As a current member of the heathcare workforce, there was one subject discussed in Chapter 2 of Dr Todd Kelsey's book Surfing the Tsunami that struck a particular chord with me. In the subsection Health and Diagnosis, Dr Kelsey introduces and discusses AI in healthcare and particularly the progression of machine learning and AI in performing complex decisions or diagnosis that currently we have to rely on a physician's expertise to determine. Dr Kelsey mentions that "...there have already been instances of algorithms that have been proven to be more accurate than doctors with diagnosis" (Kelsey 2018). I wondered to myself, "How far has this already progressed?" and "Could AI/ML/computers really be the next generation of medical doctors?"
As advanced as healthcare is these days, in many other ways, it would be considered to be relatively archaic compared to the technology and advancements that other industries have leveraged. Medical records were primarily recorded and kept with pen and paper only a few years ago! So while AI and technology advancements continue to astonish in many industries, healthcare felt to me like a slow adopter and relatively safe field to be in. Not only does it seem to be a late adopting industry, medicine and the application of therapeutic intervention is an art form based in scientific knowledge. The difficulty with medicine is people. And by people, I really mean variance. Everyone is different, has different genetics, responds differently, etc. It's almost impossible to know the right answer when there is so much variance from person to person and how one person reacts to a therapy, medication, or treatment, is completely different than the next. So, it seems to me that the skills and knowledge of physicians, cultivated by years of education and practice couldn't possibly be replaced by computers, right? Well, to be honest, I'm not so sure anymore.
Google has multiple AI models that are being implemented in medicine. In fact, my own organization, Northwestern Medicine, partnered with Google to develop a model designed to identify lung cancer from screening tests. A skeptic may wonder, as I did, Ok, fine. But how well did it perform and is it really that accurate to be relied upon for such a serious concern? This model can identify cancer better than a radiologist with 8 years of experience and achieved expert-level accuracy (Wiggers 2019). See that picture above? This is a chest x-ray of pneumothorax, commonly explained as a collapsed lung. The highlighted area within the image on the right was generated by another one of Google's AI models. This was missed by multiple radiologists but accurately identified by the AI model (Wiggers 2019). Google has also developed DeepMind which is an AI model that can recommend the proper line of treatment for 50 eye diseases with 94% accuracy, thanks to advances in processing visual information through eye scans (Wiggers 2019).
Although incredible feats of advanced AI exist in medicine today and will certainly develop and progress further, I'm not convinced that we will see AI replace physicians. At least, not yet. While the data above certainly serves as strong and convincing evidence for the progression of AI and machine learning within healthcare, the gap is too large to close rapidly. I expect to see more and more integration of AI in medicine and a vast improvement in accuracy and quality of care. But I also expect physicians and clinical professionals to use these models available as tools in assisting them to provide high-quality healthcare, not replace them altogether.
Dr Kelsey points out that machine learning involves the performance of complex algorithms, supplied by an incredible amount of data and analysis (Kelsey 2018). Dr Kelsey compares this learning technique to the way the human brain creates neurons, developing a network of decision pathways and optimal answers through the development of a new generation of computer chips called neuromorphic chips (Kelsey 2018). Could a development like this shorten the timeline or close the gap? Yes, of course. I fully expect to see changes to both hardware and software that we currently use and from that, we most certainly could progress to the point where machines can not only perform all functions and decisions of a physician, but do them better. The parabolic advancements in technology, AI, and machine learning is just getting going and certainly warrants the attention of leaders in healthcare. Computers don't get tired, they don't have 'a bad day', they don't act on emotions, and for many other reasons, can be much more dependable than humans. I foresee a heavy AI integration over the next decade and fully expect to see physicians utilizing AI to assist them in not only performing surgeries or procedures, but making in-depth, detailed, and accurate clinical diagnosis and treatment plans. Will they get to the point where we no longer need humans in medicine? We shall see.
Until next time,
References:
1. Kelsey, T. (2018). Surfing the Tsunami; An Introduction to Artificial Intelligence and Options for Responding.
2. Wiggers, K. (2019, December 3). Google details AI that classifies chest X-rays with human-level accuracy. VentureBeat. https://venturebeat.com/ai/google-details-ai-that-classifies-chest-x-rays- with-human-level-accuracy/
No comments:
Post a Comment