Analyze Particles in ImageJ, explained (every option, with examples)

Analyze › Analyze Particles… counts and measures every separate object in a thresholded image or binary mask. Threshold the particles first (Image › Adjust › Threshold…), choose a size range and a circularity range, tick Display results, Summarize and Exclude on edges, and click OK: the Summary window gives the count, total area, average size and %Area, and the Results table gives one row per particle with the columns chosen in Analyze › Set Measurements….

The worked example below is a sample TEM image of gold nanoparticles (synthetic, generated for these guides, 1024 × 1024 pixels at 0.25 nm per pixel). With Otsu, watershed, a minimum size of 20 nm² and Exclude on edges, ImageJ 1.54f counts 136 particles with a mean Feret diameter of 13.777 ± 1.585 nm.

What Analyze Particles needs before you run it

Analyze Particles does not find particles by itself. It traces the pixels that are already selected, either the red pixels of an active threshold or the foreground of a binary mask, and treats each connected group as one particle (pixels touching at a corner count as connected).

1. Set the scale, so that sizes come out in nm² or µm². A TIFF from ImageJ or many microscopes carries it; check Image › Properties…, or set it from the scale bar as in how to set the scale in ImageJ.
2. Choose the measurements in Analyze › Set Measurements…: Area, plus Shape descriptors and Feret's diameter for particle sizes and shapes.
3. Threshold with Image › Adjust › Threshold…: for dark particles on a bright TEM background, untick Dark background; for bright particles (SEM secondary electrons, fluorescence), tick it. Check that the red covers the particles and nothing else.
4. Clean the mask if particles touch: Apply (or Process › Binary › Convert to Mask), Process › Binary › Fill Holes, then Process › Binary › Watershed.

On the TEM sample, without the watershed, touching particles merge: ImageJ counts 74 "particles" with a mean Feret of 21.7 ± 9.7 nm instead of 136 at 13.78 ± 1.59 nm. More on that in nanoparticle size distribution from TEM images.

Every option in the Analyze Particles dialog

The sections below take each group in turn, with what it changes on a real run.

Size: calibrated units or pixels

The Size range is an area, and on a calibrated image it is in calibrated units: the label says Size (nm^2) or Size (micron^2). Type a single number for a minimum (20 means 20-Infinity), or a range such as 20-500. Tick Pixel units to give the range in pixels.

Choose the minimum from the image. On the TEM sample the smallest real particle is about 80 nm² and the noise specks are a few pixels, so anything from 5 to 50 nm² gives the same 136 particles. Write the minimum in your methods: it decides what counts as a particle.

For a size distribution, keep the upper limit at Infinity. Setting a maximum to "remove aggregates" also removes the largest single particles, and the distribution narrows for the wrong reason.

Circularity: a shape filter

Circularity is $4\pi \times \text{area}/{\text{perimeter}}^2$, 1 for a perfect circle and close to 0 for a thin line or a ragged outline. A range of 0.80-1.00 keeps round particles and drops clusters, rods and debris.

Use it with care. On the TEM sample without watershed, a 0.80-1.00 range drops the merged clusters, but the particles inside them go too: 39 particles are left of the 136 that watershed separates. If isolated particles are smaller or larger than clustered ones, the mean shifts. Circularity also depends on the pixel size: a digitised circle 200 pixels across measures 0.903 in ImageJ, not 1, and small particles of a few pixels score lower still. Feret diameter, circularity and solidity shows why.

Show: what Analyze Particles draws

• Outlines draws the outlines in a new image, each labelled with its row number in the Results table. Use it to find a particle from the table.
• Bare Outlines: the same without the numbers.
• Ellipses: the ellipse fitted to each particle (the one behind Major, Minor and AR).
• Masks: a new binary image with only the particles that passed the filters (shown with an inverted LUT, so particles look black). Useful as a clean mask for the next step.
• Count Masks: a 16-bit image in which every pixel of particle n has the value n: a label image for other tools.
• Overlay and Overlay Masks: outlines or filled shapes on the original image, without changing its pixels. This is the best check: zoom in and look for merged pairs, cut particles and missed ones.

Display results, Clear results and Summarize

Display results writes one row per particle, with the columns ticked in Set Measurements. Clear results empties the table first; untick it to collect particles from several images in one table (tick Display label in Set Measurements so you know which row came from which image). Summarize adds one row per image to a separate Summary table.

The Summary's %Area is the total area of the counted particles over the image area. With Exclude on edges on it leaves out the cut particles, so it is not the area fraction of the image: for that, see porosity from SEM images.

Add to Manager

Add to Manager puts every particle in the ROI Manager (Analyze › Tools › ROI Manager…). From there Measure measures them again, on this image or on another channel (redirect the measurement or select the other image and click Measure), and More › Save writes a RoiSet.zip. The ROIs are also how you delete a wrong particle by hand: select it in the list, Delete, then Measure again.

