2.8 Image segmentation & Structural network analysis
During the culture lifespan, the neurons grow new connections and
self-structure themselves into a complex network, whose structural
evolution is the objective of the study. Therefore, from the large-scale
photographs acquired during the monitoring days, we extracted the
connection scheme and mapped it into a graph using an image segmentation
algorithm coded in MatLab to identify both the neurons and their
connections. The whole procedure is illustrated in Figure 2. Panels A1-3
show the growth evolution of a small part of the whole culture between
day 1 and 15. Panels B1-3 show the result of the image segmentation
algorithm after detecting neurons (and their aggregates) in red colour
while the segmented neurites are marked in green. This information is
used to construct a connectivity graph (panels C1-3) where nodes (green
circles) are single neurons or aggregates and links (blue connecting
lines) are either direct paths between nodes or through a branching
path.
Our aim is to compare the evolution of CNNs cultured on a chip to those
cultured on Petri dishes as previously reported in Refs.. The network
evolution involves the creation of new links connecting nodes over time.
We monitor the progress of several commonly studied graph parameters
such as the average clustering coefficient (C), which accounts for how
densely connected is each local vicinity of the network, and the
shortest path length (L), computed in the largest connected component,
that quantifies how far are the nodes in the topological sense. These
two last parameters are used as an indicator of the balance between the
local and long-distance connectivity in the network.