American College Of Surgeons - Inspiring Quality: Highest Standards, Better Outcomes

Artificial Intelligence is Coming Fast. Are Surgeons Ready?

OCTOBER 29, 2019
Clinical Congress Daily Highlights, Tuesday Second Edition


Artificial intelligence has become reality. Once the domain of science fiction and futurism, the technology now translates web pages online, trades stocks on Wall Street and drives automobiles on city streets. Increasingly, it also supports surgeons and other health care providers in decision-making, diagnosis and other tasks.

“AI is moving so fast right now; it’s amazing,” said Marc Melcher, MD, PhD, FACS, Stanford University School of Medicine, Stanford, CA. At the 2019 Clinical Congress of the American College of Surgeons, he and his colleagues were among a number of teams who considered the impact of AI on surgical practice.

Human vs. Machine

AI often makes news when a computer is able to outperform a human expert in the performance of an intelligent task. IBM’s Watson computer famously defeated two “Jeopardy!” champions on the television quiz show in 2011. A widely covered paper1 published on Oct. 1 in The Lancet Digital Health concluded that diagnostic systems based on a form of AI called deep learning can match or outperform health care providers in accuracy. The authors performed a pooled analysis of 69 studies published between Jan. 1, 2012 and June 6, 2019 that developed deep learning models for disease diagnosis based on medical imaging or histopathology.

A number of studies presented at the 2019 Clinical Congress came to similar conclusions when looking at the performance of AI systems. For example, Dr. Melcher and his colleagues at Stanford found that a computer vision algorithm could identify fat globules in livers for transplantation as reliably as trained pathologists.

In another application, researchers from the New York University School of Medicine, New York, NY developed a decision-making AI system to determine whether a patient would require postoperative ICU care. The system was trained on 512 patients, and used 87 clinical variables to predict whether the patient would later qualify for ICU care based on 15 criteria such as re-intubation, prolonged hypotension or new-onset arrhythmia. When tested on 50 patients undergoing major surgery, the AI system made the correct choice 82 percent of the time. In comparison, a surgeon, anesthesiologist and intensivist correctly assigned these patients 70 percent, 58 percent and 64 percent of the time respectively. Most of this difference was due to health care providers assigning more patients to the ICU who did not end up needing it.

“The algorithm was less conservative than the physicians, and this is very important,” said researcher Francesco Maria Carrano, MD, New York University School of Medicine, New York, NY. “They prefer to have the patient well monitored in the ICU even if it’s not required.”


Call the Chaplain

A human expert can consider up to seven factors when making a decision, said Joseph Firriolo, Boston Children’s Hospital, Boston, MA. In comparison, AI can use thousands.

Yet that impressive power of discrimination requires some supervision, Dr. Firriolo added. The designers of AI systems need to assure themselves that these algorithms are not taking unjustified shortcuts in their internal processing. This can be challenging, because by its nature the logic beneath the most advanced algorithms cannot be fully characterized. In one case, Dr. Firriolo said, researchers using AI to predict which whether ICU patients would die within two days discovered that the algorithm was seizing on a single word in the electronic medical records: chaplain. The computer was picking up a reliable signal, but one without medical relevance.

Beyond that, Dr. Firriolo noted, an AI system is only as good as the data used to train it. As the classic computer science aphorism states, “garbage in, garbage out.”

Dr. Melcher said that the need for high-quality, abundant data is the main challenge he faces in further developing his system for analyzing steatosis in donor livers. He has shown that AI can clearly match pathologists in identifying fat globules, and surpasses them in estimating the total fat content of a donor organ. But in order to develop a functional algorithm, he needs thousands of slides from livers of known fat content. And because his project is ultimately trying to determine whether donor liver fat content affects the success of transplants, that dataset also needs to include patient outcomes as well.

“My next investment really is building up a bigger database of well-labeled images that are connected to concrete outcomes, so that I can run the analysis,” Dr. Melcher said.


The Robot Will See You Now

AI is poised to change the practice of medicine dramatically, said Marcovalerio Melis, MD, FACS, New York University School of Medicine, New York, NY, who collaborated with Dr. Carrano on the study of ICU admission decisions. He imagines a future in which AI algorithms will have real-time access to electronic medical record data as it accumulates, and tell physicians when a patient should go to the ICU or receive a transfusion.

Faced with this prospect, Dr. Melis said, physicians should start thinking now about how to react when an AI-driven system offers advice at odds with their intuition.

“The first question that comes to mind is, if I know that an AI has better judgement than I do, when am I allowed as a physician to disregard this alert,” he said.

The profession also needs to think about ethical issues like liability, Dr. Firriolo said. He and his colleagues searched the medical literature for papers considering the safety and ethical concerns surrounding AI and found none. The computer science literature, however, is rich with explorations of such issues. For example, the rapid advancement of automated vehicles raises the issue of how to convince the public to adopt AI technology once it’s demonstrably safer than a human operator. Ultimately, the same issue will face surgeons as robotic systems gain reliability in performing more sophisticated operations.

“Once we’ve proven that it’s safe and can produce a superior outcome, people should be required to use it,” Dr. Firriolo said.

That may sound unnerving to some of his colleagues, he acknowledged. But the reality is that if computers do eventually achieve general intelligence they will inevitably begin to outpace human surgeons, and we should begin to think now about how best to incorporate the technology into surgical practice.

“If we as surgeons are in it from the beginning, we can dictate how it’s used,” Dr. Firriolo said. “I think that’s the best way to make it safe and make sure that humans still have a role in the care of other humans.”

References:
1. Liu X, Faes L, Kale AU et al. A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis. Lancet Digital Health. 2019;1(6):E271-E297. Abstract