Exclude on edges: count and size

A particle cut by the frame is measured smaller than it is. On the TEM sample, Exclude on edges takes the count from 164 to 136, and the mean area from 124.9 to 133.9 nm². For sizes and shapes, always tick it.

For a number per area (particles per µm², pores per cm²), excluding edge particles undercounts, and large particles are excluded more often than small ones. Two fixes: count each edge particle as half (run once with and once without Exclude on edges; the difference is the number cut), or measure inside a smaller rectangle and count the particles whose centre lies in it.

Include holes and Composite ROIs

Include holes decides whether the inside of a particle counts. Ticked, every particle is measured by its outer outline, so a speck of bright noise inside a dark nanoparticle does not take a bite out of its area; this is what you want for solid particles. Unticked, holes are left out of the area, which is what you want for the shell itself (the metal in a nanoshell, the wall of a ring). On the synthetic shells the mean area is 1,058 nm² with Include holes and 457 nm² without it.

Process › Binary › Fill Holes before Analyze Particles does the same job as Include holes for solid particles, and makes the mask itself clean.

Composite ROIs matters only when holes are left out and you keep the ROIs (Add to Manager or Overlay). In ImageJ 1.54f a particle with holes is then stored as a composite ROI, the outline minus the holes. Without it the stored ROI is the plain outer outline, which includes the hole if you measure it again.

Overlay

The Overlay checkbox puts the outlines of the counted particles on the analysed image itself, as a non-destructive overlay, so you can check them without a second window. Image › Overlay › Remove Overlay takes them off.

Why does Analyze Particles count zero, or one giant particle?

The black-background setting deserves its own line. A mask is foreground on background, and ImageJ reads which is which from Process › Binary › Options… › Black background. If a mask made with one setting is analysed with the other, the background becomes the particle. When a mask looks right but the count is 1, this is the reason.

Reporting it

A methods sentence to adapt:

Particles were analysed in ImageJ 1.54f (Schneider et al., 2012). Images (0.25 nm per pixel) were thresholded with the Otsu method, holes were filled and touching particles separated with the watershed algorithm; particles smaller than 20 nm² and particles touching the image edge were excluded. The Feret diameter of each particle was measured (n = 136 particles from one image).

Give the count of particles and of images, the threshold method, the size filter, whether edge particles and touching particles were handled, and the mean ± SD with the histogram.

How do I run Analyze Particles on many images?

Analyze Particles is one command in a scriptable program, so it scales. Record the steps once with Plugins › Macros › Record…, then run the macro on a folder with Process › Batch › Macro…: each image gets its own row in the Summary table if Summarize is on. Keep the threshold method, size range and options identical across the series, and look at the outlines of a few images from each condition before you trust the numbers.

When a threshold cannot separate the particles, Fiji adds more: MorphoLibJ for label images and 3D particles, Trainable Weka Segmentation for images no threshold can split, and StarDist for crowded round objects.

An easier way: count particles in LabCanvas

LabCanvas Image Analysis does the same job in the browser, with nothing to install and no plugins: threshold, clean-up and count sit in one panel.

1. Open the TIFF (drop it on the window or press Ctrl O); the pixel size is read from an ImageJ TIFF.
2. In Count › Count particles, choose the Otsu threshold with Dark objects, tick Fill holes, Separate touching objects (watershed) and Exclude particles on the edges, and set the smallest particle to 20 nm².
3. Tick Feret's diameter in Set measurements and click Count.
4. Click Send to Graphs and pick the Feret column for a histogram with mean ± SD.

On the same TEM sample LabCanvas counts 136 particles with a Feret diameter of 13.77 ± 1.589 nm; ImageJ gave 136 and 13.777 ± 1.585 nm.

Why it is a good choice for this job:

• Nothing to install: it runs in the browser on any computer or tablet.
• ImageJ's own names: Fill holes, watershed, size, circularity and exclude on edges, and a Summary row with Count, Total Area, Average Size and %Area. Watershed is a tick box, not a separate command.
• An image check flags clipped pixels, JPEG compression and uneven light before you count.
• Results go straight to Graphs for the histogram and statistics, and each table carries a methods sentence to copy.
• Free with a LabCanvas account (CSV download is part of Pro), and every measurement is checked in more than 4,000 automated tests.

There is no macro language; Count › Batch runs the same Count particles settings on many images. Open Image Analysis.

ImageJ and Fiji are independent open-source projects; LabCanvas is not affiliated with them.

References

• Schneider CA, Rasband WS, Eliceiri KW (2012). NIH Image to ImageJ: 25 years of image analysis. Nature Methods 9(7): 671–675. doi:10.1038/nmeth.2089