Research: Spine Detection and Classification

A fundamental challenge in understanding how dendritic spine morphology controls learning and memory has been quantifying three-dimensional (3D) spine shapes with sufficient precision to distinguish morphologic types, and sufficient throughput for robust statistical analysis. The necessity to analyze large volumetric data sets accurately, efficiently, and in true 3D has been a major bottleneck in deriving reliable relationships between altered neuronal function and changes in spine morphology.

We introduce a novel computational approach for detection and shape analysis of neuronal dendritic spines from confocal and multiphoton laser scanning microscopy (CLSM and MPLSM) images, that operates fully in 3D, and is faster and more accurate than existing semi-automated technologies. The algorithm is a module of our NeuronStudio software application, an integrated system for semi-automated digitization, morphometry and analysis of complex neuronal morphology at high resolution.

Image Segmentation and Voxel Processing

Figure 1: Individual voxels in successive layers of different colors. A) Voxels of first layer shown in red. B) Fourth iteration of the cluster-building produces the green layer. C) In the last iteration the green layer floods into the dendrite. D) Schematic of cluster layer-building.

The spine analysis module utilizes a previously computed model of the dendritic tree which is compatible with many existing neuronal morphometry applications. At each node along the dendritic model we define a cubic section of data centered at the node, that is then thresholded. Once each node has been assigned a threshold value, any voxel in the dataset may be segmented by linearly interpolating a local threshold value between nodes along the closest dendritic segment.

Spine Detection Using Voxel Clustering

Individual spines are detected by clustering candidate spine voxels, starting from the tips and moving towards the dendrite. The cluster-building algorithm can be described as an iterative 3D flood-fill of the structure. Each iteration builds a layer that limits cluster growth towards the dendritic segment but does not constrain sideways growth. If at any point during layer-building the spread exceeds a user-provided maximum spine width, the cluster building stops.

Calculating Spine Profiles

Figure 2: A) XY view showing the rays of a 2D Rayburst run at the center of mass of a layer. B) Side profile of the Rayburst core, demonstrating how the rays extend in the XY plane only. C) Bar graphs showing the blue Rayburst diameters calculated at each layer for three representative spines of type, mushroom, thin and stubby. These profiles are used to classify spine types. HD: head diameter; ND: neck diameter.

For each layer built in a particular cluster, we maintain a profile of measures for later use by the shape classification routine. These measures include the spread of the layer, the Rayburst diameter, and the depth of the layer.

Spine Shape Classification

After clustering, spine shapes are classified into three types, mushroom, stubby and thin, using the profile of 2D Rayburst diameters computed in consecutive layers along the length of the spine. Use of 2D Rayburst within each layer avoids the effects of residual optical smearing in the Z direction