Computer Methods and Programs in Biomedicine

Volume 95, Issue 1, July 2009, Pages 62-71 



Interactive image analysis programs for quantifying injury-induced axonal beading and microtubule disruption

Devrim Kilinc (a), Gianluca Gallo (b) and Kenneth A. Barbee (a)

(a) School of Biomedical Engineering, Science, and Health Systems, Drexel University, 3141 Chestnut St., Philadelphia, PA 19104, USA

(b) Department of Neurobiology and Anatomy, Drexel University College of Medicine, Philadelphia, PA 19129, USA


Received 15 February 2008; 
revised 9 January 2009; 
accepted 9 January 2009. 
Available online 13 March 2009.


Abstract

Focal axonal beading and focal disruption of microtubule structure are characteristic to traumatic axonal injury. We have recently reproduced these morphological and structural changes in our in vitro model system [D. Kilinc, G. Gallo, K.A. Barbee, Mechanically induced membrane poration causes axonal beading and localized cytoskeletal damage, Exp. Neurol. 212 (2008) 422–430]. In order to measure bead formation objectively, an observer-independent quantification of beading was necessary. In addition, a quantitative measure for the extent of co-localization of axonal beads and microtubule disruptions was required to establish a causal relationship between focal cytoskeletal damage and bead formation. In this paper we describe Matlab-based, interactive image analysis programs for axonal beading quantification and co-localization analysis. Injury-induced increases in the axonal beading could be successfully detected using the bead analysis program.

Keywords: Axonal injury; Axonal beading; Morphometric analysis; Image analysis




1. Introduction

Diffuse axonal injury (DAI) is the diffuse form of traumatic brain injury that is a continuum of neurochemical events initiated at the time of the trauma due to mechanical forces [1]. DAI is characterized by increased membrane permeability, disturbance in the ion balance, damage to cytoskeletal elements and axonal beading morphology leading to disconnection from target tissue and subsequent cell death [2] and [3]. Axonal beads, focal swellings along the length of the axon, are reflective of the accumulation of the membrane-bound organelles that are normally transported on intact microtubule tracts [4]. Axonal transport impairment following brain trauma has been shown to localize to axonal beads [5], suggesting a causal link between focal structural damage and morphological changes.

We have recently developed an in vitro model system to study the effects of mechanical trauma on cultured primary neurons [6]. By applying uniform fluid shear stress on chick forebrain neurons, we have induced structural and morphological changes that mimicked in vivo DAI. We have found that mechanically induced injury caused axonal bead formation and focal disruption of microtubules that co-localized with beads [6]. We have also determined that mechanoporation is the underlying event in fluid shear stress injury and could be reversed by post-injury application of the membrane sealant Poloxamer 188 [6].

An image-based, observer-independent quantification of axonal beading was necessary to objectively measure bead formation and compare uninjured, injured, and treated neurons. In addition, a quantitative measure for the extent of co-localization of axonal beads and microtubule disruptions was necessary to support our hypothesis that focal cytoskeletal damage is the underlying factor in bead formation.

Image analysis algorithms are not uncommon in neuroscience research [7] and [8]. To classify neuron types in the rat brain, image-based methods have been developed to quantify morphological parameters such as soma shape and number of main dendrites [9]. The majority of the neuronal morphometric analyses have been conducted to elucidate complex structure of the dendrites and dendritic spines [10]. Methods have been developed to classify dendrite branching in cat retinal ganglion cells [11] and in pyramidal neurons of the monkey [12], as well as to compare dendritic structure, spine geometry and branching patterns in normal and pathological human brain [13].

Axonal morphometry was extensively studied in histological sections of peripheral nerves. A semi-automated method has been developed to determine axon diameter and myelin thickness in normal and pathological human superficial peroneal nerves [14]. The number and the area of axons in rabbit motor and facial nerves have been analyzed by semi-automatically cleaning artifacts using Adobe Photoshop and automatically detecting particles using ImageJ image analysis software [15]. Axon and myelin areas have been analyzed in electrically stimulated cat sciatic nerve by combining automated and manual methods [16] and in toxically induced rat peripheral neuropathy by using customized macros in ImageJ [17] and [18]. Customized routines have also been established in Metamorph software to determine the number of axons per cell body and branching characteristics of cultured peripheral sympathetic neurons [19]. In vivo [20] and [21] and in vitro [22] morphology of central neurons have been characterized using semi-automated methods. However, none of these methods are suitable to detect axonal beading morphology, for it involves local changes along the length of the axon.

