NeuronMetrics: software for semi-automated processing of cultured neuron images - PubMed (original) (raw)

NeuronMetrics: software for semi-automated processing of cultured neuron images

Martha L Narro et al. Brain Res. 2007.

Abstract

Using primary cell culture to screen for changes in neuronal morphology requires specialized analysis software. We developed NeuronMetrics for semi-automated, quantitative analysis of two-dimensional (2D) images of fluorescently labeled cultured neurons. It skeletonizes the neuron image using two complementary image-processing techniques, capturing fine terminal neurites with high fidelity. An algorithm was devised to span wide gaps in the skeleton. NeuronMetrics uses a novel strategy based on geometric features called faces to extract a branch number estimate from complex arbors with numerous neurite-to-neurite contacts, without creating a precise, contact-free representation of the neurite arbor. It estimates total neurite length, branch number, primary neurite number, territory (the area of the convex polygon bounding the skeleton and cell body), and Polarity Index (a measure of neuronal polarity). These parameters provide fundamental information about the size and shape of neurite arbors, which are critical factors for neuronal function. NeuronMetrics streamlines optional manual tasks such as removing noise, isolating the largest primary neurite, and correcting length for self-fasciculating neurites. Numeric data are output in a single text file, readily imported into other applications for further analysis. Written as modules for ImageJ, NeuronMetrics provides practical analysis tools that are easy to use and support batch processing. Depending on the need for manual intervention, processing time for a batch of approximately 60 2D images is 1.0-2.5 h, from a folder of images to a table of numeric data. NeuronMetrics' output accelerates the quantitative detection of mutations and chemical compounds that alter neurite morphology in vitro, and will contribute to the use of cultured neurons for drug discovery.

PubMed Disclaimer

Figures

Fig. 1

Fig. 1

Images of a fluorescently labeled neuron and graphic representations of the output generated by NeuronMetrics. (A, B) Double-labeled images of a cultured neuron. (A) Anti-HRP labels all neuronal membranes with uniform, high signal intensity, which is optimal for accurate skeletonization. Even the broad neurite region (marked with arrow; enlarged in panel D) shows uniform signal intensity. This is an ideal input image for NeuronMetrics. The dark pixels on the right side of the cell body are an artifact resulting from overexposure during image acquisition. Asterisk, cell body; n, noise enclosed by neurites. (B) Anti-GreekBetagal immunostaining identifies this neuron as a mushroom body GreekGamma neuron but the signal intensity is too low and patchy for optimal skeletonization by NeuronMetrics. (C) Graphic output of NeuronMetrics computation. The neurite skeleton (dominant primary neurite in magenta, remainder in green) provides an excellent representation of neuronal morphology. The cell body region of interest (ROI) is indicated in dark blue and the neuron territory is bounded by the orange polygon. The skeleton, cell body ROI, and polygon have been thickened to 3 pixels to improve visualization. (D) Enlargement of the broad neurite region in (A), showing uniform, high signal intensity throughout. (E) Cell body region after Sobel edge detection and rolling-ball background subtraction, showing the computed cell body ROI.

Fig. 2

Fig. 2

Schematic overview of the processing steps in NeuronMetrics. Square brackets indicate optional steps and asterisks indicate steps that require user input. Main modules (bold text along left side) are run by selecting them under the ImageJ “Plugins” [sic] menu. When processing a batch of images with the Neuron & Cell Body module, Neuron ROI, Noise ROIs, and Click Cell Body run sequentially on each image in the folder because the user performs a brief manual task in each step. Subsequently, all images in the folder are automatically run through Cell Body ROI. Thereafter, the user initiates the desired module, in sequence, to process all images in the folder in batch mode.

Fig. 3

Fig. 3

Skeletonization steps performed by NeuronMetrics. (A) Fluorescent anti-HRP image of a cultured neuron. (B, C) Two images created after pre-processing. (B) Thresholded image detects the high intensity fluorescent signal. (C) Image created using Laplacian edge detection enhances fine neurites. (D) Merged image created by combining the black regions of the thresholded and Laplacian images. (E) Preliminary skeleton generated by skeletonizing the merged image. Boxed regions have gaps in the skeleton or disconnected noise. (F) Final skeleton generated after gaps are filled, disconnected noise is removed, and the cell body skeleton is cleared. (G–J) Enlargements of boxed regions in (E) and (F), showing gaps and noise in the preliminary skeleton (left panels) and the outcome of gap filling and noise removal in the final skeleton (right panels). (K) Final skeleton (white) superimposed on a false-colored neuron (magenta). The inset is an enlargement of the region marked by the asterisk, showing the one-pixel-wide skeleton and a spot of enclosed noise that was manually excluded using the optional Noise ROIs module. The skeletons in E, F, and the main panel of K were thickened to three pixels to improve visualization.

