(a, d) and (b, e) for all the 6250 rater images of the (a, b) 2D and (d, e) 3D testing sets as a function of . A linear fit was performed for both scatter plot types (black) showing the quasilinear relationship between and both and . (c, f) plotted as a function of . We see that the scatter plots also follow a quasilinear trend. This graph demonstrates that, compared to a point with given and , a neighbor point with a higher (worse) can still give a higher (better) or similar , especially for high , questioning the validity of as a performance measure for label fusion.