In this paper, we describe interactive image analysis programs for axonal beading quantification and co-localization analysis. These programs can be easily modified to analyze different aspects of axonal morphology in culture systems and therefore possess the potential to become useful tools in neurobiology research.


2. Methods and theory

We applied mechanically induced injury on cultured chick forebrain neurons to mimic DAI. Details of cell culture methods and experimental setup can be found elsewhere [6]. Isolated neurons were cultured on indexed glass coverslips (Bellco Glass, Vineland, NJ) and were subjected to fluid shear stress injury. Indexed coverslips allow tracing of individual neurons during the experiment, enabling 7–8 neurons per coverslip to be imaged. Neurons were imaged with phase contrast microscopy before and 5, 20, and 60 min after the injury to follow the changes in their morphology. Images were taken with an inverted Nikon Diaphot Eclipse TE300 microscope (Optical Apparatus, West Chester, PA). Following injury, cultures were fixed at designated time points for further processing. Simultaneous fixation and extraction method allows fixing microtubules while extracting free tubulin monomers out of the axoplasm [23]. Fixed cultures were treated with 2 mg/ml sodium borohydride (Sigma, St. Louis, MO) and stained to reveal tubulin with FITC-conjugated DM1A anti-tubulin (1:100, Sigma) and actin filaments with rhodamine phalloidin (5:100, Invitrogen, Carlsbad, CA). Images were acquired using a Zeiss 200M microscope (Zeiss, Gottingen, Germany) and captured with AxioVision software (Zeiss).

To quantify axonal beading, individual beads that emerged following injury were counted and normalized by the length of the axon. Manual counting is subject to observer bias and can provide inconsistent results due to differences in lighting conditions, presence of pattern marks of the coverslip, and local concentration of the neurons. In order to eliminate the human influence in the counting process, a set of Matlab (MathWorks, Natick, MA) programs was created. Details of these interactive image analysis programs are described in Section 3. We recently showed that both manual and software-based methods produced similar results in detecting the increase in beading due to shear stress injury [24].

We previously showed that in sham controls, microtubules exhibit the characteristic bundled appearance in the central spine of the axon. Shear stress injury did not affect total microtubule levels during the post-injury period; however, examination of individual axons revealed that the presence of beads was related to local decreases in axonal microtubule staining [6]. To quantify this phenomenon, we created a set of Matlab programs (1) to interactively analyze microtubule staining images and (2) to determine the extent of co-localization. Details of these interactive image analysis programs are described in Section 3.


3. Program descriptions


3.1. Axonal beading quantification program

The input file for axonal beading quantification program beading.m is a 480 × 640 8-bit grayscale (pixel values range from 0 to 255) image. This program returns the length and the average radius of the axon (before filtering) and the ‘beading vector’ that contains location and size information of individual axonal beads. Beading.m runs a number of subroutines, some of which are interactive. The flowchart of beading.m is shown in Fig. 1.

Fig. 1. The flowchart of beading.m, a program that receives the phase contrast image of the neuron (Input Image) and returns spine vector S, raw and filtered bead vectors Z and Z′, and category vector C. Matlab programs (m-files) are represented in solid gray, image files in solid white, and output files in marble pattern. Dashed round line indicates an interactive process. 


Beading.m assigns the pixel values of the input image to the two-dimensional matrix ‘rawImage’. ‘RawImage’ is then sent to cleanup.m for the interactive process of removing unrelated objects from the image. On patterned coverslips, axons usually traverse index numbers or the peripheries of the index boxes. Although these lines are not on the same plane with the neuron, they create contrast and need to be removed from the image. Cleanup.m runs an interactive loop where the Matlab function ‘roifill’ fills in user-specified polygons by interpolating inward from the pixel values on the boundary of these polygons  [25]. Once the user is satisfied with the cleaned image, Matlab function ‘imcomplement’ creates a complement (pixel values subtracted from pure white) image.

