ECCO21: Moderate accuracy of AI in detecting different forms of IBD

Research presented at the European Crohn's and Colitis Organisation (ECCO) congress suggests that use of an artificial neural network (ANN) showed "moderate accuracy" in recognising different forms of inflammatory bowel disease (IBD). "Enlarged sample size and inclusion of morphologic images are needed in order to improve the accuracy of ANN," the authors said.

Investigators explained that "it usually takes more than one year to diagnose ulcerative colitis (UC) and more than two years for Crohn's disease (CD)," due to the lack of a gold standard and clinical similarities, so they set out to optimise existing methods in diagnosis with the help of an ANN.

To develop the ANN, 163 patients with UC, 53 patients with CD of the large bowel and 34 patients without endoscopic findings underwent colonoscopy, during which 856 images were obtained, including images of mild, moderate and severe IBD activity. All images were captured in high-resolution with the same Olympus endoscope. The ANN consisted of a multilayer perceptron model that determined a presence of pathology and a convolutional neural network model that differentiated UC and CD.

77% accuracy for pathology differentiation

The endoscopic images were downloaded to the ANN, which was trained three times due to the decrease of its accuracy. The model in differentiation of pathology showed an average accuracy of 77%. The result "could be associated with an insufficient number of CD images or ANN training on continuous data set of images with UC and CD features without clustering them to particular patients," the authors said. They noted that a limitation of the study was that all the images were taken with the same endoscope.

To ensure you don't miss other Top Stories like these and news on key healthtech industry developments, sign up for our free daily e-newsletter here.  

Did you like this article?