Morphology-Based Distinction Between Healthy and Pathological Cells Utilizing Fourier Transforms and Self-Organizing Maps

This method can help answer key questions in basic research and clinical applications, such as cancer research and cancer treatment, but also in the immunology field. The main advantage of this method is that it is a static method and easy to facilitate. This technique is able to automatically identify diseased cell types based on their shape and movement characteristics.

The implications of this technique extend towards diagnosis and therapy of cancer or inflammatory diseases. This method can provide insights into actions between immune cells and cancer cells but it can be also applied in any field where three-dimensional microscopy data are necessary. It might be that individuals new to this method will struggle because of their problems to describe and validate the three-dimensional microscopy data.

We thought that a visual demonstration of this method is critical because the software tools that we utilize here are not familiar to many of the scientists. To begin, obtain a high-resolution, deconvolved three-dimensional microscopy data set as described in the accompanying text protocol. Load the 3D image data into the reconstruction software, and begin to create a 3D surface for each object.

To accomplish this, select the 3D View option and click on Surfaces, then click on the Next button to proceed with the Surface Creation Wizard. Now, select the image channel for the surface reconstruction. Choose a smoothing value that does not hide the details of the surface but also avoids porous surfaces.

Apply the smoothing function by clicking on the Smooth Option Checkbox and providing a smoothing radius. When creating three-dimensional imagery constructions of microscopy data, it is especially important to pay attention to the smoothing factor and the proper threshold method, in order to not lose any cellular characteristics or shapes. Next, select the thresholding method to find the surfaces.

Use an absolute intensity threshold when the objects, like the ones shown here, are well-separated from the background and have an approximately uniform brightness level. When the objects vary in their intensity but can still be separated from the local background and from the other objects surrounding them, apply a local contrast threshold. Set the Local Threshold Search Area according to the value of the expected diameter of the reconstructed objects.

Next, select from a list of options to filter the reconstructed surfaces according to morphological parameters of interest. This includes volume, sphericity, surface-to-volume ratio, and more. Save and export the generated surfaces in a format such, as VRML, that is compatible with the 3D animation software that will be used in the next step.

Start Blender, and go to the Output tab on the right side of the window. Select the TIFF format from the Dropdown Menu, and set the Color Depth to 8-bit RGBA. Next, switch into Scripting Mode, and navigate to the provided script file Titled:GUI_Autorotate.

py, which is downloadable from the github repository for this work. Back in the main window, click on Run Script, and choose the folder of the wrl files when prompted for input. Then, go to the Default Menu.

Here, set the rotations to a value at or above six. Run the script by clicking on the Rotate button. A rotation of six different angles is usually sufficient to distinguish the different cellular populations, however, we do not recommend to use fewer than six rotations, in order to not lose any information about shapes or characteristics.

Save the projections of the individual surfaces in the same folder that was used for the input. By default, the images are saved in an 8-bit TIFF format, which is the format required by the FIJI plugin, Shade. Open Fiji, and select Shade in the Plugins menu.

Start with the default values, and fine-tune the parameters later on. Click Okay when ready to run the program. Next, choose the source folder that contains the TIFF files that were created in the previous section, and click on Select.

Then, provide an Output Data Folder. When the first image appears, draw a rectangle surrounding the cell, and begin by clicking on Okay. As the plugin runs, you see the pre-processing of the image in the periphery finding of the cells indicated by red lines.

The coordinates of the found-edges are now used to calculate the discrete fourier components. To train self-organizing maps for the first time, begin by loading MATLAB. Load the TrainSOM MATLAP script, and then select Run to begin the training.

Make sure to correctly path the file if needed. When it starts running, and additional window will pop up to display the progress. Wait until the training is completed before proceeding.

By default, the script is set to run 2, 000 iterations, or, epics. After the training is finished, examine the network's topology plots. Here is an example of the neighbor-distances plot, the sample-hits plot, and the input-planes plot.

Self-organizing maps are an important tool to discover hidden relationships in your data set. They cluster your data objectively, and they have the advantage that they don't need a training data set, because they learn unsupervised. The network is now trained.

Right click on the file in your workspace, and save it for future use. Load in the self-organizing maps, when using an already trained map, in order to cluster a data set. Then, import the CSV file that is to be tested with the pre-loaded trained maps.

Here, we will select the CSV output of the Shade Plugin. When the data loads, change the output type to Numeric Matrix, and then select Import Selection. After the classification is finished, use the command window to evaluate the various plots.

The image show here is from a deconvolved intravital multi-photon miscroscopy data set of microglial cells. Under physiological conditions, the microglia presented a rather complex shape with multiple, highly-branched processes. When placed in a cancerous environment, such as this cortical tumor model, the microglia changed to a simpler, more spindle-like shape.

20 of these forier-shaped descripting components were used as inputs to train the self-organizing map. The trained map was then tested in order to evaluate its ability to distinguish between healthy and cancerous cells. The healthy cell population was projected onto a single area shown here, whereas, the cancerous microglia data set presented as a dumbbell-shaped active region.

Maps can also be trained by medical experts to identify distinct cell-groups. Here, resting cells, phagocytosing cells, interacting cells, and mobile cells, were identified, reconstructed, and used to train a 12x12 map. This combined map shows groups of high-hit value artificial neurons, especially in the bottom-left and middle areas of the map.

The robustness of the mapping approach was tested by using the trained self-organizing map with 3 random subsets of the same resting cell type that was not a part of the training data set. The response of the S.O.M to this input exhibits a very similar response which indicates the correct cell type. After development of this technique, now we have a toolkit, with which we can simply categorize cell-shape changes.

These changes occur, during cancer, for example, or other immune system responses, and we can measure them with microscopy, using whole animals, excised tissue, or even organs on chip.