Neuronal morphometry directly from bitmap images · number of primary branches, ... Nature Methods:...

Supplementary Figure 1 | Overview of the Sholl Analysis software.

(a) User interface, version 3.4.1. Sub-modules noted in blue. (b) Maximum intensity projection of a Drosophila class IV dendritic arborization sensory neuron (ddaC) labeled by the ppk1.9-GAL4-driven reporter UAS-mCD8::GFP, a sample image distributed with the plug-in. The annulus formed by Starting radius (the first sampled distance) and Critical radius is indicated. Outer arc depicts Enclosing radius, the distance to the most distal dendritic tip. (c) Semi-log plot of the cell depicted in (b), where the number of intersections was normalized to the area of each sampled shell. Two regression lines are shown to demonstrate that curves can be fitted to all data points, or to a subset restricted to the 10th to 90th percentiles of the data (shaded in gray). R2 is the coefficient of determination, and k the Sholl regression coefficient. (d) Linear Sholl plot of the cell in (b). Key metrics include Critical value (Nm), Critical radius (rc) and Mean value (Nav). Differences between sampled and fitted maxima are shaded in gray. The centroid of the sampled profile is marked (×). Schoenen ramification index (RI) is the ratio between number of branches at the maximum and the number of primary branches, using either sampled data or the fitted Nm. The number of primary branches can be entered manually, or drawn from the number of intersections at Starting radius.

Nature Methods: doi:10.1038/nmeth.3125

Supplementary Figure 2 | Accuracy of bitmap image-based Sholl Analysis.

(a) Maximum intensity projection of an Alexa 594 dye-filled layer-5 pyramidal neuron of juvenile mouse visual cortex1. Arrowheads highlight the apical tuft (top) and the soma (bottom). (b) Manually reconstructed dendrites (blue) and axon (magenta) of the neuron in (a). (c) Linear Sholl plots from bitmap images (dots), following either manual segmentation (“user segm.”) or automated segmentation of the image stack. For comparison, results from reconstruction-based analysis are plotted for the axon (dashed magenta line) and dendrites (solid blue line). The plot of the left corresponds to the basal region populated by both axonal and dendritic arbors, while the plot on the right corresponds to apical dendrites. (d) Tukey mean-difference plot to examine the agreement between the bitmap approach and traced reconstruction approach for three different neurons (numbered and color-coded). Each data point represents the count of intersections at a particular distance from the apical bifurcation: pairwise differences between the two approaches at each distance are plotted against each mean. The 95% limits of agreement for individual cells are shown to the right, as is the average for all three cells and the average bias (dotted lines).

1. Buchanan, K.A. et al. Target-specific expression of presynaptic NMDA receptors in neocortical microcircuits. Neuron 75, 451-466 (2012).


Supplementary Figure 3 | Resilience of bitmap image-based morphometry to image degradation over a range of synthetic noise.

(a) Maximum intensity projection of an axonal arbor of a Drosophila olfactory projection neuron from the DIADEM Challenge dataset (OP_1)1,2. We contaminated the original stack (voxel size: 0.33×0.33×1.0µm) with Poisson noise, using increasing multiples of the stack standard deviation (σ = 9.55) as the probability mass function of the Poisson random variable. This noise was either added (+) or subtracted (−), over a range from −18σ to +18σ. Images are shown with the coefficient of determination below each image (R2, mean ± s.d.) to indicate the degree of similarity with the original; R2= 1 corresponds to identical images. Arrowheads indicate the Sholl analysis center (see b). (b) Each graph represents one of 7 metrics (3 from sampled data, 4 from fitted data) that were calculated from Sholl plots generated directly from bitmap images with Sholl Analysis (“Bitmap”), or from their respective reconstructions traced with Neural Circuit Tracer37 (“Reconstruction”). The parameters and routines used to retrieve the data were fixed (See Supplementary Methods and Supplementary Software for details). Shaded areas in each graph (light gray) indicate the range of noise (−14σ to +8σ) over which metrics were largely consistent with those calculated from the original image (zero noise, dark grey vertical bar). Dashed lines indicate two references calculated at zero noise: one from a user-derived manual segmentation of the stack (“Bitmap reference”, blue) and one from the DIADEM gold standard reconstruction (red). For each metric, the concordance correlation coefficient (ρc)3 between the bitmap approach and reconstruction approach are shown.

