An automatic severity grading system using confocal laser endomicroscopy to evaluate inflammatory activity of...

In the treatment of ulcerative colitis (UC), persistent inflammatory activity increases the risk of recurrence and poor prognosis. Probe-based confocal laser endomicroscopy (pCLE) can accurately assess the UC inflammation than conventional endoscopy. The aim of this study was to develop and evaluate an automatic severity grading system (ASGS) for assessing the inflammatory activity of UC using pCLE. First, we developed ASGS based on a convolutional neural network (CNN) for grading UC inflammation activity using pCLE images retrospectively. Next, we prospectively recruited UC patients for image and video validations of ASGS. The remission patients were then followed up for at least 12 months to evaluate relapse according clinical disease activity score. In the prospective image testing set, the sensitivities of ASGS for predicting Grades A, B, C, and D of UC inflammatory activity were 97.38%, 84.56%, 86.22%, and 97.74%; the specificities were 99.25%, 99.36%, 99.90%, and 97.60%. In the video testing set, the sensitivities of ASGS for predicting Grades A, B, C, and D of UC inflammatory activity were 75.56%, 94.00%, 92.31%, and 95.24%; the specificities were 98.17%, 93.25%, 98.00%, and 98.10%. The ASGS showed a good correlation with pathology, the quadratic weighted kappa (QWK) was 0.780. During follow-up, endomicroscopy score serve as an superior parameter to predict the relapse of UC patient than white-light endoscopy (WLE). ASGS enables fully automated grading of the inflammatory activity of UC, which offers the possibility to predict the remission and prognosis of UC patients.