Today’s Clinical Lab asked Mateusz Tylicki, BSc, product manager at Visiopharm, about the ways artificial intelligence (AI) can help deliver reliable cancer diagnosis and high-quality patient care.
What makes a robust predictive marker for directing cancer treatments?
Predictive markers help determine and separate the responders (patients) from the nonresponders to specific therapy. These could be genes, proteins, or specific molecular signatures of predictive value. What makes robust markers are the ones that are very good at separating the true negatives from the true positives.
In your experience, what are the challenges of developing better and more predictive markers?
Tumors can be quite heterogeneous. Discovering a potential marker and getting some signals is one step, but to be used in the clinical space, a marker has to go through a very robust and long clinical validation process. These steps are complex and time-consuming. There are economic factors involved: High costs are associated with the current R&D. Inter- and intra-laboratory inconsistencies and a significant 10–30 percent analytical error rate in fields like immunohistochemistry (IHC) add to these factors.
How can AI help refine current approaches?
AI can definitely help but it won’t solve everything. What AI offers is an objective and unbiased way of quantifying a response but the wider ecosystem also needs to be robust. AI can act as a stimulus or a driver to highlight where the issues are and help the industry improve.
Are there any instances in your field research where AI helped make the processes more reliable?
Very much so! Using AI and next-generation standardized reference material, our team at Visiopharm was able to identify a number of fixable issues in instrumentation, standard quality control procedures, etc., in the labs we work with. Left undetected, these issues would have generated erroneous results and adversely affected clinical decision-making. There are other instances where our client labs used certain clones that proved to be fairly unstable. Having the quantitative evidence generated by AI helped them make the decision to go with a different, more stable clone. So yes, there are several instances that we've encountered where AI has come in handy.
Are we now able to rely on AI-based decisions to analyze large data sets and predict/diagnose cancers?
AI is employed in mapping and spatial biology studies in drug discovery stages to understand molecular mechanisms. In the studies that I have been involved in, AI is quite mature and fairly robust in delivering insights that correlate with expert assessments and pathologists. However, we still need consistent components, reference materials, and processes in place to ensure reliability.
What's one technology/method in your area of research you’re looking forward to and why?
I'm going to be slightly biased: Applying the combination of AI and next-generation reference materials and calibrators in tandem with IHC is one of the best bets to get the analytical error rate in IHC vastly down by an order of magnitude to where clinical chemistry and other clinical disciplines are today. This is an exciting time with plenty of opportunities in both diagnostics and therapeutics. Next-generation technologies will somehow redefine how cancer is perceived—no longer as a chronic, incurable, terminal disease but as a condition with reasons to be optimistic about.