STOCKHOLM, Sweden — An artificial intelligence (AI)–assisted image-enhanced endoscopy system accurately assessed vascular healing and predicted long-term clinical relapse in patients with ulcerative colitis (UC), according to data from the study of a novel investigational tool.
Clinical relapse was predicted in 3% of patients identified as having vascular healing in all segments compared with 23.9% in those with vascular activity (ie, one or more segments were active), reported Yasuharu Maeda, MD, gastroenterologist from Showa University Northern Yokohama Hospital, Digestive Disease Center, Yokohama, Japan.
In patients with a Mayo Endoscopic Score (MES) ≤ 1, the clinical relapse rate was 3% and 18.6% in the vascular healing and vascular active groups, respectively, he said.
Endoscopic remission is a crucial treat-to-target goal in patients with UC, and image-enhanced endoscopy is spreading in routine practice as a way to detect inflammation and to predict outcomes, Maeda said.
“Image-enhanced vascular findings lead to a stronger correlation with histological activities and long-term prognosis compared with white light endoscopy assessment,” he explained. “It also means that assessment can be done on-site without biopsy, pathologist effort, and associated costs; however, specialist training is required to achieve a high accuracy in outputs.”
Maeda presented the data (Abstract OP16) at the 19th Congress of the European Crohn’s and Colitis Organisation (ECCO).
Stratifying the Relapse Risk
Maeda and colleagues developed a novel AI-based narrow-band imaging system, training it by using 8853 images from 167 patients with UC.
The AI system, EndoBRAIN-UC (Cybernet System Corp, Tokyo), is in use and currently adapted for only one endoscope, the Endocyto CFH290EC (Olympus EMEA, Tokyo), but for the purpose of this study, it was trained on images from five different scopes.
“By combining narrow-band imaging and AI, we developed a system where we can differentiate between vascular activity and vascular healing. This allows us to predict relapse,” Maeda said.
In an open-label, prospective cohort study, they tested the system with the aim of assessing the efficacy of AI-identified vascular healing to stratify the relapse risk in 100 patients showing clinical remission of UC (ie, partial MES ≤ 1).
Patient characteristics were similar between both groups with an average disease duration of 10 years.
In the vascular healing group (n=33), the average age was 52 years, 20% were men, 58% had extensive colitis, and 52% had a MES score of 0.
In the vascular active group (n=67), the average age was 56 years, 32% were men, 61% had extensive colitis, and 25% had a MES score of 0.
Colonoscopy was performed using the AI system to identify mucosa as healing or active for six colorectal segments of each patient. The MES and histologic assessment for these segments were also recorded.Patients were then followed for up to 12 months and assessed for clinical relapse.
The clinical relapse rate was higher in the vascular active group than in the vascular healing group as identified by AI.
“We only evaluated the diagnostic output of the AI but obtained white light endoscopies and biopsies for contrast studies,” Maeda noted.
They also looked at whether the endoscopist’s level of experience (ie, trainee or expert) was important but found that clinical relapse predictive values were independent of the endoscopist’s experience.
Still in the Early Stages
AI-assisted colonoscopy work is still at an early stage , said session co-moderator, Monika Ferlitsch, MD, head of Internal Medicine Department II, gastroenterology and hepatology, Evangelical Hospital, in Vienna, Austria.
We now have initial results, but “I suspect it will take 10-20 years for implementation into routine clinical practice,” she said.
The best outcome for our patients is to be able to predict response to therapy and recurrence rates, “and we see this is possible now with AI. But of course, we need more clinical data to support it,” Ferlitsch said.
Maeda and Ferlitsch have declared no financial disclosures.
>>> Read full article>>>
Copyright for syndicated content belongs to the linked Source : Medscape – https://www.medscape.com/viewarticle/ai-identified-vascular-healing-can-predict-clinical-relapse-2024a10003rm
