A study published in Gastroenterology evaluated the use of a machine learning tool to evaluate endoscopic disease activity in patients receiving treatment for ulcerative colitis.
Data was collected from an international phase 2 trial of 249 patients. In total, 795 full-length endoscopy videos were prospectively collected, comprising 19.5 million image frames. Expert readers assigned each video an endoscopic Mayo score (eMS) and an Ulcerative Colitis Endoscopic Index of Severity (UCEIS) scores. The researchers constructed a recurrent neural network based on these scores. The primary outcome was quadratic weighted kappa (QWK) comparing the alignment between machine-read scores and reader-assigned.
The researchers reported “excellent” agreement between the neural network and human-graded scores, with a QWK of 0.844 (95% confidence interval, 0.787-0.901) for eMS and 0.855 (95% confidence interval, 0.80-0.91) for UCEIS.
Via: Gastroenterology https://linkinghub.elsevier.com/retrieve/pii/S0016508520352835