The complement image is then sent to thresholding.m that runs an interactive loop where the image is transformed to a black-and-white (BW) image (a binary matrix of the same size). First, the gray level of the image is calculated using Matlab function ‘graythresh’; then Matlab function ‘im2bw’ converts the grayscale image to BW image to be displayed. The user is asked to change the gray level to assure that the BW image contains all neuronal segments. Once the user is satisfied with the thresholded image, the image is sent to selector.m, which runs another interactive loop where the user picks white areas that belong to the neuron. The user aims to exclude non-neuronal areas from the BW image; however due to the variations in the pixel value of the neuron in the grayscale image, the BW axon is usually segmented. Selector.m utilizes Matlab function ‘bwselect’ that returns a binary image of the white objects that are adjacent to the user-selected pixels [25]. These objects are collected in a temporary binary image until the user is satisfied with the selection.

The next step is filling the gaps between selected segments of the neuron. Filling.m runs an interactive loop that utilizes the intrinsic dilation and erosion functions of Matlab to provide connectivity to the neuron. Here the user is asked to pick pixels to fill the gaps between separate segments. These pixels are then added to the image before Matlab functions ‘imdilate’ and ‘imerode’ are run. Filling.m returns a binary image that represents the neuron as a single white object on black background. This image is stored in the hard disk under the name ‘Filled Image’.

‘Filled Image’, the binary representation of the neuron, is then sent to the subroutine trimming.m, where the so-called “spine” of the neuron is created and trimmed to contain the axon alone. ‘bwmorph’ is a Matlab function that applies morphological changes to binary images. When used with the argument ‘thin’ and when the number of repetitions is not limited, this function returns the spine of the neuron [25]. The spine consists of connected lines that remain when the neuronal area is shrunk to single pixel thickness. Since the spine contains lines that represent the soma and the axon branches, further processing is required. In an interactive manner, the user is first asked to select those pixels that are to be discarded and then to select a pixel on the desired portion of the spine. The subroutine spine.m creates the spine vector S that contains the indices of the pixels that make up the axon spine.

The subroutine beadcount.m takes ‘Filled Image’ and the spine vector S as inputs and returns the beading matrix Zbead as output. The flowchart of beadcount.m is shown in Fig. 2. Vector O contains the numbers that correspond to areas of circles with radii from 1 to 10. For each point along the spine, a circle whose radius varies from 1 to 10 is created (makedisk.m) and superimposed with the neuron image (overlap.m). For each point along the spine, the radius of the largest circle that is overlapped by the neuronal area more than 85% is recorded to the beading vector Z. Therefore, Z contains bead radii (or half the thickness of the axon if there is no bead) for each point along the axon spine. Since not all the values stored in Z necessarily correspond to an axonal bead, a filtering step was required to distinguish beads from the rest of the axon. The subroutine beadfilter.m filters out all pixels whose radius is smaller than twice the average radius value of the entire spine. This function also eliminates the points in close proximity to a local maximum (i.e. if a pixel has a radius of 6, next pixel is discarded if its radius is ≤5 and the further next pixel is discarded if its radius is ≤4 and so on). Filtered beading vector Z′ is saved to the hard disk. Finally, the subroutine beadcategorize.m categorizes beads with respect to their size and returns the category vector C that counts the number of pixels for each radius value from 1 to 10.

 

Fig. 2. The flowchart of beadcount.m, a program that calculates the beading vector Z using filled binary image of the axon and the spine vector S. O(i) is the area (in pixel squares) of a circle with radius i (in pixels). Briefly, for each pixel along the spine, a disk with increasing radius is created and added to the binary image at this pixel. If 85% or more of the disk area overlaps with axonal regions, the radius value is stored in Z(i). 

The image-analysis part of the axonal beading quantification program receives an intensity image and delivers the axon length L, the unfiltered and filtered beading vectors Z and Z′, and the category vector C. Use of these vectors can be adjusted according to the purpose of the analysis. We used Microsoft Excel (Microsoft Corp., Redmond, WA) software to obtain a single number to represent the “beading state” for each axon. The difference in these calculated “beading scores” for the same axon pre-injury and at 60 min post-injury is considered as the increase in beading and used in statistical analyses. The “beading score” is calculated by


     Beading score =

where i is the bead radius, ni the number of beads with size i, read from vector C, and L is the length of the axon in pixels. Taking 3rd power of bead size weights the contribution of each bead by its volume, providing a better representation of the beading of an axon. Axonal beading is characterized by the accumulation of transported cargo along the axon, and the volume of the bead directly reflects the amount of local damage  [26].


