Figure 5.

IRR confounding and sensitivity. (a) Estimated coefficients (and 95% confidence interval) for the variables in a linear model that predicts IRR metrics. The variables were normalized to a sensible scale for a fair interpretation of the effect sizes. The coefficients for the regions quantify the difference in intercept from that of the default region (legs). (b) Leave-one-rater out sensitivity analysis of the average pixel-level ICC measure. Avg, average; ICC, intraclass correlation coefficient; IRR, inter-rater reliability; KA, Krippendorff’s alpha; min, minute.