Fig. 4

Fig. 4

Diagramatic representation of NeuronMetric’s gap-filling method. Each panel shows two portions of curvilinear neurite skeleton, c1 and c2, with endpoints p and q, respectively. (A) A gap, for example within a neurite, that will be filled. (B) The solid red line segment drawn between p and q and the dashed-red extended line segment to endpoints p’ and q’. Because the distance between the line segment extensions and the nearby curves is less than some specified maximum deviation, the gap between c1 and c2 will be filled. (C–D) A gap, for example between two different neurites, that will not be filled because pixels on curve c2 fail the 'close to' test.

Fig. 5

Fig. 5

Skeletonization of a neurite region with non-uniform fluorescent signal. (A) Anti-HRP image of a neuron with regions of broad, non-uniform signal (e.g., arrow). (B) Skeleton (red), obtained using the non-uniform signal mode, superimposed after thickening to three pixels to improve visibility. (C–E) Enlargement of a region, marked by arrow in panels A and B, which has non-uniform signal. For comparison, see Fig. 1D which shows an enlarged broad region with uniform signal. (C) Fluorescent image. (D) Overlay of the skeleton (one pixel wide) created using non-uniform signal mode, which results in the desired representation of neurite branching. (E) If uniform signal mode is used, the resulting skeleton encloses the darker gray sub-regions instead of traversing them, which would cause spurious length and branch number computations.

Fig. 6

Fig. 6

Detection and assessment of faces in neurite skeletons for use in the branch-count correction. (A) A cultured neuron demonstrating three types of simple neurite contacts: t, tip-to-neurite; d, tip-to-tip; x, neurite crossover. (B) Neurite skeleton depicting all faces resulting from neurite contacts. The unbounded face is beige and bounded faces are other colors. The tip-to-neurite contact at t causes the light green face; the tip-to-tip contact at d causes the lavender face; and the crossover at x causes the red face. (C) Enlargement of a self-fasciculating region of a cultured neuron. Note that the distance between adjacent neurites varies, resulting in periodic contacts in the neurite skeleton. (D) View of the whole neuron. (E–F) Corresponding neurite skeletons with each face a different color. Faces are labeled based on their classification by size and shape: s, small; m, medium; e, elongated medium; L, large. (E) Enlarged view of faces along a region of self-fasciculation. The series of adjacent small faces results from self-fasciculation rather than from tip touching. They are categorized as F1 faces and are excluded from the branch-count correction as desired. The two elongated faces are categorized as F2 faces and are excluded from the branch-count correction. This is the desired outcome for the purple elongated face, but not for the rose face which results from a tip-to-neurite contact. (F) View of all faces in the neuron. Note that the large faces, non-elongated medium faces, and isolated small faces often result from tip touching. Using categorized faces to correct for obscured neurite tips improves the branch count estimate: 13 of the 15 faces in panel (B) and 4 of the 12 faces in panel (F) are added to the respective endpoint counts.

Fig. 7

Fig. 7

Validation of branch number estimate. Comparison of manual branch counts based on neuron images with NeuronMetrics automated branch number estimate based on skeletonized representations of the same images (n = 59 neurons from a single experimental data set). The dark line represents identical values (x = y). The majority of points fall within the thin lines which represent (+) and (−) 20%.

Fig. 8

Fig. 8

Primary neurite count. (A–D) Computing the root of each primary neurite. All or part of the skeletonized neurite arbor is shown inverted, with the cell body ROI in light gray. See Fig. 1A for original neuron image. (A) Final skeleton of neurite arbor, thickened to improve visibility. Asterisk marks the cell body. (B) Enlarged view of cell body ROI and the proximal portions of the primary neurites. (C) The expanded cell body ROI (red), as computed by NeuronMetrics and added to the image in (B). (D) NeuronMetrics has removed skeleton pixels distal to the expanded cell body ROI, leaving only the skeleton pixels that connect the two ROIs. The root of each primary neurite is found by traversing each skeleton fragment from the expanded cell body ROI to the first skeleton pixel that contacts the cell body ROI. (E–G) Assessment of primary neurite count. Three different neuron cell bodies, showing the cell body ROI (gray) and adjacent skeletonized neurites (black) superimposed on the corresponding fluorescent image. Skeleton has been eliminated from the cell body regions. The automated primary neurite counts vs. manual counts are 6 vs. 5 (E), 17 vs. 18 (F), and 11 vs. 10 (G). Errors in the automated counts, relative to human interpretation of the neuron, result from skeletonization error (see e in E) and a neurite tip touching the cell body (see t in G). In F, a very short neurite (see s), that did not meet the primary-neurite length threshold, had been counted manually. (H) Validation of primary neurite count. Comparison of manual and automated primary neurite counts performed on a set of 59 neurons (images are from the same experiment depicted in Fig. 8). Box-plot distributions with the median indicated by the line inside the box. Top and bottom of the box represent the 75th and 25th percentiles, respectively, and the upper and lower crossbars represent the 90th and 10th percentiles, respectively. The median values are identical and the Mann-Whitney rank sum test shows no statistically significant difference (P = 0.66).

