video


Dr. Daniel Hashimoto: Using Machine Learning to “Moneyball” Surgery

08.19.19
Developing the Collective Surgical Consciousness

At Worrell’s 2019 Clinical Advisory Board event, we were excited to host the Associate Director of Reseach, Dr. Daniel Hashimoto. In his presentation to Worrell’s team of researchers and designers, he shared how his team of collaborators from top academic establishments in the nation are using artificial intelligence to improve the quality of surgical training and democratize access to surgical expertise. He argued that the future of surgery will not be defined by machine automation, but by augmentation. Humans will work side by side with machines to create the best possible surgical outcomes.

SOURCES OF INSPIRATION FOR RETHINKING SURGERY

Dr. Hashimoto told the story of reading Michael Lewis’s bestselling book Moneyball while in graduate school. The book, which has since been turned into a blockbuster Hollywood movie, documents the story of the Oakland A’s executive Billy Beane. Working with a smaller budget than the rest of the MLB, Beane couldn’t afford to sign the players that hit the most home runs or threw the most strikeouts, so as Hashimoto explained, “He had to explore what are the other things that happen over a course of a game that contribute to wins and losses?” To answer this question, Beane got creative and did a statistical deep dive to find the less obvious metrics that correlated with a team’s chance of winning. Beane found players with the stats that unintuitively correlated with more wins. Then he built the A’s roster accordingly. This strategy led the low-budget A’s to unexpected success and ultimately transformed the way Major League executives develop their rosters. After reading the book, Hashimoto asked himself, “Could we Moneyball surgery?”

Dr. Hashimoto had a hunch that there are a lot of little things that happen over the course of a surgery that surgeons don’t recognize as being a big deal. He asked, “How do we find these subthreshold events that happen in the course of an operation that contribute to a complication or a death?” Unfortunately, the way the healthcare system stores data regarding surgeries is not conducive for a Moneyball approach to surgery. He explained, “We’ve been storing information in very broad and general terms. If there was a complication, our records don’t provide any information about how it happened. For example, we don’t have any idea how that bile duct injury that occurred during a gallbladder surgery happened. We just know that it did.”

“That’s a binary decision about yes or no, what we’re more interested in is how artificial intelligence can influence decision making in real time, similar to the way our cars do.”

Dr. Hashimoto needed to find a way to document the details that led to complications or successful surgeries, so surgeons could learn from that information and improve their performance moving forward. As he thought about this problem, once again he received inspiration from a source outside the field of medicine. Within healthcare AI is used for lots of useful things such as identifying whether a mole is cancerous, but Dr. Hashimoto wanted to go beyond that. He explained, “That’s a binary decision about yes or no, what we’re more interested in is how artificial intelligence can influence decision making in real time, similar to the way our cars do.” He saw his peers developing self-driving cars that could identify whether an object was a person or car, and it could also identify whether that person or car presented a safety threat. Dr. Hashimoto explained, that he doesn’t just want AI to help a surgeon identify what they are looking at, he wants AI to help surgeons decide what to do next.

BRINGING HIS IDEAS TO LIFE

Dr. Hashimoto walked the audience through the  details of how his team is developing AI that can understand and predict surgical scenarios during Sleeve Gastrectomy Surgery. For this project, Hashimoto’s team took video of 88 of these procedures. They then had two surgeons annotate the video to identify what was happening in the surgery for each frame of the video. His team then split that data into two groups. They used 70% of the data as a training set, to inform the AI system on how to identify the stages of the procedure. They used the other 30% of the data to test the AI system.

The system they built identified the stage of the surgery with 85.6% accuracy. The fact that it still got it wrong about 15% of the time may be concerning, but Dr. Hashimoto noted that the two surgeons who created the data that trained the machine only agreed with each other 86.2% of the time. Since they can only build AI systems that are as good as the humans who are developing them, his team recognized they needed to move beyond using just two surgeons to train the AI.

“We built an educational platform, so we could provide immediate value for surgeons who are doing teaching or going through residency, and also so we could train our systems on what is happening in the context of an operation.”

With the goal of creating a solution to get more surgeons to contribute to the training data, Dr. Hashimoto and his team are developing the AI analytics platform behind the original training platform called, “Think Like a Surgeon." The original platform was developed by Dr. Amin Madani at University of Toronto. He explained, “We helped develop this training platform called ‘Think Like a Surgeon.’ It’s like crowdsourcing surgical guidance.” Many surgeons willingly annotated videos of more procedures to provide training for new doctors, and that data informed the development of the AI. Dr. Hashimoto said, “We built an educational platform, so we could provide immediate value for surgeons who are doing teaching or going through residency, and also so we could train our systems on what is happening in the context of an operation.” As Hashimoto’s research teams continue to increase the number of surgeons contributing to the training data, they are developing more helpful and accurate AI systems that can identify and predict the events that will contribute to complications.

THE BIG PICTURE IMPLICATIONS

In a world where most decisions are supported by ratings and reviews from people with relevant experience, Hashimoto’s teams believe surgery should be the same way. Regarding surgeons approaching new or complicated surgeries he said, “It’s silly to depend on your own intuition in today’s technology driven environment.” The projects Dr. Hashimoto shared in his presentation are just the beginning of what’s possible for AI assisted surgeries. He explained, “We are pushing for a centralized repository of surgical data.” He wants to build systems that can collect and distribute the expertise of all the surgeons in the world. This would improve the quality of healthcare as every surgeon could have access to information from the best surgeons in the world.

“The near future of surgery will not be automation. It will be augmentation."

Dr. Hashimoto concluded his presentation saying, “The near future of surgery will not be automation. It will be augmentation.” Surgeons will be working in collaboration with machines and they will be supported by the collective expertise of everyone that has done that surgery before. Dr. Hashimoto refers to this idea as The Collective Surgical Consciousness

DESIGNING THE FUTURE OF HEALTHCARE

At Worrell, we’re grateful to be partnering with healthcare’s leading innovators like Dr. Hashimoto and the rest of the members of our Clinical Advisory Board. If you’d like to learn more about the experts and institutions we collaborate with and how we are working to design a brighter future for healthcare, please reach out. We look forward to starting the conversation.