Is Laboratory Medicine Ready for Artificial Intelligence?

Newswise — Some say that the invention of the abacus more than 2,500 years ago was the first step on the long road to artificial intelligence (AI). The reason why this simple calculator made of beads and wires is often called the precursor to the modern computer is because the principle—performing repeated calculations faster than the human brain—is the same. Now scientists are harnessing the speed of modern computers and programming them with cognitive algorithms and sophisticated decision trees to help solve complex medical problems. This rapidly advancing science was the focus yesterday of the Chair’s Invited Session, “Is Artificial Intelligence in Genomics Ready for Prime Time?”

Stanley Lo, PhD, chair of the 2017 AACC Annual Meeting Organizing Committee, said he selected the topic because “AI can be considered disruptive technology that significantly changes how we perform laboratory medicine.”

The first of three distinguished speakers, Michael Berger, PhD, examined the use of the MSK-IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets) assay to look for mutations in some 468 cancer genes. Gene mutations associated with a particular cancer can sometimes be found in other cancers too, sometimes making it amenable to the same treatment. Berger’s team takes DNA samples from the tumor and the patient’s blood to help differentiate between germline and somatic mutations and identify the genes causing the cancer. Their technique can more easily identify clinically relevant gene mutations and also pinpoint the pathways affected by the altered gene(s). This targeted gene therapy can potentially cure the cancer and opens up exciting possibilities.

Next, Patrick McNeillie, MD, clinical lead and senior architect of IBM’s Watson Health project, discussed how advanced clinical diagnostics algorithms can allow computers to relate diverse clinical literature with patient data to generate an “outcome report.” To achieve this, AI systems need to “see” unstructured data, such as radiologic images, graphs, slide images, clinic notes, and literature texts, and then convert this to a knowledge base in a structured format. Software must also associate this with other similar “knowledge banks.” For example, cortisol can be either a hormone or drug, and can be differentiated by looking at context clues. So if cortisol is associated with “secretion” it indicates a physiological process, ruling out its use in this example as a drug. This contextual information-based approach helps the system be intelligent.

Complementing these two speakers, Nirali Patel, MD, explained how IBM Watson is helping her institution to help fulfill the promise that “healthcare should be tailored to meet the needs of the individual.” However, Patel noted that “AI can be used to aid physicians in making their diagnosis, but not replace them entirely.”