Fig. 9

Fig. 9

Fluorescent anti-HRP image of a cultured neuron, with superimposed neurite skeleton (magenta) and cell body ROI (dark blue). The territory (orange) of the neuron, computed by NeuronMetrics, is the area of the smallest convex polygon that encloses the skeleton and cell body ROI. The skeleton, cell body ROI, and polygon have been thickened to 3 pixels to improve visualization. This random neuron is from a wild-type (OreR-C) culture prepared from the central brain region of a larval CNS.

Fig. 10

Fig. 10

Example of a cultured neuron with extensive self-fasciculation which necessitates length correction. (A) Fluorescent anti-HRP image of the neuron. (B) Improved neurite skeleton at same scale as in A (note that the cell body skeleton is present). (C–D) Enlargements of two regions of self-fasciculation from the areas indicated in A. (C) A bundle of three strands runs clockwise; at the arrow, one strand breaks away while the other two remain very close. (D) Three adjacent strands.

Fig. 11

Fig. 11

Isolating the dominant neurite for polarity index (PI) determination. (A) Fluorescent anti-HRP image of a cultured neuron in which the dominant neurite contacts other neurites. (B) Improved neurite skeleton, at the same scale as in A, with the dominant neurite colored red and cell body skeleton automatically removed. The skeleton has been thickened to 3 pixels to improve visualization. (C–E) Enlargements of the boxed region of A. (C) Fluorescent image. Arrows indicate tip-tip contact (s), tip-neurite contact (t), and a neurite crossover (x). (D) Neurite skeleton overlaid on the fluorescent image before isolating the dominant neurite. (E) Skeleton overlay after isolation of the dominant neurite, which is shown in red. Arrows indicate how the dominant neurite was isolated: s, simple break; t, trimmed break; x, temporary (hence no longer present) break.

Fig. 12

Fig. 12

Comparison of data obtained using NeuronMetrics and previous SimplePCI-based image-analysis method. (A) Portion of a cultured anti-HRP-stained neuron imaged with the SpotRT camera. Background is higher than in other images in this report which were acquired with a more sensitive camera. (B) Skeleton generated by NeuronMetrics provides a good representation of the neurites, although some noise is skeletonized (asterisks). The skeleton has been thickened to three-pixels-wide to improve visibility. (C–E) Quantitative morphometric data obtained using a set of neurons from a single experiment, all imaged with the SpotRT camera (n = 50). Comparison of semi-automated NeuronMetrics method with the SimplePCI-based method that required extensive manual editing. (C, D) Pairs of box-plot distributions, with the Mann-Whitney rank sum test showing no statistically significant differences between results obtained by the two methods. (C) Total neurite length (P = 0.12). (D) Branch number (P = 0.70). (E) The PI profiles obtained with the two methods were very similar. This is the typical skewed PI profile of _201Y_-marked mushroom body GreekGamma neurons.

Similar articles

Cited by

References

    1. Abdul-Karim MA, Roysam B, Dowell-Mesfin NM, Jeromin A, Yuksel M, Kalyanaraman S. Automatic selection of parameters for vessel/neurite segmentation algorithms. IEEE Trans Image Process. 2005;14:1338–1350. - PubMed
    1. Al-Kofahi KA, Lasek S, Szarowski DH, Pace CJ, Nagy G, Turner JN, Roysam B. Rapid automated three-dimensional tracing of neurons from confocal image stacks. IEEE Trans Inf Technol Biomed. 2002;6:171–87. - PubMed
    1. Al-Kofahi KA, Can A, Lasek S, Szarowski DH, Dowell-Mesfin N, Shain W, Turner JN, Roysam B. Median-based robust algorithms for tracing neurons from noisy confocal microscope images. IEEE Trans Inf Technol Biomed. 2003;7:302–17. - PubMed
    1. Al-Kofahi O, Radke RJ, Roysam B, Banker G. Automated semantic analysis of changes in image sequences of neurons in culture. IEEE Trans Biomed Eng. 2006;53:1109–1123. - PubMed
    1. Andrew AM. Another efficient algorithm for convex hulls in two dimensions. Inform Process Lett. 1979;9:216–218.

Publication types

MeSH terms

Substances

Grants and funding

LinkOut - more resources