Automated Three-Dimensional Detection and Shape Classification of Dendritic Spines from Fluorescence Microscopy Images (original) (raw)
Figure 3
Cluster Declumping and Spine Stem Reattachment.
A) Volumetric data showing cluster of three spines that appear merged due to limited image resolution. B) Result of clustering algorithm using the cluster declumping routine described in the text. The merged spines have been properly detected as separate clusters indicated by differently colored voxels. C) Local 3D gradient vectors used to declump two merged spines. Vector heads are colored red or blue; black tails point in direction of increasing intensity. At the valley marked by the white arrow, the gradients reverse direction: red gradients point leftward, toward the center of mass of the red spine; blue gradients point rightward, toward the center of mass of the blue spine. D) Volumetric data showing an apparently detached spine head, and its spine stem attached to the dendrite. Because of poor resolution of the spine neck, this thin spine is detected as two separate spines. E) Bell-shaped region (transparent red) used to detect the stem of a spine whose neck cannot be adequately resolved. The voxels of the attached (red) and detached (yellow) spines are represented as cubes within a volume rendering of the dataset. Because the tip of the attached spine falls within the bell-shaped region, the algorithm merges these two clusters into a single spine. White line between yellow detached and red attached spines represents an approximate scale of 0.8 µm. F) Schematic showing parameters of the spine stem reattachment routine. p0: lowest-DTS voxel on a detached spine; p1: closest point to surface of the dendritic model: m: maximum point on an attached spine within the bell-shaped domain shown in transparent red; pm: projection of m onto line segment [p0, _p1_]; Dmax:: limit of the bell-shaped domain that encloses the spine stem.