1. Brown, K.M. et al. The DIADEM data sets: representative light microscopy images of neuronal morphology to advance automation of digital reconstructions. Neuroinformatics 9, 143-157 (2011).

2. Jefferis, G.S. et al. Comprehensive maps of Drosophila higher olfactory centers: spatially segregated fruit and pheromone representation. Cell 128, 1187-1203 (2007).

3. Lin, L.I. A concordance correlation coefficient to evaluate reproducibility. Biometrics 45, 255-268 (1989).


Supplementary Figure 4 | Sholl-based metrics of Type-1 and Type-2 PV interneurons.

Metrics loading on the first principal component (71.6% of observed variation), used as clustering variable in Figure 1c: (a) Sholl regression coefficient, (b) Sum of intersections, (c) Distance associated with at least two intersections (a modified Enclosing radius), (d) Centroid radius, (e) Centroid value, (f) Critical value, (g) Mean value, (h) Critical radius. Values enclosed by brackets depict factor loadings. Because pipette fluorescence could not be entirely eliminated near the soma, the number of primary branches and Schoenen ramification indices were excluded from the analysis.


Supplementary Note

Overview of the Sholl Technique The Sholl technique describes the branching patterns of dendrites and axons by plotting the number of branches as a function of the distance from a defined location, usually the cell soma. In his original publication1, Sholl achieved this by superimposing a set of concentric circles onto camera lucida tracings of stained neurons, and by counting the number of branches intersecting each circle. From plotted profiles of this data, he derived metrics that he used to classify neurons from different regions of cat neocortex. This way, Sholl was the first to quantify the anatomy of individual labeled cells2. Though his groundbreaking technique ignores certain features such as dendritic taper3, it provides useful morphological metrics to characterize neuronal arbors. It is used extensively to quantify the effects of genetic and pharmacological manipulations on neurons, and has also been successfully applied to the study of retinal vasculature4,5, angiogenic sprouting6,7, and mammary ducts8, where the center of analysis was the optic nerve, a microcarrier bead, or the nipple, respectively. Despite this proven utility in cellular morphometry, previous methods to apply the Sholl technique have required reconstruction9-12, or were limited to two-dimensional (2D) images of 3D arbors13.

Implementation Sholl Analysis is an open-source plug-in for ImageJ14 programmed in Java and bundled in Fiji15, a distribution of ImageJ focused on biological image analysis. The plug-in performs the Sholl technique directly on 2D and 3D images of fluorescence-labeled cells. It employs an improved algorithm to retrieve data from pixel-based connectivity; while in development its algorithm has been employed for analysis of 2D binary images in studies of dendrite morphogenesis16-20 and angiogenesis4. The plug-in features a straightforward interface (Supplementary Fig. 1a), and is accompanied by a detailed user guide (http://fiji.sc/Sholl). It is also complemented by a Fiji update site (http://fiji.sc/BAR) providing auxiliary tools for data analysis and image pre-processing (Supplementary Methods). As part of the ImageJ framework, Sholl Analysis can perform batch processing and analyze any grayscale image supported by the Bio-Formats library21. Sholl Analysis was designed with flexibility in options for data sampling, normalization and curve fitting, and it is scriptable for automated execution of subroutines. With these features it calculates key Sholl metrics (see below) while accommodating a wide diversity of arbor morphologies (Supplementary Fig. 1a). It allows continuous sampling at every distance from the center of analysis, or interval-based sampling with integration of repeated measures. Users can also confine the analysis to specified sub-regions of an image. For users without access to imagery of cells labeled in spatial or spectral isolation, the plug-in can also analyze raw data from Sholl profiles obtained from Simple Neurite Tracer11, software for semi-automated reconstruction that is also bundled in Fiji.

Requirements Sholl Analysis requires only that neurons be labeled or resolved in isolation (spatially or spectrally, as with Brainbow-labelling). The resilience of Sholl Analysis to simulated shot


noise (Supplementary Methods) suggests it can accommodate a range of different imaging systems for fluorescence microscopy. However, as detailed in the Pre-processing section of the user guide (http://fiji.sc/Sholl), users first ought to mitigate labeling artifacts and sources of image degradation that can affect segmentation, such as uneven illumination along the Z-axis. Acquaintance with ImageJ and experience with fluorescence microscopy are recommended to successfully use the software.

Normalization, Sholl Plots and Metrics The quantitative description of a Sholl profile obtained from the initial count of intersections may require normalization of values against a chosen property of the sampling shell such as its perimeter or area (in 2D), or its surface area or volume (in 3D)22 (Supplementary Fig. 1a).

