The NeuronCyto Package is available HERE . Our full Paper is available at Cytometry Part A. Some online demonstrations shows how it works.
The study of neuronal morphology and neurite outgrowth has been enhanced by the combination of imaging informatics and high content screening, in which thousands of images are acquired using robotic fluorescent microscopy. In order to understand the process of neurite outgrowth in the context of neuroregeneration, we used mouse neuroblastoma N1E115 as our model neuronal cell. 6000 cellular images of four different culture conditions were acquired with twochannel widefield fluorescent microscopy. We developed a software package called NeuronCyto . It is a fully automatic solution for neurite length measurement and complexity analysis. A novel approach based on topological analysis is presented to segment cells. The detected nuclei were used as references to initialize the level set function. Merging and splitting of cells segments were prevented using dynamic watershed lines based on the constraint of topological dependence. A tracing algorithm was developed to automatically trace neurites and measure their lengths quantitatively on a cellbycell basis. NeuronCyto analyzes three important biologically relevant features, which are the length, branching complexity and number of neurites. The application of NeuronCyto on the experiments of Toca1 and Serum starvation show that the transfection of Toca1 cDNA induces longer neurites with more complexities than serum starvation. 
Cellular morphogenesis and its relationship with cell functions are significant and critical in many biological studies, for example the study of neuronal morphology and neurite outgrowth. The combination of imaging informatics and high content screening enables the acquiring of thousands of images using robotic fluorescent microscopy. Manual processing of those images is subjective, labor intensive and inaccurate. Automatic analysis and extracting quantitative information are challenging, however vital for inferring new biological insights. We developed a software package called NeuronCyto, which is a fully automatic solution for neurite length measurement and complexity analysis.
Accurate segmentation of the neural cells is the prerequisite of a cellbycell based analysis, however it is common that the neural cells are clumpy and touched with each other, which makes the automatic analysis very challenging. In NeuronCyto, a novel segmentation algorithm based on level set method and topological analysis is presented to separate the clumpy cells. Merging and splitting of cells segments were prevented using dynamic watershed lines based on the constraint of topological dependence.
The basic concept of level set in segmentation is to define a level set function in higher dimensional space, evolve this function and then use a zero level () to cut the level set function such that the region in the image will be divided into two parts, which are for inside and for outside. The points of are known as zerolevel set, which consist a number of curves called level curve, as shown by the green line in Figure 1.




Figure 1. Initialization and evolution of the level set function. The green solid curves are the level curve , e.g. the boundaries of the level sets. The level curve will start from the segmentation of the nuclei based on our initialization of Eq. (4) and Eq. (5). The level curves then will evolve according to Eq. (2). The speed of level curve will determine by the variation of the image intensity, say, brighter regions will be segmented as foreground earlier.
Level set approach is a powerful tool in the image processing and computer vision communities due to its superior properties among many other active contour approaches, which include:
Regionbased and therefore robust against noise.
Continuous and smooth contours of the object;
No need to parameterize of the curve;
Good numerical stability;
Straightforward extension from two dimensional formulation to higher dimensions.
The level set method is also well known for its ability of naturally handling topological changes, for example merging and splitting, however this merit becomes a liability in our study, since the cell segmentations should conform to the sought nuclei. Our approach solved such problem while preserve other merits of level set approach. We introduced a topological constraint called topological dependence. According to this constraint, the number of cells is equal to the number of nuclei.
As we know, both the variation of gray level intensity of cell image and geometric information are essential for cell segmentation. Traditionally, the level set function is initialized as distance function in many works. However, we initialize level set function based on the image intensity and its evolution according to EulerLagrange equation also assimilates the clue of image intensity. Both the nucleus image and cell image are normalized to [0 1]. We first saturated the cell image intensity in the regions of nuclei to 1 and then we shift the modified function to negative. Figure 1 shows an example for the level set function initialization and its evolution. The initialization of the level set function is shown in Figure 1(a). The evolution will pull the level set function little by little as shown in Figure 1(b) to 1(d). The cell segments will grow outwards according to the variation of the image intensity.
We present the topological dependence as a constraint to prevent the undesired merging. Its idea is simple but novel. An example of topological dependence is shown in left side of Figure 2 and a few examples that violate this constraint are shown in the right side. Topological dependence actually means that one cell segments should contain one and only one nucleus. For more completed definition of topological dependence, please refer to our full paper.
Figure 2. Topological dependence and some violations. The left figure shows a topological dependence. The three connected regions contain one and only one black regions. The right figure shows some examples of the violation of topological dependence.
This initialization of the level set function assures when artificial time is 0, the cell segment equal to the nuclei segments as shown in Figure 3, e.g. topological dependence. In our original work, the topological dependence is preserved by the relabeling procedure according to the longest common boundary. However, it is expensive when the number of the cells is big. In this package, we use smarter way to evolve the watershed line. It is faster, more efficient and much easier for implementation. The evolution of the watershed line is shown in Figure 3. We can see that the watershed lines are rather straight in Figure 3(a) and (b). It becomes more and more accurate with the increase of time in Figure 3(c) and (d). A movie demonstrates how the watershed line is evolving based on the topological dependence.








Figure 3. Evolution of the segments and the watershed lines. The cell segments will start from the nuclei segments and evolve speed is determined by the variation of image intensity. Brighter regions will be segmented as foreground earlier. The watershed line is dynamically evolving based on the constraint of topological dependence to prevent the topological changes.
At last, a tracing algorithm was developed to automatically trace neurites and measure their lengths quantitatively. NeuronCyto analyzes three important biologically relevant features, which are the length, branching complexity and number of neurites. Neurites are traced by following the skeleton structure from the root points (red dot) in Figure 4. To facilitate tracing, points are defined as different types: Background; Untraced neurite points; Cell bodies; Cell body boundaries; Neurite Root Points. In Figure 4, those points are illustrated by different color. The black regions are the cell bodies; the blue solid lines are the boundaries of the cell body; the red dot are the root points of the neurite, which is defined as a point that the neurite starts from the cell body; and the yellow lines are the untraced neurites. Some neurites are already traced and shown in magenta, green and black. The region within the circle of dash line in Figure 4(a) is amplified in Figure 4(b). Different parts of the neurites are labeled when they are being traced and thus measure the length and complexity of the each neurite.


Figure 4. Illustration of Neurite Tracing. (a) The cell bodies and the neurites after we differentiate them by the morphological operation CLOSING. The black regions are the cell bodies; the blue solid lines are the boundaries of cell body; the red dot are the root points of neurite; the yellow lines are the untraced neurites and traced neurite are in magenta. (b) The tracing of the neurite. Green dots are the node points; magenta dots are the body points; black dots are the leaf points.
Recommended System Configuration of NeuronCyto:
1. CPU: Pentium 2.0 G above;
2. Memory: > 1 G;
3. Matlab with Image processing toolbox;
Before you download the package, please read the GNU LICENSE and Read Me .
To download the NeuronCyto Package, please click HERE.(URL of download).
Some more materials are available online:
Our original paper is available at:
http://dx.doi.org/10.1002/cyto.a.20664
Movie shows the traditional level set approach without the topological dependence constraint is available:
http://web.bii.astar.edu.sg/~yuwm/CytometrySubmission/Movies/iShowUMovie_LevelSet.mov
Movie shows the level set approach with the topological dependence constraint is available:
http://web.bii.astar.edu.sg/~yuwm/CytometrySubmission/Movies/iShowUMovie_LevelSet_Watershed.mov
Movie shows how the neurites are traced and measured is available:
http://web.bii.astar.edu.sg/~yuwm/CytometrySubmission/Movies/iShowUMovie_Tracing.mov