Can artificial intelligence determine prognosis in patients with ulcerative colitis?

Medicom Medical Publishers interviewed Dr Yasuharu Maeda (College of Medicine and Health, University College Cork, Ireland and Digestive Disease Center, Showa University Northern Yokohama Hospital, Japan) about his innovative research at the intersection of artificial intelligence (AI) and endoscopic imaging. His work primarily focuses on enhancing the accuracy and efficiency of diagnostic procedures for gastrointestinal diseases, with a special emphasis on inflammatory bowel diseases such as ulcerative colitis (UC).

At the 19^th Congress of ECCO, Dr Maeda presented a study titled "A novel artificial intelligence-assisted Image Enhanced Endoscopy assesses accurately 'vascular-healing' and predicts long-term clinical relapse in patients with UC: a prospective cohort study" [1]. This presentation highlights his latest research findings, showcasing the potential of AI-assisted image-enhanced endoscopy in the management of UC. By accurately assessing vascular healing and predicting long-term clinical relapse, Dr Maeda's work offers hope for improved patient care and outcomes in the field of gastroenterology. Medicom contacted him to answer some questions.

Can you explain the significance of microvascular findings in managing UC and how they've traditionally been identified?

“With image-enhanced endoscopy (IEE) and magnifying endoscopy, microvascular findings on the colorectal mucosa have been observed in detail. This approach has a stronger correlation with histologic activity and long-term prognosis compared with white light endoscopy (WLE) assessment. Furthermore, it allows on-site assessment without the need for biopsy, thus reducing pathologist effort and cost compared with histology. However, special training is required to achieve high accuracy,” Dr Maeda noted. “Therefore, it is not yet widely used in clinical practice.”

What challenges do specialist endoscopists face in detecting inflammation in UC, and how does AI-assisted IEE address these issues?

“There are 2 main challenges,” Dr Maeda said. “First, even among specialists, endoscopic scoring by conventional WLE varies widely between examiners. Secondly, there are cases where inflammatory activity remains histologically even when the patient is judged to be in remission by WLE.”

“AI-assisted IEE provides an objective assessment, reduces inter-operator variability, and identifies microinflammatory findings not seen with WLE,” highlighted Dr Maeda.

How does the new AI-based NBI system work, and what makes it adaptable to various commercially available endoscopes?

“When narrow-band imaging (NBI) images are acquired, 2 class classifications –Healing and Active– are the immediate output. To adapt the AI algorithm to different endoscopes, training images were collected using different commercially available endoscopes.”

Could you elaborate on the concept of 'vascular healing' identified by AI and its importance in predicting UC relapse?

“Vascular findings are known to correlate with inflammatory activity.” However, Dr Maeda noted, “vascular atypia may persist even when the patient appears to be in remission on WLE. This is the same as histologic inflammation. It has been reported in the past that residual vascular atypia is a risk factor for UC recurrence. The AI assists the endoscopist by providing an objective diagnosis.”

In your study, how did the AI-based system's identification of vascular-healing and vascular-active mucosa correlate with clinical outcomes in UC patients?

“In patients in clinical remission, there was a significant difference in the cumulative relapse rate at 12 months after endoscopy between the 2 groups (3.0% in the vascular-healing group vs 23.9% in the vascular-active group).”

How did the performance of the AI system compare between expert and trainee endoscopists in terms of sensitivity, specificity, and accuracy? 

“AI-assisted vascular healing was evaluated during colonoscopy by 27 endoscopists, comprising 11 experts and 16 trainees, who evaluated 35 and 65 patients, respectively,” summarised Dr Maeda. “The sensitivity, specificity, and accuracy of the AI-assisted vascular-healing predictions of clinical relapse within 12 months did not differ significantly according to endoscopist experience: 100%, 39.3%, and 51.4%, respectively, for experts and 90.0%, 38.2%, and 46.2%, respectively, for trainees (P>0.95, P>0.95, and P=0.84, respectively).”

Based on your results, how do you envision the AI-based NBI system changing the future management of UC in clinical practice?

“The AI-based NBI system provides precise, objective, and on-time treat-to-target. It allows treatment changes to be made on the spot. In recent years, the number of patients with UC has increased worldwide, and it is difficult to create an environment where specialists can necessarily perform endoscopy,” Dr Maeda mentioned. “The emergence of objective treatment targets through AI is expected to help equalize treatment.”

Are there any limitations of your study that should be considered when interpreting the results?

“Two main limitations are that it is a single-centre study and there is a lack of direct comparison between AI-based vascular healing and histological remission in predicting relapse. We are planning a multicentre randomised-controlled trial to remedy these issues,” Dr Maeda assured.

What are the next steps in research or development for AI-assisted diagnostic tools in gastroenterology?

“The next goal of this research is to obtain regulatory approval, market the product, and introduce it into clinical practice. In the meantime, we are developing an automated detection system for UC-associated dysplasia.”

Read our Peer-Reviewed Conference Report article on this subject: Predicting relapse in ulcerative colitis with AI-assisted endoscopy

1. Maeda Y, et al. A novel artificial intelligence-assisted Image Enhanced Endoscopy assesses accurately "vascular-healing" and predicts long-term clinical relapse in patients with ulcerative colitis: a prospective cohort study. OP16, 19^th Congress of ECCO, 21–24 February 2024, Stockholm, Sweden.

 

Copyright ©2024 Medicom Medical Publishers



Posted on 23 February 202419 April 2024