What does AI mean for GI?

Small Talk, Big Topics host Dr. Matthew Whitson is joined by Drs. Renumathy Dhanasekaran, Alexandra T. Strauss and Daniel Penrice for a conversation about the rise of artificial intelligence (AI) and how it will impact gastroenterology and the medical field. Dr. Dhanasekaran is an assistant professor at Stanford University, Dr. Strauss is an assistant professor at Johns Hopkins University and Dr. Penrice is a gastroenterology fellow at the Mayo Clinic in Rochester, Minnesota.

To kick off the episode, Drs. Dhanasekaran, Strauss and Penrice share how they first became interested in AI. Dr. Dhanasekaran said she was skeptical of AI at first, but she started to hang out with people who saw its potential. Dr. Penrice said he was always interested in computers at a young age and received opportunities to apply these skills in his professional career. Dr. Strauss said she became interested in medical school when things were coming together for her and what she was passionate about all collided.

Next, they discuss AI’s current role in gastroenterology. For Dr. Dhanasekaran, she thought she didn’t use AI, but realized everyone uses AI to an extent, through products such as Google Maps, Amazon and even Instagram, which uses the AI algorithm to determine content people are interested in. For gastroenterology, Dr. Dhanasekaran sees AI’s benefits and usage with image analysis endoscopy, which can enhance diagnoses and become faster and more efficient. AI makes it easier to reach patients, and for them to reach providers, Dr. Dhanasekaran said. However, there is a need to be aware of AI’s pitfalls. She also mentioned that medical societies such as AGA will be instrumental in navigating the ethical cadres as AI develops.

Dr. Penrice made an analogy between AI and basic science, saying AI is on the bench and still writing experiments and figuring out what works and what doesn’t from an experimental standpoint. The GI community doesn’t know yet if these will really improve patient care outcomes since it’s just a potential, but needs to make sure it will be fair, just and equitable.

The guests also discuss tools currently being used in practice. Dr. Penrice said AI predictive tools are the most common, like predicting risks, but these tools are not mainstream yet. Dr. Dhanasekaran talked about AI assisted colonoscopy screening where users can choose to do procedures with AI assistance or without. With AI assistance, it could flash and stop the user if they are rushing, but if they are doing a good job already, this won’t be as useful. AI is also used in electronic medical records (EMR), where it makes a summary of the records. AI can screen everything and provide a high level summary with potential risks for the patients. It can also be used to monitor patients – videos are monitored by AI and alert the nurses stations instead of having an individual always monitor the videos. AI is not widespread at Johns Hopkins, according to Dr. Strauss, but can be used to assist colonoscopy at centers. AI is just in the beginning stages and there is a need to see randomized, controlled trials. For AI in hepatology, there is predictive modeling and AI is developing tools for this. These models are great for giving outcomes and who should get transplants, but it could be discounting other factors. There is an argument for some predictive models in machine learning for HCC.

Dr. Dhanasekaran stated that polyp detection devices are in the trial phase and not aware of data. There needs to be a wider update. If this could potentially increase ADR and withdrawal time, it will be very attractive to folks. However, it needs to convince and pass the trial with flying colors. Dr. Strauss had a conversation with private practice groups at a conference and they are really excited about having polyp detection devices for quality metrics and AI in general for resources. The community is excited about how well we can automate and pull data from the EMR.

Next, they discuss where AI is going in gastroenterology in the next 10 years and what they are most excited to see. Dr. Dhanasekaran is optimistic about AI changing medicine. Automation can make things more efficient, but jobs could be affected. The EMR is completely changing, but clinicians need to participate instead of wait for companies to solve problems. Clinicians need to be here from the beginning. Dr. Penrice predicts a period of struggles in the next five years since some AI innovations will work, while others won’t. He is excited about more time to spend with patients and tools to help with tougher diagnoses. He emphasizes that AI isn’t here to take jobs, but become assistants. Dr. Strauss says the field needs to determine how to embrace AI and let it augment, and not replace, care. She looks forward to seeing physicians learn how to trust AI in open mindedness and is excited for how we can use it to help trainees. AI will have an aspect to medical education that will be interesting. She thinks it won’t make trainees immediately proficient but may make the education gap smaller. They also talk about the potential of trainees relying on AI tools as a crutch instead of using their own physical skills.

Lastly, the guests examine about how and when to teach AI. As a current fellow, Dr. Penrice talks about the importance of exposing the trainees to what others are doing. Trainees need to be part of the conversation and know the vision. He also says teaching AI will depend on what technologies and tools are integrated into practice. Dr. Dhanasekaran says that practicing clinicians need to be part of the research space now. The programs developed need to be based on broad, diverse training and that trainees need to start now and partner with engineers. Trainees need to understand all these concepts so they can meaningfully work with engineers and people developing the AI tools. Dr. Strauss states that the trainees will have a spectrum with a wide range of skills. The majority of people won’t care about designing tools but will want to use them. There is a need to build a curriculum around AI and start teaching in medical school.

To close the episode, they discuss ethical considerations in AI, including not using what is accessible and easy, but using all races and all languages in larger models.