Sholl himself originally described three types of plots1:

• Semi-log, i.e., log(normalized intersections) vs. distance • Log–log, i.e., log(normalized intersections) vs. log(distance) • Linear, i.e., intersections vs. distance

The common goal of the two logarithmic plots (semi-log and log-log) is to interpolate a straight line from the normalized data points using simple linear regression. The choice of which of the two logarithmic plots is most appropriate depends upon arbor complexity22, which can be quite diverse among different types of neurons. However, for a particular type of neuron, one of the two logarithmic plots is expected to produce an unambiguous linear regression. In Sholl Analysis, results for both are obtained concurrently, and their linear regressions can be directly compared. In addition, users can define whether linear regression is fitted to all data points, or only to a subset (Supplementary Fig. 1c). The slope of the linear regression (when multiplied by −1) is known as the:

• Sholl regression coefficient1, or Sholl decay. The rate of decay of branch density with distance from the center of analysis. Abbreviation: k

In linear Sholl plots, polynomial regression can be used to mitigate the influence of local variations in the sampled data23, and thereby better describe the relationship between intersections vs. distance. Linear Sholl plots of neurons having different arborization patterns might be best fitted to polynomials of different orders23, and so we have included within the Sholl Analysis plug-in a heuristic algorithm that predicts the polynomial of best approximation for each analyzed arbor (Supplementary Fig. 1d).

Three key metrics can be derived from the polynomial function23 (Supplementary Fig. 1d):

• Critical value. The maximum of the polynomial function, an indication of maximal branching. Abbreviation: Nm

• Critical radius. The distance at which the Critical value occurs. Abbreviation: rc

• Mean value of the polynomial function. The average number of intersections between the Starting radius and Enclosing radius. Abbreviation: Nav


Sholl Analysis also calculates additional metrics from linear plots (Supplementary Fig. 1d), including descriptive statistics of sampled data, its geometric center of mass, as well as the:

• Shoenen ramification index. The ratio between the maximum of the profile (either the sampled maximum or the fitted Nm) and the number of primary root branches emanating from the center of analysis24. Abbreviation: RI

Supplementary Methods

Comparison of Sholl profiles from bitmap images and manually reconstructed arbors To assess the accuracy of Sholl Analysis in counting intersections directly from images of neurons, we examined cortical pyramidal neurons that had been labeled in acute slices of mouse visual cortex by Alexa-594 dye during whole-cell recording25. Pyramidal cells are characterized by two distinct dendritic arbors: a single apical dendrite extends from the cell apex towards the pial surface, while several basal dendrites arborize from the base of the soma26, 27. Their axon projects away from the pial surface and has an arbor of collaterals that approximately overlaps with the basal dendrites. Therefore, bitmap images of dye-filled pyramidal cells can be analyzed in two non-overlapping regions: an apical tuft populated only by apical dendrites, and a basal region populated by both axonal and dendritic arbors (Supplementary Fig. 2a,b). We obtained profiles for the apical and basal regions independently, capitalizing on the plug-in’s built-in option to do so (Supplementary Fig. 2c). Using the main apical bifurcation as the center of analysis for apical dendrites, we compared Sholl Analysis to a conventional approach where arbors were manually reconstructed with the Neuromantic freeware28 and then subjected to the Sholl technique using Simple Neurite Tracer11 (see Image Processing below). Using three different pyramidal neurons, we made pairwise comparisons of the two approaches at every sampled distance. We calculated the difference in the number of intersections counted by the two methods, and then plotted this difference against their respective average in Tukey mean-difference plots (Supplementary Fig. 2d).

In the apical region, there was good agreement between the Sholl profiles obtained using automated bitmap analysis versus manually reconstructed dendrites (example in Supplementary Fig. 2c, right panel). For each of the three neurons examined, variability between the two approaches was distributed evenly across the range of data (Supplementary Fig. 2d). The limits of agreement for all three cells combined (calculated as the mean of differences between the two methods ±1.96✕ standard deviation29) fell between −3.2 and +3.9 intersections. From this distribution we assessed bias, the mean value of the difference between the two approaches, and found that the bitmap approach over-sampled by an average of 0.34 intersections (1.5%). Taken together, these results suggest good agreement between our bitmap image-based approach and the traditional approach using manual reconstructions. Others have assessed the accuracy of an outdated version of our software to measure dendrites of retinal ganglion cells labeled by diOlistics in 2D, and deemed it unfavorable compared to traditional methods requiring reconstruction30. We caution users that bitmap accuracy will necessarily deteriorate with discontinuous labeling of neuronal processes, or with poorly-segmented images that do not reflect the complete topology of the cell. The Sholl Analysis plug-in allows users to


visually inspect for segmentation artifacts in color-coded masks that overlay sampled data upon the segmented arbor. Where images are unavoidably sub-optimal with respect to signal-to-noise, Fiji features specialized workflows that could improve segmentation31.

