This transcript has been edited for clarity.
I'm David Kerr, professor of cancer medicine at University of Oxford. Today, in this brief video, I'd like to talk a little about the application of artificial intelligence in oncology. I'm sure that, like me, you've seen an extraordinary amount of information — misinformation — being presented about artificial intelligence in the mainstream media.
It seems impossible to avoid it, whether in television, podcasts, or print media. Wherever we look, we're being warned that the end of the world is nigh, and that artificial intelligence will soon provide some form of robotic armies that will take over the world and lead to the end of the human race. In fact, some of the founding fathers of artificial intelligence are warning us all.
I have no doubt the application of artificial intelligence for nefarious means, such as hacking, political destabilization, and military applications, have potential for harm. Ain't no maybes about it.
But I guess I see the good.
There's a nice paper published recently in Annals of Oncology by an excellent Norwegian group in which they used machine learning — artificial intelligence — to develop a very robust model capable of looking at the immune phenotype of patients with resectable non-small cell lung cancer and coming up with some very strong prognostic models.
With my dear friend and fantastic collaborator, Professor Håvard Danielsen from the University of Oslo, we've done similar work that we've published in The Lancet and The Lancet Oncology looking at artificial intelligence and machine learning to come up with prognostic models for colorectal cancer. We're using digital pathology — not molecular biology, not cell biology — to interrogate pixels from digital images. The machine can learn to differentiate good from bad pixels: that is, good from bad prognosis.
This is a very strong example of why I think digital pathology and artificial intelligence are partners made in heaven. This same AI approach, of course, is being used by our colleagues in radiology, who par excellence, have access to large digital databases. Again, they're interrogating pixels to determine tumor from nontumor, to determine good from bad prognostic tumors, and so on using these extraordinary mathematical models to separate good from bad, black from white, and up from down.
The other area where I think we will see more use of artificial intelligence is to look at large health databases. Electronic patient records collect extraordinary amounts of what we would call routine data, and by that I mean conventional demography describing the patient's gender, age, renal function, liver function, hematologic function — the things we collect without thinking very much about.
Other data include the number of times patients present themselves to the healthcare system and the number of underlying conditions they have.
Without requiring us to do any clever genetics or molecular biology, could we use these enormous databases to come up with predictive algorithms that can identify patients most at risk of developing a whole range of different cancer types? Or use these databases to predict who's at danger of developing a myocardial infarction within a given period of time? I think the answer is yes.
Given the size and extent of these databases and their applicability, or I might say cleverness of the machine learning systems, I think we'll be able to take a large amount of data from those electronic patient records.
Note that I said the "cleverness" of AI — so that's anthropomorphism. That's the ascribing of human characteristics to inanimate objects. Already, we're wondering if artificial intelligence can become conscious; can it become capable of making its own moral decisions? Could we develop artificial intelligence-driven chatbots that would allow us to support our patients such as those who are suffering with mental health problems?
Is it possible that chatbots could play a role when we come to end-of-life care, supporting patients in that way, also? There is so much to explore. I feel, to be honest, a tingle of excitement rather than fear.
I'd be very interested in your comments or ideas about how AI could be applied to help support us in the wider oncology community.
Thanks for listening, as always. For the time being, Medscapers, over and out. Thank you.
David J. Kerr, CBE, MD, DSc, is a professor of cancer medicine at the University of Oxford. He is recognized internationally for his work in the research and treatment of colorectal cancer and has founded three university spin-out companies: COBRA Therapeutics, Celleron Therapeutics, and Oxford Cancer Biomarkers. In 2002, he was appointed Commander of the British Empire by Queen Elizabeth II.
Medscape Oncology © 2023
Cite this: David J. Kerr. AI in Oncology: Sorting Through Misinformation - Medscape - Dec 21, 2023.
Comments