Figure 2

Workflow of SpineS. Z-stacks (M \(\times\) N \(\times\) Z) from multiple time points (T) are analyzed. Each z-stack is imported and registered. MIPs are then computed and filtered using median filtering. Dendritic spine heads are detected using a CNN and segmented using a watershed segmentation algorithm followed by a graph-based clustering. Each spine volume is estimated using the integrated fluorescence intensity (IFI) of the spine head within the segmentation region. The intensities in the region are normalized using the median fluorescence intensity of the dendrite at the corresponding time point after background fluorescence subtraction. Neck paths from the spine head center to the closest geodesic point on the dendrite are computed using the fast-marching algorithm by imposing empirically set constraints.