Comparison with an arbitrary sampling

The set of measures (9 images) is arbitrarily split into two sub-groups, containing 960 and 976 fibers respectively. Figure shows the fiber distributions found for each sub-group with the manual and automatic procedures.

  

Figure 4.11: Comparison of fiber distributions for automatic ( above) and manual ( below) measures for data sets 1 ( left) and 2 ( right) ( $bin = 0.5 \mu m$)

The two peaks in the 4 histograms are at the same place (around $5 \mu m$ and $11 \mu$) and their pattern is equivalently biphasic. However, histogram shape is different for the two groups and the $\chi^2$ statistical test, performed on manual normalized histograms, shows significant differences for them ( p < 0.0001). In contrast, the automatic histograms correspond very closely to their manual counterparts (p > 0.01 and p > 0.05).
This last test shows that, even if it does introduce some bias, the automatic procedure is much less detrimental than the arbitrary choice of a limited number of images in the nerve cross-section, a critical decision in the manual morphometric study.