Fig. 4.

Visualization of parameter space for image analysis. (a) The overview shows sampled parameter values as a clustering hierarchy. (b) The refinement view shows scaled previews of the image-based outcomes for selected subtrees in the clustering hierarchy. (c) The reference image view superimposes the output that currently has the focus in the refinement view on a reference image. (d) Selected subtrees are highlighted in blue in the overview and (e) shown in distinct regions in the refinement view. (f) Users can specify a level in the clustering tree at which to position subtrees side-by-side in the refinement view. Further interactive features include: (g) brushing; tagging of (h) high-quality and (i) low-quality outcomes; and (j) filtering. The data set shown here was generated by sampling the parameter space of the IdentifyPrimaryObjects module in CellProfiler for human HT29 cells (5 parameters, 4 samples each, yielding 1,024 outcomes). At this point the user has identified and tagged a number of high-quality outcomes where all cells have been detected (coded in green in the overview and detail view). The user has also tagged low-quality outcomes where all cells have not been detected (magenta). The distribution of green and magenta in the overview indicates which parts of parameter space to consider and which parts to disregard.