3.2. Co-localization analysis program

Co-localization analysis for an axon is done by the program coloc.m that uses the beading data obtained from beading.m and the microtubule intensity data obtained from microtubule intensity analysis program MT.m. MT.m is similar to beading.m up to the point where the axon spine is created, except that the complement image is not required since the neuron already has a black background. We obtained 1008 × 1280 microtubule staining images and converted these to the standard 480 × 640 images using bicubic sampling interpolation method of Adobe Photoshop (Adobe Systems Inc., San Jose, CA). This conversion was necessary to facilitate the use of the subroutines that were developed for the beading quantification program. The output of MT.m is the intensity vector V that contains intensity values of the raw image along the axon spine S. Once the spine vector S is created using the same subroutines as in beading.m, the subroutine getvalue.m is run, which reads the intensity value of the raw image for pixels listed in the spine vector and assigns them into the intensity vector V.

To what extent beads co-localize with minima of relative microtubule content is determined by running the program coloc.m. This program uses the beading vector Z and the intensity vector V as input and returns the co-localization matrix. The flowchart of coloc.m is shown in Fig. 3. As explained in Section 2, phase contrast images are taken at designated time points following injury. Extraction, which removes the membrane during fixation of neurons, is required for detection of microtubules independent of soluble tubulin. Removal of the membrane makes it impossible to use a later phase-contrast image for morphological analysis. Therefore, the phase contrast image that is used for determining Z and the fluorescent image that is used for determining V contain the same neuron with different orientations. Also, in some cases, the neuron can only be imaged using more than one microscope frame. If there are more than one phase contrast image for a neuron, beading vectors are joined into a single vector using attacher.m. Similarly, the intensity vectors that belong to the same axon are joined into a single intensity vector. The subroutine find_an.m extrapolates the shorter vector to the size of the longer vector and for every bead it calculates the average intensity value of the pixels within the bead diameter (e.g. pixel #16 is assigned the average intensity of pixels from #13 to #19, if its bead size equals 3). The flowchart of find_an.m is shown in Fig. 3. During image processing interactive operations, especially the handpicking of the pixels where the axon starts may result in a shift between beading and intensity vectors. Also, the direction of spine might be opposite in these vectors due different orientations in the raw images. Find_an.m optimizes relative positioning and corrects vector directions if necessary using the output of the subroutine cumulratio.m. The goal of the optimization is to maximize


where Z is the beading vector, V the intensity vector and n is the length of the axon in pixels. A potential drawback exists since the target of the optimization favors our hypothesis that beads co-localize with microtubule losses. To overcome this drawback, we have limited the range of positioning to 10% of the length of the axon such that the program can position the two images without significantly altering the outcome. Once the positioning and directionality is optimized, bead radius and intensity of those pixels that have a significant bead size is retained. Coloc.m returns a matrix that contains position, bead size, intensity, and average intensity (within bead diameter) values for all significant beads.

Fig. 3. The flowchart of coloc.m, a program that takes the beading vector Z and the intensity vector V and returns the co-localization matrix AN. Briefly, beading and intensity vectors are assigned to vector BK whose last column is the average intensity within the extent of the bead diameter. Vector AN contains the rows of BK where the bead radius is non-zero. The relative position and direction of V is optimized for maximum co-localization. 

 

4. Sample program runs

To demonstrate the actual computational process, sample runs for all programs described will be presented in this section. The phase contrast image (Fig. 4A) and the microtubule staining image (4B) of an injured neuron were obtained using the methods described in Section 2. For demonstration purposes, both images were rotated and cropped to isolate the neuron from the rest of the image and to achieve similar orientation.

Fig. 4. Images of the example neuron. (A) Phase contrast image, aligned and cropped for demonstration purposes. (B) Microtubule staining image. Boxed segments of panels A and B are magnified 3× and shown in panels C and D, respectively. Bar = 20 μm.

Beading.m reads the image file from memory and assigns the pixel values to the ‘rawImage’ (Fig. 5A). Cleanup.m interactively removes unrelated objects from the image (Fig. 5B). The complement image is created (Fig. 5C). Thresholding.m interactively transforms the image to a black-and-white (BW) image according to the parameter entered by the user (Fig. 5D). Selector.m helps the user pick white areas of the BW image that belong to the neuron (Fig. 5E). Filling.m interactively adds pixels to the selected areas to provide connectivity for the neuron, returning an image where the neuron is a single white isle on black background (Fig. 5F). Trimming.m calculates the spine of the neuron (Fig. 5G) and interactively isolates the axonal portion of the spine (Fig. 5H). Fig. 5I shows the BW image of the filled neuron along with the super-imposed axon spine. Beadcount.m calculates the beading vector Z using the novel algorithm explained in Section 3. Z is filtered using beadfilter.m (becomes Z′) and axonal beads are categorized according to their size by beadcategorize.m.

