From addressing challenges with diagnosis to predicting potential treatment response, researchers share their thoughts about leveraging artificial intelligence (AI) technology for managing this form of inflammatory bowel disease (IBD).
As AI research has advanced in recent years, there has been a growing interest in how to apply these technologies to help improve the management of people with complex medical conditions such as ulcerative colitis (UC).
“In the past 5 years alone, there have been over 50 published research studies looking at this in UC,” noted John M. Gubatan, MD, a gastroenterologist and researcher at Stanford University School of Medicine in California.
At its core, AI uses computers to process information and make informed decisions using advanced machine learning algorithms. In the case of UC, various information can be analyzed by these algorithms to better predict diagnosis and outcomes of UC. This information may include:
- medical history
- lab values
- imaging and biopsy results
- gene expression or sequencing data
This article explores the emerging role of AI in the management of UC, including how researchers are leveraging these technologies to improve outcomes and clinical decision-making.
Currently, there are no models to reliably predict who will develop UC. In addition, diagnosis is not always straightforward. Currently, a diagnosis of UC comprises:
- the presence of chronic or recurring symptoms
- markers of inflammation in the blood or stool
- imaging and endoscopy results
- biopsy results
“In some people, not all of these factors are present, which makes diagnosing UC challenging — especially differentiating UC from other forms of colitis,” said Gubatan. “AI can be used to create a robust scoring system based on these clinical and other patient-specific factors, which could make diagnosing UC more quantitative and efficient.”
“These systems have been shown to be highly accurate and can reduce reading time and improve efficiency,” agreed Nayantara Coelho-Prabhu, MBBS, a gastroenterologist and IBD expert at Mayo Clinic in Rochester, MN.
“Previous efforts have focused mainly on diagnosis and staging,” noted Coelho-Prabhu. “Utilizing natural language processing and AI technology, the focus is now on developing clinical decision support tools.”
“We are very fortunate to have a wide range of FDA-approved therapies for UC,” added Gubatan. “Given the many therapy options, it is sometimes difficult to decide which therapy to start for which patient and [predict] how they will respond to these therapies.”
As there are so many treatment options and because there are no specific rules for how to choose between them, Gubatan noted treatment selection in UC can sometimes be “trial and error.” This means that it may take a while to find the right medication for each person.
Research suggests that
“AI could be applied to patient datasets to create ‘precision medicine’ models that could help clinicians and patients decide on therapies based on a person’s specific clinical background and genetic makeup,” suggested Gubatan. “Likewise, AI could be used to predict [who] will have more severe UC and guide clinicians to choose more potent therapies upfront.”
Inflammation and tissue damage caused by UC can lead to various complications. Some of these may be serious and may require emergency care. At present, there is no good way to predict who will develop these complications. But Gubatan believes that AI may be able to help change that.
“AI could be used to review thousands of UC patient datasets to reveal novel hidden patterns differentiating UC patients with or without the specific complication and then use this as a biomarker for risk prediction,” he explained.
Coelho-Prabhu is especially excited about this potential use of AI, as it aligns with her own research interests. Her work focuses on the detection of irregular, potentially precancerous cells in people with IBD.
Chronic inflammation in UC can damage and stress the cells of the intestines, which can eventually lead to the development of cancer.
“I am excited about the possibility of developing IBD-focused algorithms and computer-assisted detection systems utilizing large datasets, thus helping to prevent colorectal cancer in this high risk population,” she said.
But she also noted that this work is still in the early phases of development. Additionally, risk prediction models developed using data from precancerous lesions in people without UC may not be as accurate in those with UC. The mechanisms of cancer development are different in those with UC versus those without, meaning the lesions look and behave differently.
Coelho-Prabhu emphasized that AI models for risk prediction in UC — for colorectal cancer or other potential complications — need to be developed using information from people with UC.
The use of AI in the management of UC has come a long way in a short time. Some uses of AI have already made it into clinical care settings, whereas others are still in the research and validation phases.
“AI-assisted endoscopic disease detection in capsule endoscopy and disease assessment scoring in colonoscopy are already available in clinical practice in some parts of the world,” noted Coelho-Prabhu. “AI applications in cross-sectional image analysis are also helping us to detect early disease noted on [imaging].”
Gubatan predicts that the use of AI will soon expand to help support the interpretation of clinical findings that may be ambiguous.
“Sometimes, there could be some variability in interpreting the endoscopic imaging or histologic slides and assigning severity scores,” he noted. “With AI, this process can be streamlined and made faster, more efficient, and more accurate. This is already starting to happen with AI-guided image recognition with colon polyp detection during our screening colonoscopies.”
In the more distant future, Gubatan suggested that AI systems may be able to integrate more information from electronic health records and molecular datasets, such as genetic screens, to support a personalized approach to UC care.
“We live in an age of big data, and molecular datasets have traditionally been difficult to analyze,” he said. “What’s exciting is not only more robust prediction models or clinical scoring systems but also a deeper understanding of the pathogenesis of UC itself, which in turn could lead to more novel therapies.”
In addition, he emphasized that any attempt to personalize the treatment of UC must also consider the unique needs of the individual being cared for.
“Patient preferences and values should also be incorporated into these AI models to maintain a wholistic and patient-oriented approach,” he said.
Coelho-Prabhu added that more research is necessary before many of these AI systems can be part of clinical care.
“To foster equitable and widely applicable algorithms, there is a need for larger and more diverse datasets,” she said. This will help ensure that the models that are developed are accurate for many different groups with UC, regardless of age, sex, race, or ethnicity.
AI has already begun to help with some aspects of UC care, such as interpreting imaging and biopsy results. Ongoing research aims to understand how AI can support a personalized approach to treatment selection, but Gubatan noted that these more ambitious efforts will take time to make it into the clinic.
“Before they can be reliably applied to clinical practice, AI models need to be robustly validated in large, independent cohorts,” he said, echoing Coelho-Prabhu. These types of studies will help ensure that AI can effectively help diverse groups of people with UC and help prevent disparities in medical care.
“The integration of AI into clinical workflows is inevitable and will transform healthcare,” concluded Coelho-Prabhu.