At first, it sounded impossible. But that’s exactly the kind of challenge that Dr. Chih Lai and Jihun Moon ’24 MS, two artificial intelligence experts at the University of St. Thomas in Minnesota, enjoy more than anything.
In fact, it’s their motto: “Making the impossible, possible.”
In collaboration with colleagues at the University of Minnesota Surgical Center, Lai and Moon developed an AI-based surgical analysis system. The system demonstrates remarkable precision, helping surgeons implant a device during a live operation to an accuracy of approximately 0.4 mm.
For Lai, a professor of software engineering and data science, it sounded out of reach … until it wasn’t.
“Those kinds of requests on artificial intelligence, they make me grow,” Lai said. “That kind of pressure pushes me to keep asking questions, and it’s there, in the research, where we can make the impossible, possible.”
The breakthrough may someday aid medical teams around the globe. The University of St. Thomas and the University of Minnesota have recently filed a joint patent for the innovative technology.
“We’re looking all the time for ways to improve medicine,” said Dr. Dwight Nelson, a senior fellow in functional neuro-orology at the University of Minnesota and a member of the research team. “One of those ways we became interested in was Vision AI, which can analyze an image much like our own eyes do.”
The AI-based analysis program, which examines fluoroscopy images taken during surgery, is designed to improve the implantation of neuromodulation therapies. Once implanted the treatments give off electrical impulses to control overactive bladder or chronic urinary retention, among other issues.
“When you’re looking to apply AI for medical purposes, it cannot be roughly correct; it must be precise,” said Lai, who has previously partnered with industry on a variety of patented AI projects, including the detection of counterfeit products, predicting power generation and monitoring Parkinson’s disease patients.
For more than a year, Lai and Moon, a recent data science graduate student at St. Thomas, worked alongside the surgical team to identify possible solutions. They poured over images from prior surgeries, using the data to develop their new AI framework. Eventually, their goal is to train the AI to go beyond static photos and provide analysis during live surgeries.
They faced several hurdles along the way. The team only had access to 56 images from prior patients, a relatively small amount of data to train artificial intelligence on. The images themselves were also noisy and full of various surgical equipment that their system would need to learn to recognize.
“At first, there was not an answer in front of us,” said Moon, who now serves as the university’s AI Innovation Fellow. “Each day we tried to find the pathway there. It was not a direct path, but we found it.”
One of the biggest challenges was less technology focused: learning how to communicate with their fellow professionals at the University of Minnesota.
“In many ways, we spoke different languages,” Moon said. “On one side, we had the artificial intelligence expertise, and on the other, we had medical professionals. In order to move forward, we had to find ways to meet in the middle.”
In the end, it paid off. The surgical analysis tool, and its unprecedented precision, is expected to give surgeons a significant new tool to improve success rates.
“What we found was, for the patient, using visual artificial intelligence leads to a better outcome, something that’s much more predictable, and can be easily measured over time,” Nelson said.
Nelson, a former professor at the University of St. Thomas, originally reached out to his old colleague, Lai, to explore the potential cross-university partnership. Both believe that artificial intelligence will play a significant role in a new generation of medical breakthroughs.
“We’re basically just touching the tip of the iceberg when it comes to AI, but these improvements are not going to happen overnight,” Nelson said. “You need a lot of data to do this work, and it’s often hard to get access to.”
The AI-based analysis software is currently under review with the U.S. Patent and Trademark Office. The team plans to continue refining their methods in the coming years.
“As we continue to make these advancements, AI will be able to make medicine and medical treatments much more personalized,” Lai said. “The potential for AI helping human life has a very bright future.”
