Microglia Morphometry

This video will demonstrate the usage of [ Plugins › Microglia › Measure Microglia Morphometry ]

Screenshots

Manual correction of the automated segmentation

Automated skeletonization

Plugins > Microglia > New Microglia Segmentation And Tracking

Semi-automated microglia segmentation and tracking.

Input:

Calibrated(!) single color TIFF stack time lapse with microglia signal, each TIFF plane corresponding to one time point
- Example intensity image input

Output:

TIFF stack with label mask images, each TIFF plane corrsponding to one time point. Cells keep their label across time, thereby encoding the tracking results.
- Example label mask output

Parameters

There are few parameters that can be set to adapt the algorithm to different cell types. Below are the values that were used for the data in the above publication.

Minimal cell size [um^2] = 200 micrometer^2
- This is used in the automated Semantic Segmentation
Maximal cell skeleton length [um] = 450 micrometer
- This is used in the automated Object Splitting Note:
As the parameters are in micrometer units it is critical that the input images have the correct calibration (pixel size)!

Automated semantic segmentation

For all time points, the images are smoothed using an anisotropic diffusion filter (www, doi) and then individually binarised by means of an intensity threshold that is determined automatically from the respective image intensity histogram: The intensity at the histogram mode (intensity_mode) is computed as well as the right hand side (towards higher intensities) intensity where the count decreases to half the count at the mode (intensity_rightHandHalfMode). The threshold is then computed as intensity_mode + 1.5 *( intensity_rightHandHalfBin - intensity_mode ). On the resulting binary image connected components smaller than minimal_microglia_size are removed.

Semi-automated instance segmentation and tracking

The instance segmentation (object detection) and tracking are linked and preformed with human interaction. The main challenge here is that microliga can be touching making it difficult to separate them fully automatically. To address this challenge, the binary mask is first subjected to an automated object splitting algorithm, followed by the opportunity to manually correct the segmentation.

Automated object splitting

The binary masks are converted to label masks using connected component analysis. Each connected component is skeletonized and the total skeleton length (skel_length) is measured. A likely number of microglia within each connected component is computed as n = ceil( skel_length / skel_maxLength). For each connected component with n > 1 the position of local intensity maxima (corresponding to likely microglia soma) is computed. The n brightest local intensity maxima are used as seed points to split the connected component using an intensity based watershed algorithm. Splits are only accepted if the resulting objects are larger than minimal_microglia_size.

Manual label mask correction

For each time point, the result of above automated steps are presented to the user for manual correction. The user sees the intensity image and the label mask image side by side and has the ability to draw into the label mask image. Using the ImageJ ROI tools (e.g. the polyline tool) one can select a region and either set pixels to 0 ([ Edit > Clear ]) for splitting or removing parts of cells or set pixels to some arbitrary value > 0 ([ Edit > Draw ]) for joining cells. The [ Update labels ] button will first convert the corrected label mask image to a binary image (thus, for drawing or erasing it only matters whether pixel values are 0 or > 0 ) and then recompute a new label mask image by means of a connected component labeling. Alternatively, there are two buttons in the user interface: [ Draw ] and [ Erase ]. Clicking them will set the drawing color in Fiji such that segments can be joined (draw) or separted (erase), e.g. using the pencil tool. This enables freehand drawing for segmentation correction (see the above video tutorial for a demonstration). Please note that also here the drawing color does not matter as long as it paints pixels values > 0. Upon pressing the [ Update labels ] button the colors will be corrected to reflect the cell identity.

Automated tracking and splitting

Once the above manual segmentation correction is finished the user clicks the [ Next frame ] button. This will trigger the tracking of cells from the current frame (t) to the next frame (t+1). In frame t+1 the algorithm starts from the automatically segmented cells as explained above. Next, it takes into account the (manually corrected) information from the current frame t to split cells that are likely wrongly connected in frame t+1. To do so, for each connected component in frame t+1 it determines the potentially multiple label regions that it overlaps with in frame t. Using those overlap regions as seeds it performs a watershed splitting to draw dividing lines in frame t+1. Using the resulting binary mask image it then assigns the labels in frame t+1 based on the label that each connected component maximally overlaps with in frame t. If the overlap is zero, a new label index will be assigned. This way of assigning the label indices in frame t+1 implements the tracking such that cells belonging to one track have the same label index throughout the time series.

Saving the label mask images

The output of the segmentation and tracking are a label mask TIFF stack, with each TIFF plane corresponding to one time point. This label mask stack can be saved by clicking the [ Save ] button. The [ Stop and save ] button also terminates the program and closes the user interface windows.

Plugins > Microglia > Continue Microglia Segmentation And Tracking

Semi-automated microglia segmentation and tracking. This command implements the same functionality as the one described above [ Plugins > Microglia > New Microglia Segmentation And Tracking ]. However, one can continue working on a partially finished segmentation.

The segmentation will continue on the last frame of the label mask image. Note that this allows to also correct the segmentation of a data set only containing a single frame.

Input:

Single color TIFF stack time lapse with microglia signal, each TIFF plane corresponding to one time point
- Example intensity image with 5 frames
TIFF stack with label mask images, each TIFF plane corrsponding to one time point. That is, the output of this command however, typically only partially finished with less frames than the input.
- Example label mask image with 3 frames

Output:

Same as in [ Plugins > Microglia > New Microglia Segmentation And Tracking ]

Plugins > Microglia > Measure Microglia Morphometry

The intensity and corresonding label mask images are used to compute shape and intensity features for each segmented cell.

Input:

Single color TIFF stack time lapse with microglia signal, each TIFF plane corresponding to one time point
- Example intensity input
TIFF stack with label mask images, each TIFF plane corrsponding to one time point. Cells keeping their label across time, thereby encoding the tracking results .
- Example label mask output

Output:

CSV file containing various cell shape and intensity features (see below for details)
- Example cell features CSV
TIFF stack with point annotation images, each TIFF plane corrsponding to one time point.
- Example annotation output
TIFF stack with cell skeleton images, each TIFF plane corrsponding to one time point.
- Example skeleton output