Neuronstudio

NeuronStudio is a non-commercial program created at Mt. Sinai School of Medicine by the Computational Neurobiology and Imaging Center. This program is designed to allow automatic tracing and reconstruction of neuron structures from confocal image stacks. Using a point and click interface you can load a confocal dataset in TIFF format and select and trace dendritic structures. The resulting model can then be exported to a file using standard formats for further processing, modeling, or for statistical analyses.

The system is capable of handling morphologic details on scales spanning local spine geometry through complex tree topology to the gross spatial arrangement of multi-neuron networks. NeuronStudio provides a much-needed system for automated digitization that avoids the subjective errors that arise during manual tracing.

Amongst the main features you will encounter in this version are the following:

• Support for 8 and 16 Bit TIFF files
• Automatic file name sequence detection
• Tree building with dynamic threshold selection
• Accurate 2D and 3D measurements using the #Rayburst Algorithm
• Dendritic spine detection and reporting
• 2D and 3D data visualization
• Integrated 3D image filters
• 3D length and width measuring tool
• 64-bit version available for large datasets

Digitized trees and their branch diameters are output in SWC file format for easy importing to other morphometric software, and to compartment modeling programs such as NEURON and GENESIS.

Additional recommended knowledge

How to ensure accurate weighing results every day?

How to quickly check pipettes?

Daily Sensitivity Test

Deconvolution

Deconvolution of imaged data is essential for accurate 3D reconstructions. A blind deconvolution system uses an estimated and iteratively refined theoretical point spread function(PSF) that can be adjusted locally to account for variations in refractive characteristics of the tissue with depth and sample characteristics. For automated use with large, tiled tissue blocks, this is faster and more accurate then using an experimentally determined PSF.

Skeletonization and Diameter Estimation

Quantization errors arise in standard skeletonization algorithms from the integer nature of digital images. The requirement for accurate representation of fine dendritic geometry has required the development of novel adaptations of standard skeletonization and diameter estimation algorithms to correct for these quantization errors. Iterative thinning skeletonization methods can provide a distance in voxels from each tree node to the surface of the object. This distance is the D6 metric, obtained by counting the number of voxels as they are removed in the minimal 6-connected path from the surface to the medial axis. In existing skeletonization or vectorization algorithms for dendritic morphometry, the branch cross-section at any node is approximated as circular, with the D6 metric providing the single diameter estimate. The precision of this diameter estimate is limited to the physical size of the voxels. For small structures such as thin dendrites and spines, comprising only a few voxels even at maximal imaging resolution, the error can be significant if this measure is used directly (see figure). To minimize quantization error and evaluate more precisely the geometry of the nodes, a new estimation technique exists, the Rayburst Sampling Algorithm that uses the original grayscale data rather than the segmented images for precise, continuous radius estimation, and multidirectional radius sampling to more accurately represent non-circular branch cross-sections and non-spherical spine heads.

Rayburst Algorithm

  The Rayburst Sampling Algorithm uses the original grayscale data rather than the segmented images for precise, continuous radius estimation, and multidirectional radius sampling to more accurately represent non-circular branch cross-sections and non-spherical spine heads. The algorithm precomputes an array of unit vectors which sample the data in multiple directions, (the Sampling Core) from which an estimate of the node's geometry is computed. Accurate representation of each direction by the sampling core requires that the N vectors should be uniformly spaced over the unit sphere. The algorithm uses a particle physics simulation in which a set N of randomly oriented unit vectors is generated, resulting in a random, nonuniform distribution of points on the sphere. Each particle then receives a repulsive force from every other particle, proportional to the inverse square of the distance between them. By iteratively displacing the particle in the direction of the resultant forces, the particles rearrange themselves. This system will tend to a stable, minimum energy configuration within approximately 40 iterations, where each particle is maximally separated from its closest neighbors. The information can be used to reconstruct 3D branches of arbitrarily irregular shapes. The diameter of an equivalent circular cross-section is computed in the plane normal to the medial axis using the Median Lower Band Diameter(MLBD). To compute the MLBD, take the set of samples and add the corresponding pairs of opposite vectors. Sort the vectors by size, define the lower band as the lower 50%, and use the distance at position N/4, representing the median of the lower band, to estimate the diameter.

See also

• Confocal microscopy
• Voxel
• Point Spread Function
• Deconvolution
• TIFF
• Skeletonization

References