So I think that AI in myeloma is still nascent and still evolving. I think artificial intelligence has had a more profound impact in other diseases and in other cancers, particularly lung cancer and breast cancer, which is where a lot of my research has been up to this point.
But when I was asked to come and present the keynote lecture on AI in multiple myeloma, I was both intrigued and flattered. At the same time, I was a little hesitant and nervous because we hadn't done a lot of work in multiple myeloma with applications of AI. As I started to do some research, I realised that there’s not a lot going on, but I think that represents a tremendous opportunity.
A lot of the work that our group has done has been with the analysis of digital pathology images to predict outcomes across multiple different cancers, both solid and non-solid cancers. I showed some data today on acute myeloid leukaemia, and I think that’s a tremendous opportunity. One of the things I’ve already picked up at the symposium is the need for better ways of characterising risk, particularly high-risk myeloma. I think there are tremendous opportunities there with artificial intelligence and routine data to come up with more granular, quantitative ways of characterising risk.
One piece of data I presented in my keynote talk this morning was on the use of AI with routine fundus images—images of the eye. The data suggests that we can predict which patients are likely to develop multiple myeloma up to ten years in advance. This is very early data, so we’re still wrapping our heads around what it means and delving more deeply into the findings. However, I think this kind of application area is where AI could really start to have an impact in multiple myeloma.
I think there’s a huge opportunity with routine data. That’s been the main mantra of our group in particular: how do you leverage routine data to go beyond the status quo? How can we get to better diagnoses? How can we achieve more granular, quantitative risk stratification, predict the progression of multiple myeloma, and further think about response to therapy? I think that’s another huge frontier for us to explore.
I showed data this morning on using AI in the context of lung cancer to predict response to immunotherapy. Similarly, with CAR T therapy in multiple myeloma, there’s a big opportunity to develop predictors of response and identify who will and won’t respond to therapy. I think there are multiple opportunities across the diagnostic, prognostic, and predictive realms. As we think about precision medicine in multiple myeloma, I believe AI will play a significant role across this continuum of unmet needs.
I’m looking forward to the next three days to continue learning about these unmet needs and how AI could have an impact on multiple myeloma. Another important aspect to consider is the setting we’re in. Here in the beautiful city of Rio, in Brazil, resources might not be as abundant as in Western Europe or North America. One of the things our group has focused on is how to leverage technologies like AI to level the playing field. How can we create opportunities for these technologies in the context of low- and middle-income countries?
Beyond that, we need to think about addressing disparities. We know that many tools trained on data from one population may not generalise to others. As we explore opportunities with AI, we must be intentional about how we develop and validate these tools. It’s important to ensure they are equitable and unbiased, working across diverse populations. At the same time, we need to develop them in ways that allow for scaling and deployment in lower-resource settings. This means leveraging routinely acquired data, such as radiology scans and pathology images, that are already part of the clinical workup for patients.
By doing this, we can exploit the power of technologies like AI while also thinking about dissemination and global deployment. I believe this approach is critical as we look to address health disparities and ensure equitable access to these advancements.
