That’s because up to 25 per cent of the millions of people who undergo inpatient operations each year around the world experience negative complications either during or after surgery, according to the World Health Organization. These adverse affects can range from soreness at the incision site to internal bleeding to death.
Although not all of these adverse events are caused by the actions of surgeons, some are, and Madani — a general surgeon with the Sprott Department of Surgery at the University Health Network (UHN) in Toronto — wants to reduce that risk.
He was researching the techniques and thought processes used by the most elite, highest-skilled surgeons, when a group of data and computer scientists suggested he could use artificial intelligence (AI) to mimic their minds.
“I was a big skeptic, actually, for the longest time,” Madani said. “That’s a big statement to make.”
The resulting collaboration produced a prototype that uses computer vision — a field of AI that trains computers to interpret and understand images — to identify in real time areas of an organ that are safe to dissect, and those where it is dangerous to do so.
WATCH | How artificial intelligence identifies safe dissection areas during gallbladder surgery:
It’s part of a flurry of activity in recent years among researchers, health-care workers and companies who are attempting to harness the power of digital technology to provide better medical care.
Madani’s technology is still in the early stages, and currently only applicable to gallbladder surgeries. But, he says it has the potential to improve surgery around the world, particularly in rural communities, remote areas and lower-income countries that lack surgical expertise.
Other experts agree, although they say there are still challenges to overcome before its potential can be realized.
How the technology helps guide surgeons
When surgeons perform gallbladder removal surgery, they make a “keyhole incision” in the patient’s stomach, insert a camera into the abdomen and use specialized tools to cut away and remove the organ.
Madani’s technology projects coloured areas onto the video monitor the surgeon uses to see inside the patient’s body. Green means that area of the organ is safe to cut, red means it’s not.
Another iteration uses a heatmap-style projection that changes colour based on the model’s confidence as to where the safe area is.
WATCH | Dr. Amin Madani explains how the prototype could help guide surgeons:
The prototype was developed by feeding hundreds of hours of videos of gallbladder surgeries into a software program and integrating annotations from expert surgeons identifying where they would dissect. After analyzing the data frame by frame, the algorithm starts to recognize patterns and develops the ability to make independent decisions.
The algorithm was able to consistently identify “go” and “no-go” zones as well as the liver, gallbladder and hepatocystic triangle with an accuracy ranging from 93 and 95 per cent, according to a 2020 study of 290 videos from 153 surgeons that was published in the academic journal Annals of Surgery. Madani was the lead author.
“It’s like I have a panel of experts standing, watching me over my shoulder, guiding me, navigating me and helping me not get into trouble during that operation,” said Madani.
Dr. Daniel Hashimoto, a surgery instructor at University Hospitals and Case Western Reserve University in Cleveland, Oh., who collaborated with Madani on the study, said the real promise of the technology lies in its ability to help surgeons better understand what they are perceiving when making surgical decisions.
“The hope is to say, well, can we bring in a second pair of eyes into the operating room — in this case, machine eyes — to ensure the surgeon is seeing what they think they’re seeing?” said Hashimoto.
The next question is: will it actually improve the performance of surgeons in the operating room and reduce complications?
That’s a difficult question to answer from a research perspective, according to Hashimoto, because clinical trials studying adverse events require large numbers of patients to participate. But Madani is determined to find out.
His team has already tested the prototype during live surgery to make sure it works properly, and now they are seeking approval from UHN’s ethics board to conduct further research.
The AI needs more data, videos
Another challenge lies in expanding the technology to other surgical procedures.
Gallbladder surgery is one of the most common operations, so it was relatively easy for the researchers to procure videos of successful surgeries. But tracking down useful videos could become more difficult with less common procedures.
“Most modern machine learning fundamentally depends entirely on the data,” said Frank Rudzicz, a computer science professor at the University of Toronto and an expert in artificial intelligence in health care.
“If it has very few examples of something, it just won’t learn the characteristics of that thing, and it’ll … perform very poorly.”
WATCH | The challenges of integrating artificial intelligence into healthcare:
Rudzicz said another challenge is designing the technology so that it augments the surgeon’s performance, without causing a distraction.
“One thing we don’t want is the surgeon to be looking at this video and everything’s lighting up like a Christmas tree,” Rudzicz said.
Madani said he’s well aware of the need for data and is already in talks with other experts about creating a global repository of surgical videos.
Next, pending research approval, Madani plans to test whether the technology actually improves the performance of other surgeons, thus reducing negative surgical complications.