The use of AI in gastroenterology can go beyond ‘cool tools’ to improve practice efficiency

As Artificial Intelligence (AI) technology continues to advance in the field of the digestive system, speakers at Digestive Diseases Week 2022 discussed how these tools can be put into practice to improve efficiency, reduce physician fatigue, and achieve cost savings.

As Artificial Intelligence (AI) technology continues to advance in the field of the digestive system, speakers at Digestive Diseases Week 2022 discussed how these tools can be put into practice to improve efficiency, reduce physician fatigue, and achieve cost savings.

The session, “Improving Your GI Practice with Digital Technologies,” began, with an audience member asking moderators and panelists to name the “coolest” recent development in artificial intelligence. Their responses emphasized that new technologies appear every day, but the innovations that are really worthwhile are those that will help doctors and patients. For example, supervisor Cadman Leggett, MD of the Mayo Clinic, noted that the new technology could create a “fake” image of an esophagus that could trick a gastroenterologist into thinking it was real — which is cool but has absolutely no clinical benefit.

In contrast, a presentation by Cesar Hassan, MD, PhD from the University of Humanitas in Milan, Italy, provided an overview of the use of AI in colonoscopy for automated detection and characterization of polyps. He discussed research showing that AI-assisted colonoscopy could cut the rate of missed tumors in half, and argued that even suboptimal machines could be beneficial because humans perform much worse.

“When you get distracted, you lose everything,” Hassan said, referring to a human predicament that machines avoid. And unlike humans, machines can’t lie or cheat, so randomized trials aren’t necessary to evaluate their performance. He noted that AI tools for detecting polyps have been incorporated into clinical practice almost immediately, but a paradigm shift is needed for computer-assisted characterization and diagnosis.

Despite the obvious performance advantages of AI for colonoscopy, its cost has prevented its widespread application in Europe, where Hassan said it was difficult to convince politicians to pay for an expensive tool that would save money over a very long period of time by reducing Incidence of colorectal cancer. . He called for further studies into the endoscopic practices of society, which could help demonstrate the real-world value of AI tools.

In the United States, the stress is that insurance companies pay for a technology that won’t deliver cost benefits until decades later, when recipients are likely to switch to a different payer, added the next speaker, Tyler Berzin, MD, of Harvard Medical School and Beth Israel Deaconess Center. medical.

His presentation focused on how AI can help make the lives of gastroenterologists easier by breaking the cycle of disengagement and fatigue that is often accelerated by spending a lot of time in front of a screen entering data. The exponential growth of patient data and medical knowledge, dubbed the “data deluge,” can feel crippled, and they need something to support them.

Enter Natural Language Processing (NLP) and Speech Recognition, Berzin continued. These tools can derive structure and meaning from language, allowing programs to extract and organize data for analysis. He noted that the opportunity to integrate these tools into gastroenterology practice is by creating analyzes of quality metrics in a way that is far more effective than what humans can do.

To illustrate this point, Berzin cited research published in Gastrointestinal Endoscopy The researchers developed an NLP algorithm that takes less than 30 minutes to extract data on all colonoscopy procedures performed at their institution since the introduction of electronic health records, while manual review by a human takes about 160 hours to extract data for fewer than 600 patients.1

Berzin also touched on the potential of workflow solutions to transform the clinician experience. These are triage and notification tools that can alert the radiologist which images to prioritize, rather than AI diagnostic tools. Workflow tools may not be as flashy, but they can improve efficiency and have a lower regulatory hurdle to approval.

“The goal of AI in medicine is not yet a promise, but it is an opportunity for us to improve clinical insight by leveraging data and reducing the fast and shallow work that we do, and to replace that with an opportunity for us to think clinicians.” “I would bet that would be a very effective way to combat fatigue if we could focus again on what got us into medicine in the first place.”

The latest speaker was Prateek Sharma, MD, of the University of Kansas Medical Center, who provided a glimpse into the future of artificial intelligence in gastroenterology and our current state along the timeline. For example, research like that described by Hassan has progressed from traditional endoscopy to computer-assisted detection and diagnosis, but the next steps to be achieved could be robotic endoscopy, self-driving scopes, and finally endoscopy without endoscopy.

Sharma said that while this technology has the potential to transform diagnosis, drug discovery and personalized medicine, some of the major barriers include data access, data security, and regulatory issues.

He defined the characteristics of responsible AI of the future: repeatable, secure, human-centred, unbiased, justified, explainable, and controlled.

“Don’t think that’s too demanding, because it’s the same concept that we use for clinical trials of drugs, for example,” he reminded the audience.

reference

1. Laique SN, Hayat U, Sarvepalli S et al. Application of optical character recognition with natural language processing to extract high-quality metric data in colonoscopy reports. gastrointestinal endoscopy. 2021; 93 (3): 750–757. doi: 10.1016/j.gie.2020.08.038

.