Spatial image variability analysis

In our analysis, each image is partitioned into a number of non-overlapping, spatial regions. Each spatial region is viewed as a Region of Interest (ROI). A grade is assigned to each ROI from a set of independent raters. The model is applied in lung image analysis, where a grade of 0/1 is assigned to each ROI. Here, I represents success, while 0 represents failure to detect a hypothesized pattern in the region. Methods for establishing spatial symmetry and spatial growth analysis are presented. In addition, a Bayes and a novel summation classifier are used for classifying the entire lung based on their regional grades. When compared against the readers, the two classifiers have a total misclassification error of the order of inter-rater variability error.