Lipid Droplet Counter
Finds and counts white spots in a stack and measures volume and area of them
This package contains utilities to count and measure lipid droplets or any other bright spots in a 3D stack.
It also contains:
- a 3D Watershed implementation
- a 3D Bandpass Filter
Copy the file Droplet_Counter.jar into your Imagej plugins directory. The jar file also contains the GPL-licensed source code. You can open the jar file with any archive manager, e.g. 7zip.
Author: Samuel Moll (firstname.lastname@example.org)
How to use the droplet counter plugins
write bug reports, feature requests and questions to: email@example.com
What you need:
- any ImageJ version from 2008 or newer
- the "Droplet_Finder.jar" file
- Copy all the needed files into your ImageJ/plugins folder
- Start ImageJ, open your image stack, select "Droplet Finder/Filterstacker"
- Enter the minimum and maximum feature sizes (size bounds in pixels for your droplets)
- Enter your Z/X aspect ratio
- wait until a Window labeled "BP-..." appears (could take some time)
- In the "Watershed 3D" Box enter "2.0" for all radii, check "Invert", then "OK"
- In the "Segment Analyzer" Box choose "BP-..." as image, "WS-..." as mask
- If you run Windows: before you click ok, make sure that the "Filterstacker-..."-image is not occluded by any window, otherwise you won't see the preview
- navigate the stack with the "Slice" slider while choosing optimal area and maximum threshold parameters. Click "OK"
- The Segment Analyzer will output a measurement table
What the Filterstacker actually does:
The "Filterstacker" plugin actually only does a 3D-bandpass on your stack. It will first 3D-blur the input image with a filter size of "maximum feature size" and substract the blurred image from the original. It will then eliminate small features (like noise) by blurring again with a small filter the size of "minimum feature size". The "Z/X aspect ratio" compensates for different lateral and vertical resolutions.
It will then call the "Watershed 3D" plugin that finds all local maxima (white spots) and grows regions around them, so that each region only contains one maximum. It will output a new stack where each region is labeled with a unique color. The radii control how far apart the maxima must be until they are regarded as only one maximum. Setting this to higher values makes the watershed transform more resistant to noise while increasing the running time (higher radii than 3 take VERY long) and the probability that two spots that are close together are combined into one region. Big radii also make the edges of the regions more fuzzy. Setting all radii to 2 is a good compromise. If the "Invert" option is unchecked, the plugin will find minima (black spots) instead of maxima.
Finally the Filterstacker calls the "Segment Analyzer" plugin that takes as input an image and a mask (=the watershed transform). The Segment Analyzer decides which regions contain a droplet by thresholding (explained below). It then does a FWHM threshold on all regions that passed the previous thresholding test and measures the volume, position and surface area of each droplet. The region thresholding is very simple at the moment. It only considers the maximal and summed (over the whole region) brightness (the "maximum threshold" and "area threshold" parameters). If both values are above the respective thresholds, the region is considered to contain a droplet. The Segment Analyzer has a preview. Because of ImageJ design restrictions, the stack has to be navigated with the "Slice" slider. Red marks particles, the borders between particles are green. If there is a green line through one of your particles, lower the "connect threshold".
Citing lipid droplet counter in academic papers
If you publish a paper relying on results obtained with ImageJ and this plugin, and want to add a citation to your paper, you could e.g. follow these recommendations for citing.
Q: What are the units used for the measured volume and surface area?
A: The units are pixels cubed (=voxels) and pixels squared respectively. The surface area is estimated based on the assumption that your voxels have the same height, width and length. If this is not the case (Z/X aspect ratio != 1), the estimated surface area is wrong.
Q: How can I use this plugin to find out the diameter (or other values) of my droplets?
A: The Lipid Droplet Finder outputs a table containing for each droplet the volume, surface area and position. Assuming the droplet is round and your voxels are cubic (see above question), it's diameter d can be found from the measured volume V by the simple formula d=(6V/Pi)^(1/3). The validity of the assumption that the droplet is round can be checked by comparing the measured surface area A of the droplet to the one calculated with the formula for the area of a sphere, A=Pi(d^2). Note that due to numerical errors the measured surface can actually be //smaller// than the one of a sphere of equivalent volume.
Q: Can I merge independent 2D images into one 3D stack and use the plugin to find all droplets at once?
A: This cannot be done easily since the plugin was originally designed to work with 3D images, and that it works with 2D images is more of a by-product. If you use this plugin on an image stack it will assume that it is a 3D image and connect droplets across slices.
Q: Can I use this plugin on 2D images of cells instead of stacks?
A: Yes, this works perfectly, but beware that the measured "volume" will actually be the surface area and the measured "surface area" is the circumference of the droplet.
Q: What should I put as X/Z aspect ratio?
A: The z-ratio is used to tell the plugin the ratio of your voxels, and thus the dimensions of your image stack. === === For example: Lets say you measure 30 images in the z-plane with a resolution of 512x512 each. The imaged region is 10um x 10um big, and the 30 z-planes span a distance of 2um, i.e. the lowest image of the z-stack (number 1) is right on your glass substrate and the highest image (number 30) is 2um above the glass substrate. Then your pixels/voxels have the following dimensions:
x: 10um / 512 = 19.5nm y: 10um / 512 = 19.5nm z: 2um / 30 = 66.7nm Normally, your imaging system gives you the width of a pixel (x and y values) and the stack step size (z value) directly. In our example the z-resolution is considerably worse than the x- and y-resolutions. (This is very common with CLSM images) The z-ratio is then just z divided by x, i.e. 66.7nm/19.5nm = 3.4. This is what should be put in the confusingly-named "Z/X aspect ratio" field. That said, you can experiment a bit with this setting (set it to higher/lower values than the theoretically calculated ones) to get optimal object separation.