Fig. 5. Stages of computer based beading quantification program. Briefly, the raw image (A) is cleaned (B), reversed (C), and thresholded to black and white (D). The user selects neuronal regions (E) and connects them (F) before the spine is created (G) and trimmed to reflect the axon (H). Black and white image and the spine (superimposed in I) are then used to determine the beading vector. Bar = 20 μm. 

Comparison of the beading numbers calculated by manual counting and the beading score calculated by the software program is shown for each experimental group in Fig. 6. Since the beading score is a cumulative sum of the 3rd power of bead sizes (in pixels) rather than a simple count of the beads observed along the axon, its values are approximately an order of magnitude larger than the number determined by manual counting (Fig. 6) and no correlation between these values is expected. It should also be noted that the change in the beading number (or in the beading score) for each individual axon over a period of time was used in statistical comparison of experimental groups. Statistical analyses based on the beading number or the beading score revealed significant increases in the injury group compared to other experimental groups [24]. The average change in the injury group based on the manual counting was 2.62 times higher than the average change in the sham controls (1.59 ± 0.19 vs. 0.61 ± 0.13 beads per 100 μm). On the other hand, the average change in the injury group based on the beading score was 70.0 times higher than the average change in the sham controls (7.82 ± 2.63 vs. −0.11 ± 1.77 pixel2). This finding suggests that the automated method has a higher discrimination power against beads compared to the manual counting method. The underlying reason for this variation is that the automated method considers the size of all variations in the axon diameter, rather than categorizing them as beads or non-beads, based on a threshold set by the eye of the observer.

 

Fig. 6. Comparison of the beading numbers calculated by manual counting method and the beading scores calculated by the image analysis program for different experimental groups. Change in the beading score between pre-injury and 60 min post-injury images are plotted against the change in the beading number for 106 neurons. N = 25 for sham control (A), N = 30 for injured (B), N = 18 for P188-treated sham control (C), N = 33 for P188-treated injured (D). 

We have also compared the bead numbers detected by manual and automated methods. The average number of manually counted beads per axon increased from 2.64 to 3.88 in sham controls and from 2.81 to 6.00 in injured neurons. The average number of software-counted beads increased from 47.04 to 54.52 in sham controls and from 48.09 to 50.06 in injured neurons. When only beads larger than size 3 are considered taken into account, the average number of software-counted beads decreased from 16.72 to 16.64 in sham controls and increased from 17.66 to 23.38 in injured neurons. These numbers indicate that in sham controls, the increase in the number of small beads constitute the increase in the total number; whereas in injured neurons, the number of large beads increases while the number of small beads decrease.

In order to determine microtubule intensities relative to the position along the axon, the program MT.m was used. MT.m reads the image file from memory and assigns the pixel values to the ‘raw image’. As in the beading analysis program, the raw image is cleaned using cleanup.m and transformed to a BW image using thresholding.m. Neuronal images were selected using selector.m and the connectivity of the neuron is assured using filling.m. The spine is created and trimmed to reveal the axon using trimming.m. Getvalue.m reads the intensity values of the raw image along the spine and records them to the intensity vector V.

Coloc.m reads Z′ and V vectors and returns the co-localization matrix. In the sample analysis, find_an.m extrapolates V to the size of Z and assigns the average intensity of pixels within the bead diameter to the fourth column of the co-localization matrix. Bead radii and microtubule intensities of all pixels are plotted along the sample axon segment (Fig. 7A) before filtering and optimization (Fig. 7B). We tested coloc.m for its ability to detect co-localization by comparing loss in relative microtubule content at bead locations and at random locations along the axon. For each bead, we compared the intensity of the bead region with the intensity of the adjacent proximal region (towards cell body). Also, to have a baseline, we compared intensities of randomly selected regions with the intensities of adjacent proximal regions. For beads with radii greater than three, our program detected a significant intensity loss compared to a randomly selected axonal region of the same length (Fig. 8). Therefore, the cut-off number (bead radius) to eliminate insignificant beads is determined as four. The sample axon is plotted after filtering of Z and positional and directional optimization of V (Fig. 7B). Pixels with high bead radii are associated with significant decreases in the microtubule staining intensity, demonstrating a correlation between beading and focal disruption of microtubules. Also, there exists a gradual decrease in the microtubule staining intensity from the cell body towards the tip of the axon (Fig. 7).