In the absence of subcellular labelling, the automated bitmap approach cannot discriminate axonal from dendritic segments; in the basal region therefore, our approach combined data from both segment types into a single Sholl profile (Supplementary Fig. 2c, left panel). However, thin cortical axons fluoresce less intensely than dendrites, so we minimized the contribution of axonal segments with a straightforward routine to limit the analysis to pixels above a threshold of fluorescence intensity (see Image Processing below). This led to a profile matching that of manually reconstructed dendrites alone (Supplementary Fig. 2c, left panel), showing that such routines can provide a degree of specificity for dendrites in mixed axonal/dendritic datasets.

Resilience to image degradation Shot noise is one source of random image intensity fluctuations in confocal microscopy32. To simulate this form of image degradation and assess its impact on bitmap Sholl Analysis, we contaminated the image of a neuronal arbor by adding or subtracting Poisson noise in incremental amounts (Supplementary Fig. 3a), with the image chosen from the DIADEM Challenge dataset33. This image depicts an axonal arbor of a Drosophila olfactory neuron labeled by the MARCM technique34,35. DIADEM datasets and their respective manual reconstructions (the “gold standards”) have been used in several benchmark studies33, 36-41.

We used the Sholl Analysis plug-in to obtain metrics from noise-contaminated images automatically (denoted “Bitmap” in Supplementary Fig. 3b). Over the range of noise levels, distinct metrics were calculated directly from the sampled data (three metrics), or from fitted curves (four metrics). Our bitmap approach was largely resilient to noise over a broad range (Supplementary Fig. 3b), and was accurate relative to a reference (“Bitmap reference”) for which the non-contaminated image had been segmented manually (see Supplementary Methods and Supplementary Software).

Next, we compared our Sholl Analysis results to those obtained from automatic reconstructions of those same contaminated images using Neural Circuit Tracer, a software package implementing one of the finalist algorithms of the 2009–2010 DIADEM Challenge37. As a reference for this, we used the DIADEM gold standard reconstruction34. For each metric, we determined the concordance correlation coefficient (ρc) between our bitmap approach and reconstruction42. For each of the four metrics from fitted curves in particular (Mean value ρc: 0.861, Critical value ρc: 0.915, Critical radius ρc: 0.926, and Sholl regression coefficient ρc: 0.933), there was good agreement over a range of shot noise levels (Supplementary Fig. 3b).

Programming Sholl Analysis was programmed with Eclipse SDK 3.7.2 (Eclipse Foundation) and Fiji’s built-in Script Editor. The source code of Sholl Analysis is public43 and available through the plug-in’s internet documentation page (http://fiji.sc/Sholl).


Image Acquisition Image acquisition has been described previously for Drosophila sensory neurons16, and for neocortical pyramidal cells and PV interneurons25. Neocortical cells were visualized in P12–P20 acute mouse visual cortex slices using custom-built 2-photon imaging workstations (Scientifica UK) with ScanImage44 v3.7 running in MATLAB (MathWorks). Two-photon excitation of Alexa-594 loaded via the patch pipette was obtained with a Ti:Sa laser (Spectraphysics MaiTai or Coherent Chameleon) tuned to 800–820nm, and data was acquired with Hamamatsu R3896 bi-alkali detectors and National Instruments PCI-6110 digitization boards. All recordings were in neocortical layer 5, as determined by the presence of conspicuous large pyramidal somata with characteristic apical dendrites, as visualized by Dodt contrast. PV interneurons were targeted in slices from GFP-expressing G42 mice45 by tuning laser to 880–900nm and visualizing GFP fluorescence.

