used — they’re not fully reliable yet. This is why much of the focus for AI, specifically in medicine, is on data analysis that would take far longer for a human to complete.
But while AI is being used for scanning and analytics, some scientists are reaching to apply it to corrective rehabilitation machines and robots. The robotics labs at Columbia University have been working towards devices that can help aide differently-abled people using a combination of biology, engineering, and computer science.
These devices include a corrective walking apparatus for stroke patients, shoes to help train the walking patterns of children with cerebral palsy, and headgear that can move the neck based off of eye movements for people with ALS.
“These are devices that are working in connection with the human body and they take signals from the human body… and are able to interact with you — so a machine can apply forces with human legs,” said Columbia’s Sunil Agrawal. “Part of the thought process here, is that robots are like training wheels on a bicycle. . . And that’s sort of the philosophy where you use it to retrain the brain.”
These devices have proven to be successful. But, they are still bulky, expensive, and not quite ready for commercial use. The corrective walking apparatus for instance, is a large contraption full of pulleys and wires which require the scientists’ supervision. Agrawal said it’s unrealistic for the intended subjects like the elderly and stroke patients to have these contraptions to train themselves in their homes.
It seems that aiding stroke patients is a popular goal among the Columbia robotics labs, as PhD student, Cassie Meeker’s team is sharing a similar focus. They’re working at creating a robotic hand that functions from reading small muscle movements on a stroke patient’s forearm and wrist. Through using AI, the wrist sensors can decipher the individual’s specific muscle patterns to know what the hand wants to do.
“Basically, what we want to do with these stroke subjects is help them open their hands. Often times with stroke patients they’ll have their hands clenched in a fist and they’re not able to open it anymore,” said Meeker.
Unlike Agrawal, Meeker’s team has a goal of personalizing and commercializing their product: “We’re hoping one day to have this be a take home device — we’re not there yet, but that’s the end goal.”
All of these developers have confidence and high hopes for their products in the future: “Our method could improve accuracy and consistency of evaluating breast cancer cases, and potentially improve patient outcomes,” said Google.
The issue with this confidence is that it hypes consumers up to the point where the companies feel rushed to manufacture a product. The last thing we need in medicine is rushed products that still contain faults and could potentially end up harming people.
Still, some believe that the potential benefits could outweigh the flaws. As Forbes writer, Dr. Paul Hsieh, put it “In other words, these early investigations into deep learning medical AI demonstrate that the algorithms can do as well as (if not better than) expert human physicians in some fields of medical diagnosis and prognosis.”