Fig. 7. Output of the co-localization analysis program. (A) A 160 pixel long segment of the sample axon is shown for demonstration purposes. (B) Microtubule staining intensity (line) and axonal bead radii (solid circles) are plotted along the axon. The intensity vector V was extrapolated to the size of the beading vector Z. (C) Beads whose radii were less than four were filtered out and the orientation and position of the intensity vector is optimized. Pixels with high bead radii reflect regions of abrupt decrease in MT staining intensity (arrows) relative to the adjacent more proximal (i.e. closer to cell body) regions of the axon. A gradual decrease in the microtubule intensity is evident from the cell body towards the tip of the axon (distal). Bar = 5 μm.

Fig. 8. Results of the co-localization analysis program. For each bead, the microtubule staining intensity of the bead region (black bars) was compared with the intensity of the adjacent region (white bars). Also, to have a baseline, randomly selected regions (dark gray bars) were compared with their adjacent regions (light gray bars). For beads with radii four and above, there is a significant intensity loss compared to a randomly selected region with the same length. Statistical significance was determined by Student's t-test. N is displayed on bars. Error bars represent S.E.M. 

The time required for the user to analyze one phase contrast image is 3–5 min, depending on the complexity of the image background, which is larger compared to 1–2 min that would be required for a trained eye to manually count the beads and measure the length of the axon. However, the computer-based analysis provides deeper quantitative information such as the sizes and relative locations of the beads. Moreover, computer-based analysis of the microtubule images enables the detection of the subtle changes in the microtubule staining intensity.


5. Hardware/software specifications

Matlab programs presented in this paper were created using Matlab programming language, a mathematical scripting language that is similar to C++. Matlab version 7.0.0.19920 (R14) was used. A computer equipped with Intel Pentium 4 processor (2.40 GHz) and 512 MB memory is sufficient to run these programs in negligible time.


6. Mode of availability

Matlab program codes are kept in so-called m-files, which can be edited in any text editor. All 21 m-files that are required for beading analysis, microtubule analysis, and co-localization analysis programs can be freely obtained from the authors. Files will be available at public domains such as the software depository of the International Neuroinformatics Coordination Facility (http://software.incf.org).


Conflict of interest statement

None declared.


Acknowledgements

This research was supported in part by a grant from the State of Pennsylvania Tobacco Settlement Fund (KAB); by the National Institutes Health (NS048090, GG); and by Drexel University Neuroengineering Major Research Initiative (KAB, DK). The authors thank Hasan Ayaz for helpful discussions. K.A. Barbee and G. Gallo share last authorship on this work.


References

[1] T.A. Gennarelli, L.E. Thibault and D.I. Graham, Diffuse axonal injury: an important form of traumatic brain damage, Neuroscientist 4 (1998), pp. 202–215.

[2] T.K. McIntosh, D.H. Smith, D.F. Meaney, M.J. Kotapka, T.A. Gennarelli and D.I. Graham, Neuropathological sequelae of traumatic brain injury: relationship to neurochemical and biomechanical mechanisms, Lab. Invest. 74 (1996), pp. 315–342.

[3] M. Gaetz, The neurophysiology of brain injury, Clin. Neurophysiol. 115 (2004), pp. 4–18.

[4] J.T. Povlishock and C.W. Christman, The pathobiology of traumatically induced axonal injury in animals and humans: a review of current thoughts, J. Neurotrauma 12 (1995), pp. 555–564.

[5] P.C. Blumbergs, G. Scott, J. Manavis, H. Wainwright, D.A. Simpson and A.J. McLean, Staining of amyloid precursor protein to study axonal damage in mild head injury, Lancet 344 (1995), pp. 1055–1056.

[6] D. Kilinc, G. Gallo and K.A. Barbee, Mechanically induced membrane poration causes axonal beading and localized cytoskeletal damage, Exp. Neurol. 212 (2008), pp. 422–430.

[7] N. Goto and J. Goto, Morphometric evaluations of the human nervous system, Human Cell 19 (2006), pp. 49–64.

