Semi-automated quantification of axonal densities in labeled CNS tissue
Introduction
Imaging techniques are widely used in the field of neuroscience to visualize neuro-anatomical tracings of axonal processes. However, current analyses often rely upon manual counting, a time-intensive and tedious task, or upon potentially expensive commercially available programs (Maxwell et al., 1996) or extensive programming skills (Ponomarev and Davis, 2003). Therefore, we sought to apply a computerized method to quantify labeled axons from confocal or bright-field photomicrographs using software available in the public domain.
NIH Image J with the software plug-in Feature J (Meijering, 2003) is a less expensive alternative for computerized axonal quantification and has the advantage of being able to selectively detect linear structures in images using Hessian-based edge detection. The Hessian filter is currently applied for feature detection in medical images such as MRI and CT scans, mainly to detect line-like features such as vasculature (Chen and Amini, 2004, Sato et al., 1998), and is used to selectively detect linear structures in computer simulated images (Sato et al., 1998). In the filtering process, the software calculates the eigen values of a Hessian matrix for each pixel to describe the local second-order structure of the signal, providing both edge detection and shape information (Hladuvka and Groller, 2002). Therefore, by using Feature J to select for the smallest of the eigenvalues we limit our resulting ‘eigen image’ to a display of structures with the lowest degree of curvature (Napadow et al., 2001), such as labeled axons. Traditional edge detectors, such as in Adobe® Photoshop's ‘find edges,’ use the closely related Laplacian matrix, which is the sum of the eigenvectors. While these filters can detect edges, spatial information is lost and they are less suitable for thresholding (Hladuvka and Groller, 2002). We found that application of edge detection filters in Photoshop or MetaMorph® resulted in distortion, thickening and a halo effect on the axons in our images. In our hands the application of the Hessian filter functioned well to detect labeled axons from confocal or bright-field photomicrographs of CNS tissue, and the resulting analyses compared favorably to manual methods.
Section snippets
Animal surgeries
Experiments and care of all animals were performed in accordance with approved protocols of Baylor College of Medicine.
Corticospinal tract (CST) lesions, neuronal tracing and adenoviral vector delivery were performed as previously described (Zhou and Shine, 2003, Zhou et al., 2003). Ten days after a unilateral lesion of the CST, we labeled the CST axons on the unlesioned side with cortical injections of biotinylated dextran amine (BDA). Four days later, Adv.NT-3 was retrogradely delivered to
Comparison of computer software analyses of labeled axons
In searching for a computerized method to appropriately detect and analyze fluorescently labeled axons, we evaluated the effectiveness of three computer programs (MetaMorph, Photoshop, or Image J with the Feature J plug-in) and compared them to manual counting. The use of edge detection methods in either MetaMorph or Photoshop resulted in images of hollow axons, where the outside of the axon had a broad outline (Fig. 2C and D, arrows). In these situations, the numbers of pixels representing
Discussion
We describe a computerized method for the detection of labeled axons from photomicrographs using Hessian-based edge detection. This method worked well to detect labeled axons in both confocal and bright-field photomicrographs.
There are many techniques available to quantify axonal morphology and density from data sets derived from both in vitro and in vivo experiments. In some cases, a large or complex data set requires the use of sampling techniques that describe the characteristics of a
Acknowledgement
This work was funded by grants from the NIH, Christopher Reeve Paralysis Foundation, and Mission Connect of the TIRR Foundation.
References (13)
- et al.
A confocal microscopic survey of serotonergic axons in the lumbar spinal cord of the rat: co-localization with glutamate decarboxylase and neuropeptides
Neuroscience
(1996) - et al.
Quantitative analysis of three-dimensional-resolved fiber architecture in heterogeneous skeletal muscle tissue using NMR and optical imaging methods
Biophys J
(2001) - et al.
Three-dimensional multi-scale line filter for segmentation and visualization of curvilinear structures in medical images
Med Image Anal
(1998) - et al.
Quantifying 3D vascular structures in MRA images using hybrid PDE and geometric deformable models
IEEE Trans Med Imaging
(2004) - Fisher R, Perkins S, Walker A, Wolfart E Hypermedia Image Processing Reference, 2000....
- et al.
In situ expression of brain-derived neurotrophic factor or neurotrophin-3 promotes sprouting of cortical serotonergic axons following a neurotoxic lesion
J Neurosci Res
(2005)