Artificial Intelligence Touches Cardiothoracic Surgery

Artificial Intelligence (AI) already exists in medicine. AI systems can read chest radiographs faster, cheaper, and—by some accounts—at least as accurately as skilled radiologists, according to Thomas M. Krummel, MD, of Stanford University in California. With approximately 18 robotic surgical devices in development, AI is almost certainly coming to cardiothoracic surgery.

Thomas M. Krummel, MD

Thomas M. Krummel, MD

“AI is already here in a big way,” said Dr. Krummel. “Early stage medical technology may very well have intelligent components, and we will see robotic autonomy in surgery.”

Dr. Krummel explored the growth of AI in medicine during Saturday’s Tech-Con keynote address: Artificial Intelligence­—Hope, Hype or Horror for Medical Tech. The reality, he said, is that AI contains elements of all three.

The hope is that AI can augment cardiothoracic and other surgeons to improve outcomes and reduce overall costs. The hype is that AI will replace surgeons and other skilled practitioners. The horror is that AI will out-evolve and replace humans.

How AI evolves in cardiothoracic surgery depends very much on how cardiothoracic surgeons approach, develop, and accept the technology, Dr. Krummel said. The most important barrier is simply understanding what AI is and what it is not, he added.

It is not a droid, some sort of Star WarsR2-D2 or even an artificial being. AI is nothing more than a series of algorithms: very explicit instructions designed to achieve a specific end. Explicit instructions are not, however, a perfect guide in an imperfect world filled with unpredictable events, as designers of self-driving cars have discovered in the past few months.

Algorithms can follow instructions to perfection, but instructions are not enough. Instructions must be interpreted, even for something as simple as washing hair. Most shampoo bottles carry simple instructions: lather, rinse, repeat. Without some sort of check, AI would repeat the cycle endlessly and never stop shampooing.

Another barrier is the sudden emergence of AI. However, AI is nothing new. It grew out of code-breaking algorithms designed by Alan Turing and his cryptographic team during World War II. The idea of developing and applying algorithms to solve broader problems emerged from a 1956 Dartmouth project as the thesis that anything humans can imagine can be turned into an algorithm and put to use.

It took half a century for computing technology to begin to catch up with the concept of AI, creating intelligent devices that can use knowledge to learn, to reason, to make useful decisions, and prove their value. If the Volkswagen had developed at the same speed as Intel chips, today’s Beetles would run 300,000 miles on a gallon of gas at two million miles per hour and cost four cents.

“Components matter more than we can imagine,” Dr. Krummel said. “By next year, 80% of adults worldwide will be walking around with a smart phone—a super computer—in their pockets. That has a huge impact on the development and deployment of AI.”

Companies like Amazon, Netflix, and Spotify can already predict our preferences in shopping, movies, and music more precisely than most of us can, he noted. MasterCard and VISA can accurately identify potential fraud.

In 2018, IDx became the first medical device approved by the Food and Drug Administration to diagnose diabetic retinopathy. Face2Gene uses facial recognition to diagnose a growing set of hereditary conditions in children, and Suki AI already offers physicians relief from the burdens of electronic medical record (EMR) reporting using a phone-based AI app that Dr. Krummel likened to Suri for physicians.

AI is moving more deeply into medicine than most physicians realize, he added. The Veterans Administration is using Deepmind, developed by Google’s artificial intelligence lab, to predict and prevent kidney disease. IBM’s Watson missed expectations in diagnosing cancer, largely because AI requires that answers be known in order to follow rules to find them. As more is learned about cancer diagnosis, expect AI-based diagnosis to improve.

More immediately, expect practical, cost-effective AI assistance in three areas: AI-assisted robotic surgery, AI-assisted administrative and office workflow, and image interpretation. Also in development are AI-assisted surgical assessment tools that could be used by practices, hospitals, payers, and credentialing bodies to evaluate surgical competence.

“We need to keep humans in the loop with AI,” Dr. Krummel said. “We have the kind of nuanced intelligence that machines lack, at least to date. We’d better buckle up and get into the technology.”