Inter-Grader Reliability of a Supervised Pterygium Redness Grading System

This research work outlines the methodology of a medical image grading system based on supervised learning algorithm. A total of 210 features were extracted in various color spaces and most relevant features were identified and fed into a regularized feedforward neural network. The inter-grader reliability of the supervised system was then assessed based on the manual delineation of region of interest by 2 human graders. Intra-class correlation analysis of the experiments shows excellent agreement of 0.869 to 0.954.