post-hoc soma detection

[huangziwei]

The default parameters in sk.skeletonize() are optimized for bigger cells (SACs, GCs), so they might not always be optimal for smaller ones (such as BCs). Most often, the soma detection fails because the potential soma node radii is not significantly larger than the neurites (we set the percentile (soma_radius_percentile_threshold) to 99.9, which is quite extreme, but works very well for large cells):

in this case, we might want to just re-estimate the soma instead of rerunning skeletonize() again.

skel_new_soma = skel.detect_soma(
                    soma_radius_percentile_threshold=99,
                    soma_radius_distance_factor=2,
                )
# skel_new_soma.prune(num_nodes=2) # by default we already prune the 1-node twigs but there might still be other longer artifact twigs, so prune it as you please