[8] G.A. Ascoli, Mobilizing the base of neuroscience data: the case of neuronal morphologies, Nat. Rev. Neurosci. 7 (2006), pp. 318–324.

[9] M.A. Masseroli, A. Bollea and G. Forloni, Quantitative morphology and shape classification of neurons by computerized image analysis, Comput. Meth. Prog. Biomed. 41 (1993), pp. 89–99.

[10] G.A. Ascoli, Neuroanatomical algorithms for dendritic modeling, Netw. Comput. Neural Syst. 13 (2002), pp. 247–260.

[11] C.L. Jones and H.F. Jelinek, Wavelet packet fractal analysis of neuronal morphology, Methods 24 (2001), pp. 347–358.

[12] C.M. Weaver, P.R. Hof, S.L. Wearne and W.B. Lindquist, Automated algorithms for multiscale morphometry of neuronal dendrites, Neural Comput. 16 (2004), pp. 1353–1383.

[13] S.L. Wearne, A. Rodriguez, D.B. Ehlenberger, A.B. Rocher, S.C. Henderson and P.R. Hof, New techniques for imaging, digitization and analysis of three-dimensional neural morphology on multiple scales, Neuroscience 136 (2005), pp. 661–680.

[14] S. Torch, Y. Usson and R. Saxod, Automated morphometric study of human peripheral nerves by image analysis, Pathol. Res. Pract. 185 (1989), pp. 567–571.

[15] F. Urso-Baiarda and A.O. Grobbelaar, Practical nerve morphometry, J. Neurosci. Methods 156 (2006), pp. 333–341.

[16] D.B. McCreery, T.G.H. Yuen, W.F. Agnew and L.A. Bullara, A quantitative computer-assisted morphometric analysis of stimulation-induced injury to myelinated fibers in a peripheral nerve, J. Neurosci. Methods 73 (1997), pp. 159–168.

[17] E.J. Lehning, B.S. Jortner, J.H. Fox, J.C. Arezzo, T. Kitano and R.M. LoPachin, Gamma-diketone peripheral neuropathy: I. Quantitative morphometric analyses of axonal atrophy and swelling, Toxicol. Appl. Pharmacol. 165 (2000), pp. 127–140.

[18] R.M. LoPachin, B.S. Jortner, M.L. Reid and S. Das, Gamma-diketone central neuropathy: quantitative morphometric analysis of axons in rat spinal cord white matter regions and nerve roots, Toxicol. Appl. Pharmacol. 193 (2003), pp. 29–46.

[19] L. Klimaschewski, W. Nindl, M. Pimpl, P. Waltinger and K. Pfaller, Biolistic transfection and morphological analysis of cultured sympathetic neurons, J. Neurosci. Methods 113 (2002), pp. 63–71.

[20] J. Calvet and M.C. Calvet, Computer assisted analysis of HRP labelled and Golgi stained Purkinje neurons, Prog. Neurobiol. 23 (1984), pp. 251–272.

[21] W. Gomes-Leal, G.J. Silva, R.B. Oliveira and C.W. Picanço-Diniz, Computer-assisted morphometric analysis of intrinsic axon terminals in the supragranular layers of cat striate cortex, Anat. Embryol. (Berl.) 205 (2002), pp. 291–300.

[22] D. Petite and M.C. Calvet, Morphometric characteristics of cryopreserved mesencephalic dopamine neurons in culture, Brain Res. 769 (1997), pp. 1–12.

[23] G. Gallo and P.G. Letourneau, Different contributions of microtubule dynamics and transport to the growth of axons and collateral sprouts, J. Neurosci. 19 (1999), pp. 3860–3873.

[24] D. Kilinc, G. Gallo, K.A. Barbee, Poloxamer 188 reduces axonal beading following mechanical trauma to cultured neurons, in: A. Dittmar, J. Clark (Eds.), EMBC07: Proceedings of the 29th IEEE EMBS Annual International Conference, Cité Internationale, Lyon, France, August 23–26, 2007, pp. 5395–5398.

[25] The MathWorks, Inc., Image Processing Toolbox User's Guide, 5th ed., 2004.

[26] S. Ochs, R. Pourmand, R.A. Jersild Jr. and R.N. Friedman, The origin and nature of beading: a reversible transformation of the shape of nerve fibers, Prog. Neurobiol. 52 (1997), pp. 391–426. 
 
Make a Free Website with Yola.