To produce Purkinje cell-specific expression of Brainbow, transgenic Pcp2-cre mice (Jackson Laboratory) were crossed to Brainbow 2.1 mice (line R, Jackson Laboratory). Five-month-old progeny were deeply anaesthetized (Ketamine/Xylazine, 80/5 mg/kg), and perfused with 4% paraformaldehyde. Sagittal brain slices were cut 24 hours later from the cerebellar vermis (100µm thick) using a Leica Vibratome 1000S, and mounted with Prolong Gold Antifade (Invitrogen). Images were collected with an Olympus confocal microscope with a 60✕ oil immersion objective (NA= 1.35), using the following laser wavelengths: 559nm, 515nm, and 440nm. Images (512✕512 pixels) were acquired using a z-axis step of 0.5µm and were deconvolved with AutoQuant software (Media Cybernetics).

Image Processing Image processing was performed in Fiji15. Image fields were stitched together46, and fluorescence signal from filling pipettes and dye spillage (PV cells) or fluorescence artifacts (Brainbow cells) was eliminated manually. Background was eliminated through 3D median filtering (typically radius= 2–3), and stacks were segmented using one of ImageJ’s built-in threshold methods. Due to diminished brightness, some image stacks were enhanced with the Tubeness plug-in (http://fiji.sc/Tubeness)11. The elimination of axons in mixed axonal/dendritic datasets (basal arbors of PC cells, Supplementary Fig. 2c) was performed manually, where a 3-hue lookup table was applied to the maximum intensity projection of each stack and the range of displayed intensities was adjusted by visual inspection. With background pixels and axonal/dendritic arbors segregated satisfactorily into different hues, the intensity value separating the two arbors was used to threshold the initial image stack. Remaining axonal processes were then eliminated through automated shape-based filtering using a routine from the complementary BAR update site (http://fiji.sc/BAR), which contains most of the image processing routines used in this study. Subsequently, the Sholl Analysis software was used with batch processing as described in the documentation page (http://fiji.sc/Sholl).

Reconstructions were performed using Neuromantic28 or, in the case of Brainbow tissue, with Neurolucida (MicroBrightField, Inc.). Neurolucida tracings were converted to the SWC format47 using L-Measure48. Reconstructions were imported into Simple Neurite Tracer11 using the appropriate coordinate offset so that the tracings would align optimally with the corresponding bitmap image. The Sholl technique for these reconstructions was performed using Simple Neurite Tracer11, using a manually chosen point as the center of analysis. The closest voxel to this point was chosen for subsequent bitmap Sholl Analysis. For


consistency, data was obtained using the minimum sampling distance allowed by the bitmap approach, i.e., the cube root of the product of the voxel dimensions.

Automated reconstructions of DIADEM projections were performed in Neural Circuit Tracer 2.037. Scripts and instructions on how to process the DIADEM OP_1 dataset are available as Supplementary Software.

Principal Component Analysis and Statistics To classify PV interneurons, we retrieved all the metrics calculated by the Sholl Analysis plug-in for each neuron and performed clustering analysis following Principal Component Analysis as described elsewhere49,50 (Supplementary Fig. 4). The first component (which accounted for 71.6% of the observed variation) was used as clustering variable. Hierarchical clustering was performed using Ward’s method and squared Euclidean distances as linkage metric. A 25% linkage cutoff, as normalized to the greatest separation in the data set, was used as a best-cut selection criterion for the number of found clusters50. Statistical analyses were performed in JMP 10.0 (SAS Institute) and Prism 5.0 (GraphPad Software).

Supplementary References 1. Sholl, D.A. Dendritic organization in the neurons of the visual and motor cortices of the cat. J Anat 87,

387-406 (1953). 2. Young, J.Z. D. Sholl. Journal of Anatomy 94, 564-565 (1960). 3. Rothnie, P., Kabaso, D., Hof, P.R., Henry, B.I. & Wearne, S.L. Functionally relevant measures of spatial

complexity in neuronal dendritic arbors. J Theor Biol 238, 505-526 (2006). 4. Strasser, G.A., Kaminker, J.S. & Tessier-Lavigne, M. Microarray analysis of retinal endothelial tip cells

identifies CXCR4 as a mediator of tip cell morphology and branching. Blood 115, 5102-5110 (2010). 5. Wujcik, J.M., Wang, G., Eastman, J.T. & Sidell, B.D. Morphometry of retinal vasculature in Antarctic

fishes is dependent upon the level of hemoglobin in circulation. The Journal of experimental biology 210, 815-824 (2007).

