Medtronic Endoscopy Chief Artificial Intelligence Officer Ha Hong discusses new GI Genius features, design improvements and how other device developers can jump aboard.
The latest version of the AI-powered Medtronic GI Genius system for colonoscopy is smarter, more helpful, and has room for third-party apps developed by other devicemakers.
Those third-party software-as-a-medical-device (SaMD) apps “could be anything,” Medtronic Endoscopy Chief Artificial Intelligence Officer Ha Hong said in an interview with Medical Design & Outsourcing.
“The beauty of the GI Genius system is that it’s already been deployed by thousands of hospitals worldwide,” he said. “And as long as the hospital has the system already, the third-party application developers will be able to tap into that.”
Medtronic gave the GI Genius polyp detection software a new name — ColonPro — in anticipation of other apps running on the GI Genius operating system.
“We have a vision to create an AI application ecosystem [in] the AI Access Platform,” Hong said. “We hope to host a variety of different AI applications that can be relevant for the physicians. Those apps will be downloadable into the GI Genius system … and the physician will be able to choose which app to use at a given time.”
Hong discussed ways other device developers can piggyback on GI Genius and two other key improvements in the latest update to the GI Genius system.
ColonPro’s improved polyp detection
Medtronic’s software development partner, Cosmo Pharmaceuticals, doubled the number of patients in the ColonPro training data to retain the neural networks used for the system’s various algorithms.
That led to what Hong called “a substantial improvement in the performance of the AI system.”
Firstly, the changes decreased the false positive rate by 9%, which means fewer ColonPro false alarms for physicians as they search for signs of cancer.
“The sensitivity of polyp detection before this improvement — this increase of the training data — was already very, very high, [around] 99% or 100%, depending on the test data set that you use. … Now we are literally at the ceiling,” Hong said.
On top of that, the algorithm improvements also made ColonPro better at detecting the most challenging lesions like tiny, flat polyps hiding in the corner of the camera’s field of vision that the system might have missed previously.
“The new release [has] better coverage in terms of detecting those tricky cases even better, and at the same time with less distraction because the false positive rate has been decreased,” Hong said.
Related: AI basics from Medtronic Chief Technology and Innovation Officer Ken Washington
ColonPro automated procedural highlights
The updated ColonPro software automatically summarizes colonoscopy procedures with quality metrics and other details. That includes withdrawal time, bowel cleanliness (Boston Bowel Preparation Scale, or BBPS) throughout the procedure, and key anatomical landmarks visualized within the patient’s colon.
“We also provide whether the physician actually arrived at and physically visualized the first part of the colon in an adequate way, meaning that we detect the visualization of the cecum and whether the visualization was enough,” Hong said.
Six minutes is the standard minimum time for withdrawal to properly inspect the full length of the colon.
“Under pressure or if you’re not careful enough, it is completely possible that you can withdraw too fast,” Hong said. “This kind of active monitoring system automatically checks the withdrawal time.”
That offers an automated quality metric for physicians and hospital administrators. Another feature in the works is a way to tell a physician in real-time how their withdrawal pace is progressing against that six-minute minimum.
The automatic quality metric computation system also gives physicians a breakdown of their activity during the withdrawal: navigation, inspection, instrumentation (snaring a polyp, for example) and cleaning. That information is presented as a horizontal chart with color-coded segments indicating each activity at different times of the procedure. The team is working on ways to add physical location data for the activity reporting, Hong said.
All this reduces the time spent by the physician to provide this required documentation while reducing the variability of the documentation and the procedure itself.
The new procedural highlights feature will one day be able to transmit this documentation directly with a hospitals electronic health records (EHR) system through Medtronic’s new partnership with EHR firm Modernizing Medicine (ModMed).
Preparing GI Genius for third-party apps
The latest version of GI Genius greets users and shows them available apps — including ColonPro —when they power the operating system up
Previously, Hong said, “the software component had kind of monolithic conglomeration of the AI component and the operating system, so when you turned on GI Genius hardware you just saw the GI Genius AI system running and that was it.”
There aren’t any other apps available through the Medtronic AI Access Platform yet, but the updated GI Genius now has an application programming interface (API) for the GI Genius operating system.
“This is essentially distribution-as-a-service. If you are a small startup company that has a really great and innovative idea, you can work on that, you can focus on what you’re really good at. But in reality, it’s going to be very difficult for you to distribute and sell. As Medtronic, we can help you — we have a huge footprint and because of that, if a small company is willing to join our AI Access Platform, they can take advantage of it.”
Hong sees more gastrointestinal apps as a natural starting point, such as laparoscopy or even upper GI. “I wouldn’t be surprised if we go beyond those,” he said.
To get started, third-party developers can go to Cosmo’s Innovation Center webpage to sign up, execute non-disclosure agreements, onboard data and get access to a virtual sandbox and even a physical development kit that’s a simplified version of the GI Genius system.
More with Ha Hong: How Medtronic’s using AI: Artificial intelligence insights and advice