The study used 862 endoscopic images taken between 2003 and 2021 to train visual imagery AI tool Efficient-Net-B3 to classify lesions that signal IBD.
After training, the AI tool was able to make diagnoses with 64.5 percent sensitivity, 89.5 percent specificity and 80.6 percent accuracy.
Its correct diagnosis rate was 79 percent, while the average endoscopist’s rate was 77.8 percent.
