Is AI the Killer App for Cancer?

Remember Watson, IBM’s supercomputing Jeopardy! champ and gourmet chef? It’s growing up. Watson skipped university and took a career as a super-elite MD deploying AI software and predictive analytics for research trials at blue-chip healthcare institutions. Yet Watson still needs a whole lot of schooling before making independent decisions and in the meantime, nurses and physicians have the final say. IBM, spotting this opportunity, just launched a new cognitive-computing business division called Watson Healthcare Cloud, with the lofty goal of generating $50 billion a year for Big Blue by 2018.

It’s no surprise that a tech goliath like IBM is betting big on AI in a huge marketplace like healthcare, but innovation tends to be bottom-up—and a number of startups are looking for novel ways that AI applications can improve patient outcomes and slash costs in America’s health-industrial complex. Some are exploring AI tools to design new structures for drugs based on the mechanics of the disease; others are tapping into terabytes of health data to make specific recommendations for individual patients.

So, aside from Mr. Watson, who’s to watch in this complex field? Many AI firms prefer to remain in stealth mode to stay out of incumbents’ crosshairs or refine their technology before enduring public scrutiny. San Francisco-based startup Enlitic came out of hiding last year with the promise of creating medical machines with visual intelligence capable of interpreting X-rays, MRIs, and other medical scans to spot tumors and other pathologies at speeds that an MD simply can’t. The idea is that if you show a machine thousands of pictures of a cancerous growth, it will start to understand the patterns and spot them on its own.

Here are a few others worth keeping our eyes on:

• Atomwise—Uses AI to predict how existing drugs will react with certain biological markers to treat diseases and chronic conditions.

• The Human Diagnosis Project—Initiative that aims to map all human health symptoms to every type of diagnosis—a kind of Humane Genome Project for disease detection.

• Chrono Therapeutics—An app with a wearable device, called SmartStop, that helps smokers quit by automatically adjusting nicotine doses by time of day and the smoker’s typical cravings.

• uBiome—Helps humans understand their microbiome—the thousands of species that live in and on the body that can indicate a wide range of health conditions, including diabetes and depression.

• Ekso Bionics—Developing a bionic suit that uses sensors to help people how to walk again.

• Ginger.io—App that analyzes sensor data from patients’ smartphones to identify mental health issues and help hospitals and providers manage care.

• Kyruus—Software that analyzes reams of medical records to help hospitals, health systems, and accountable care organizations get patients to the right provider and optimize care.

• 3Scan—Converts thin slices of human tissue into interactive 3D pathology slides that physicians use to detect diseases or other issues.

• Foundation Medicine—Analyzes a patient’s genomics and molecular markers to create personalized treatments for that individual’s tumors.

While these intelligent solutions show promise, they lack the bedside manner that’s critical to patient trust. Algorithm outputs should help doctors make more informed decisions—not be a prescription for treatment decisions. Enlitic CEO Jeremy Howard, previously of data modeling firm Kaggle, doesn’t see Enlitic replacing radiologists but significantly optimizing their workflow—flagging suspected tumors to save them time, but letting the physician make the final call. “The people who have the greatest chance to be successful through this machine learning revolution are those who are able to combine their area of expertise with the power of machine learning.”