6. Pan, Q. et al. Neuropilin-1 binds to VEGF121 and regulates endothelial cell migration and sprouting. J Biol Chem 282, 24049-24056 (2007).

7. Pan, Q. et al. Blocking neuropilin-1 function has an additive effect with anti-VEGF to inhibit tumor growth. Cancer cell 11, 53-67 (2007).

8. Skoura, A., Bakic, P.R. & Megalooikonomou, V. Analyzing tree-shape anatomical structures using topological descriptors of branching and ensemble of classifiers. Journal of Theoretical and Applied Computer Science 7, 3-19 (2013).

9. Glaser, J.R. & Glaser, E.M. Neuron imaging with Neurolucida--a PC-based system for image combining microscopy. Computerized medical imaging and graphics : the official journal of the Computerized Medical Imaging Society 14, 307-317 (1990).

10. Wearne, S.L. et al. New techniques for imaging, digitization and analysis of three-dimensional neural morphology on multiple scales. Neuroscience 136, 661-680 (2005).

11. Longair, M.H., Baker, D.A. & Armstrong, J.D. Simple Neurite Tracer: open source software for reconstruction, visualization and analysis of neuronal processes. Bioinformatics 27, 2453-2454 (2011).

12. Langhammer, C.G. et al. Automated Sholl analysis of digitized neuronal morphology at multiple scales: Whole cell Sholl analysis versus Sholl analysis of arbor subregions. Cytometry A 77, 1160-1168 (2010).

13. Billeci, L., Magliaro, C., Pioggia, G. & Ahluwalia, A. NEuronMOrphological analysis tool: open-source software for quantitative morphometrics. Front Neuroinform 7, 2 (2013).

14. Schneider, C.A., Rasband, W.S. & Eliceiri, K.W. NIH Image to ImageJ: 25 years of image analysis. Nat Methods 9, 671 (2012).


15. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat Methods 9, 676-682 (2012).

16. Ferreira, T., Ou, Y., Li, S., Giniger, E. & van Meyel, D.J. Dendrite architecture organized by transcriptional control of the F-actin nucleator Spire. Development 141, 650-660 (2014).

17. Xing, L., Yao, X., Williams, K.R. & Bassell, G.J. Negative regulation of RhoA translation and signaling by hnRNP-Q1 affects cellular morphogenesis. Mol Biol Cell 23, 1500-1509 (2012).

18. Stark, K.L. et al. Altered brain microRNA biogenesis contributes to phenotypic deficits in a 22q11-deletion mouse model. Nature genetics 40, 751-760 (2008).

19. Chen, Y. et al. Ankyrin repeat-rich membrane spanning protein (kidins220) is required for neurotrophin and ephrin receptor-dependent dendrite development. J Neurosci 32, 8263-8269 (2012).

20. Ultanir, S.K. et al. Regulation of spine morphology and spine density by NMDA receptor signaling in vivo. Proc Natl Acad Sci U S A 104, 19553-19558 (2007).

21. Linkert, M. et al. Metadata matters: access to image data in the real world. J Cell Biol 189, 777-782 (2010).

22. Milosević, N.T. & Ristanović, D. The Sholl analysis of neuronal cell images: semi-log or log-log method? J Theor Biol 245, 130-140 (2007).

23. Ristanović, D., Milosević, N.T. & Stulić, V. Application of modified Sholl analysis to neuronal dendritic arborization of the cat spinal cord. J Neurosci Methods 158, 212-218 (2006).

24. Schoenen, J. The dendritic organization of the human spinal cord: the dorsal horn. Neuroscience 7, 2057-2087 (1982).

25. Buchanan, K.A. et al. Target-specific expression of presynaptic NMDA receptors in neocortical microcircuits. Neuron 75, 451-466 (2012).

26. Stuart, G., Spruston, N. & Häusser, M. Dendrites, Edn. 2nd. (Oxford University Press, Oxford ; New York; 2007).

27. Ramón y Cajal, S. Histology of the nervous system of man and vertebrates. (Oxford University Press, New York; 1995).

28. Myatt, D.R., Hadlington, T., Ascoli, G.A. & Nasuto, S.J. Neuromantic - from semi-manual to semi-automatic reconstruction of neuron morphology. Front. Neuroinform. 6, 4 (2012).

29. Bland, J.M. & Altman, D.G. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1, 307-310 (1986).

30. Binley, K.E., Ng, W.S., Tribble, J.R., Song, B. & Morgan, J.E. Sholl analysis: A quantitative comparison of semi-automated methods. J Neurosci Methods 225, 65-70 (2014).

