Imagine taking a pill no one else can take, guaranteeing good health only for you. That’s the future of AI in healthcare, far beyond chat GPT. AI already analyzes medical scans of your body with an accuracy that often outperforms professionals. During my PhD, I’ve seen first-hand how AI in healthcare is transforming radiologists’ work, making diagnosing conditions like multiple sclerosis easier and more precise. And this is just the tip of the iceberg regarding AI’s potential in healthcare.
So what will be AI’s impact on healthcare? It will be a game changer, not only for radiologists but for everyone. It will make healthcare more accessible and cheaper, enhance diagnostic precision, personalize patient care, streamline administrative processes, save professionals time, save lives, and much more. Ok, that may sound too good to be true, and it’s all in the future. So what’s implemented right now in medicine that professionals, or even us, potential patients, can use based on AI technologies? Before we dive in, I just want to say that I’m not a medical expert or healthcare professional. I was a PhD student in biomedical engineering and have now switched to being a being a full-time educator and popularizer of AI. My goal is to make artificial intelligence more accessible through articles like this one and other educational resources. During my research for this video, I had the pleasure and opportunity to chat with a few medical AI experts, including Mona Flores, head of medical AI at NVIDIA, and the previous chief medical officer and certified cardiac surgeon, who gave me amazing real-world examples of how AI is currently being used from the nearby hospital up to your wrist if you are using a smartwatch. Before we dive into this episode, focusing on healthcare, I just wanted to mention that I will be attending the NVIDIA GTC event later this March, where professionals like Mona talk about AI’s applications in various industries, including healthcare. Check out her talk on how artificial intelligence is powering the future of biomedicine. All the events are completely free to attend remotely.
AI is already transforming healthcare, from research like my PhD work for automating brain lesion detection to companies like Siemens Medical and G.I. Genius revolutionizing tumor identification and improving colonoscopies. Even your Apple Watch is part of this wave, monitoring heart rhythms 24 hours a day to keep you safe. A cool real-life use case of AI Mona shared with me is a digital stethoscope from Eco-Hell that can listen for you and spot check for irregular murmurs for doctors instead of having to learn and understand all the different sounds. Yes, this already exists. Mona also shared another application that is even closer to us. In fact, it may already be on your wrist right now. The Apple Watch tracks your heart health and will notify you if something is going on, like an irregular heart rhythm, which could save you by preventing potential strokes. What is the moral of the story?
Get an Apple Watch or an Intelligent Watch just in case, before it’s too late. It may be a good investment. But you most certainly already knew about that. Unfortunately, that’s not all AI can do. There are tons of technologies already helping professionals run MRIs at a lower price and improve the patient’s experience by reducing the time it takes to run the scans. And it’s not only beneficial to experts in hospitals. We can now say there’s an app for that, even for a health issue. Thanks to the first-ever FDA-approved AI device, we can detect early signs of diabetic retinopathy, a complication that affects your eyes caused by diabetes, helping you prevent damaging your retina without waiting months, if not years, to see an ophthalmologist. AI can also help directly with drugs. For instance, DeepMind’s AlphaFold makes drug discovery faster than you can find your keys late at night. Okay, maybe not that fast, but it does that in a similar manner. While you touch and feel your keys one by one to find the right one in the dark, AlphaFold figures out the complex 3D shapes of proteins for creating new medicines.
If a protein is known to contribute to a disease, scientists can use alpha-fold to understand its shape and then design a molecule or a potential drug that can attach to the protein, inhibit its function, and thus treat the disease. The process significantly speeds up the early stages of drug development, making it quicker and cheaper to find starting points for new medicines. Likewise, AI can help with the dosages of these medicines, catching mistakes before they happen. It can do the same with food, keeping you on a good diet. It’s probably an overkill method versus just skipping the drink and dessert this time. Sometimes, medicine isn’t enough, even if you take care of your body, and you end up relying on surgery to remove or fix parts of yourself. A cool use case is one I was directly involved with in an internship I did at CAE Healthcare. I worked on robotic-assisted surgeries with incredible precision, either to help surgeons practice or allow for remote surgeries. It’s as if the surgeon has an extra pair of hands, super steady and accurate, with a second brain that can correct the small imprecision moves, making operations safer and helping patients heal quicker. And finally, of course, there’s ChatGPD, which is useful even in healthcare. We can use those text-based models to handle scheduling, transcribing what’s being said in appointments, and even help write and plan prescriptions automatically, as doctors say it in the Imagine a toon that does all the boring stuff for doctors so they can focus more on you. But there’s one problem, or one limitation, with all these super-promising AI-based applications. It requires lots of data. From automated brain images to patient information, the healthcare industry is composed of almost only sensitive and protected data, so you cannot really scrape the whole internet and build a good model like chatgpt.