31. Rizk, A. et al. Segmentation and quantification of subcellular structures in fluorescence microscopy images using Squassh. Nat Protoc 9, 586-596 (2014).

32. Sheppard, C.R., Gan, X., Gu, M. & Roy, M. Signal-to-Noise Ratio in Confocal Microscopes. Handbook Of Biological Confocal Microscopy, 442-452 (2006).

33. Lee, P.C., Chuang, C.C., Chiang, A.S. & Ching, Y.T. High-throughput computer method for 3D neuronal structure reconstruction from the image stack of the Drosophila brain and its applications. PLoS Comput Biol 8, e1002658 (2012).

34. Brown, K.M. et al. The DIADEM data sets: representative light microscopy images of neuronal morphology to advance automation of digital reconstructions. Neuroinformatics 9, 143-157 (2011).

35. Gillette, T., Brown, K., Svoboda, K., Liu, Y. & Ascoli, G. DIADEMchallenge.Org: A Compendium of Resources Fostering the Continuous Development of Automated Neuronal Reconstruction. Neuroinformatics 9, 303-304 (2011).

36. Bas, E. & Erdogmus, D. Principal curves as skeletons of tubular objects: locally characterizing the structures of axons. Neuroinformatics 9, 181-191 (2011).

37. Chothani, P., Mehta, V. & Stepanyants, A. Automated tracing of neurites from light microscopy stacks of images. Neuroinformatics 9, 263-278 (2011).

38. Narayanaswamy, A., Wang, Y. & Roysam, B. 3-D image pre-processing algorithms for improved automated tracing of neuronal arbors. Neuroinformatics 9, 219-231 (2011).

39. Turetken, E., Gonzalez, G., Blum, C. & Fua, P. Automated reconstruction of dendritic and axonal trees by global optimization with geometric priors. Neuroinformatics 9, 279-302 (2011).

40. Wang, Y., Narayanaswamy, A., Tsai, C.-L. & Roysam, B. A broadly applicable 3-D neuron tracing method based on open-curve snake. Neuroinformatics 9, 193-217 (2011).


41. Zhao, T. et al. Automated reconstruction of neuronal morphology based on local geometrical and global structural models. Neuroinformatics 9, 247-261 (2011).

42. Lin, L.I. A concordance correlation coefficient to evaluate reproducibility. Biometrics 45, 255-268 (1989). 43. Ferreira, T.A., Schindelin, J. & Maddock, T. Sholl Analysis 3.4.2. doi:10.5281/zenodo.10803 (2014). 44. Pologruto, T.A., Sabatini, B.L. & Svoboda, K. ScanImage: flexible software for operating laser scanning

microscopes. Biomed Eng Online 2, 13 (2003). 45. Chattopadhyaya, B. et al. Experience and activity-dependent maturation of perisomatic GABAergic

innervation in primary visual cortex during a postnatal critical period. J Neurosci 24, 9598-9611 (2004). 46. Preibisch, S., Saalfeld, S. & Tomancak, P. Globally optimal stitching of tiled 3D microscopic image

acquisitions. Bioinformatics 25, 1463-1465 (2009). 47. Ascoli, G.A., Krichmar, J.L., Scorcioni, R., Nasuto, S.J. & Senft, S.L. Computer generation and

quantitative morphometric analysis of virtual neurons. Anat. Embryol. 204, 283-301 (2001). 48. Scorcioni, R., Polavaram, S. & Ascoli, G.A. L-Measure: a web-accessible tool for the analysis,

comparison and search of digital reconstructions of neuronal morphologies. Nat Protoc 3, 866-876 (2008).

49. Kozloski, J., Hamzei-Sichani, F. & Yuste, R. Stereotyped position of local synaptic targets in neocortex. Science 293, 868-872 (2001).

50. Everitt, B. Cluster analysis, Edn. 5th. (Wiley, Chichester, West Sussex, U.K.; 2011). 51. Jefferis, G.S. et al. Comprehensive maps of Drosophila higher olfactory centers: spatially segregated fruit

and pheromone representation. Cell 128, 1187-1203 (2007).


Neuronal morphometry directly from bitmap images · number of primary branches, ... Nature Methods:...

Documents

Transcript of Neuronal morphometry directly from bitmap images · number of primary branches, ... Nature Methods:...