Chatgpt isn’t good for medical diagnosis; it’s just like Google, if you use it. Right now, for an irregular cough you had, you’ll be convinced to have cancer or something even worse in a few minutes. Fortunately, there are approaches to counter that. I worked on one of the available solutions for training AI models with such sensitive data in my PhD, called federated learning. This approach allows AI models to learn from multiple locations without taking the data out of the hospitals themselves. It achieves that by training individual models in each location in a step-by-step fashion, where, at each step, we combine these models into one more general and better-performing one. This way, we can develop powerful AI systems that can detect brain lesions or other diseases from MRI scans. This both preserves patient confidentiality and enhances the AI’s diagnostic accuracy, providing more data than a single hospital could ever produce, sharing the AI’s knowledge but not the data it requires to gather this knowledge. This approach is really interesting for current and future applications of AI in this field. While these current applications are all super exciting, the future looks even more promising. Think of today’s AI applications as your first BlackBerry with internet access. I remember the feeling of being one of the first in my class to have it. It was so cool. Now compare this 2007 BlackBerry with the iPhone 15.
In fact, there’s little we can compare. So much has changed. That’s what’s going to happen with every existing solution we’ve seen. The Apple Watch Checking Your Heart is just one of the first accessible monitoring applications we have access to. There will be many opportunities for remote health technologies as AI and our mobile devices improve. The first game-changer application will be Virtual Health Assistance. There will be even more on your back than an annoying acquaintance who doesn’t understand your lack of signals to go out with them. It will check on you 24 hours a day, offering constant health monitoring and personalized advice tailored to your unique genetic makeup and lifestyle. This leap forward will shift the focus from hospital-based care to home-based care. Personalized medicine will make all of healthcare similar to visiting an optometrist, where just as you receive glasses tailored specifically to your vision, AI will leverage genomics and real-time data to create treatment plans and drugs designed for an individual’s unique condition. going away from the one-size-fits-all approach and moving towards unique patient-specific treatments to the ultimate pill I mentioned at the beginning of the video. This transformation will make health care more accessible and affordable. Diagnostic tools and knowledge will be at everyone’s fingertips, just like the eye-picture example we saw for diabetes, enabling early detection and intervention like never before, directly from your home. No more six months of waiting to see a professional, and once you are finally in, hours of waiting in the waiting room. Just as electricity and the internet became indispensable, AI’s adoption will become necessary for healthcare companies to stay competitive. However, the future is not without challenges. Biases in AI and data privacy concerns are issues that will most certainly slow down progress. Healthcare has much graver consequences than generating unfair text with ChatGPD. Not that it’s not important, but small GPT mistakes might not kill as many people as a small pill dosage mistake. We need to fix the biases before deploying the models. We can’t test in the wild like OpenAI does and fix when something happens.
It may cost lives. Still, it’s important to work on implementing it in hospitals faster, as it will change professionals’ work. AI will enable them to spend more time on direct patient care, improving job and patient satisfaction, and, even more importantly, reducing overworked professionals and burnouts. So I guess it’s certainly not all that bad, even if it’s biased and imperfect. The future of AI in healthcare is not just about technological innovation or replacing professionals. It’s about transforming the field to allow for better, more efficient, and personalized healthcare for everyone. Medical advice, consultations, and even diagnosis will soon change forever, especially in remote and underserved areas. In conclusion, the transformation of healthcare by AI has already started, and it will only grow from now on. Remember that even though ChatGPT and the current companies I mentioned are super impressive, this is just the start, and what we see are the worst results we’ll ever see, as it will only keep improving. Do you still think this is a Terminator-like future?
