Image analysis
The open-source workhorse for scientific image analysis: open any format, correct, threshold, measure and script analyses as macros, with plugins for colocalisation (Coloc 2, JACoP), tracking and much more.
Our ruleSimulations prepare you for the bench but don't replace hands-on practice.
ImageJ / Fiji
The open-source workhorse for scientific image analysis: open any format, correct, threshold, measure and script analyses as macros, with plugins for colocalisation (Coloc 2, JACoP), tracking and much more.
How it differs from the standard method
- Interactive exploration plus macros that apply identical steps to every image.
- Colocalisation plugins compute Pearson, Manders and Costes tests.
- Keep the original images; work on duplicates and record every step.
1From pixels to numbers
A micrograph becomes data only when it is measured: intensities in regions, numbers and shapes of objects, overlap between channels. Image analysis is a chain of steps (correction, segmentation, measurement, statistics), and each step makes assumptions that can quietly produce a confident wrong number.
Offset, shading, bleed-through
Subtract the camera offset, divide by an illumination (flat-field) image, and unmix bleed-through measured on single-colour samples before measuring anything.
Threshold and split
A threshold separates objects from background; Otsu's method adapts it to each image. Touching objects are split by seeds from a distance transform (or intensity) and a watershed.
Pearson and Manders
Pearson's r measures how intensities co-vary; Manders' M1 and M2 measure the fraction of each channel's signal overlapping the other. Both depend on which pixels are included and on thresholds.
Undoing the blur
Deconvolution reassigns out-of-focus light using the PSF. With a correct PSF and a sensible stopping point it sharpens; with a wrong PSF or too many iterations it invents detail and amplifies noise.
2Segmentation and counting
Pipelines such as CellProfiler find nuclei first, because they are bright, compact and separated, then use them as seeds for cells. The common errors are systematic: uneven illumination makes a global threshold miss objects at the edges; a threshold copied from another experiment fails when staining varies; touching nuclei are counted as one unless declumped; a declumping distance smaller than the objects cuts irregular nuclei into pieces; debris is counted unless filtered by size.
3Colocalisation without fooling yourself
Colocalisation coefficients are easy to compute and easy to inflate. Including the black space between cells adds thousands of pixels that are dark in both channels and pushes Pearson's r up.
Key terms
Practise these as flashcards →Dividing an image by the illumination profile so equal objects give equal intensities across the field.
An automatic threshold that best separates an image's histogram into two classes.
Splitting touching objects along ridges between seed points, often seeds from distance-transform maxima.
Covariance of two channels' intensities divided by the product of their standard deviations, from −1 to 1.
M1: the fraction of channel 1 signal in pixels where channel 2 is above its threshold; M2 the reverse.
Comparing a coefficient with values from images whose blocks are scrambled, to test whether overlap exceeds chance.
An iterative maximum-likelihood deconvolution for Poisson noise; more iterations sharpen more and amplify noise more.
The image of a point source; deconvolution needs one matching the real optics, measured with beads or computed carefully.
When to use it
- Measuring intensities, colocalisation or morphology in individual images: ImageJ/Fiji, with every step recorded in a macro.
- Counting and measuring thousands of cells consistently across a screen: a CellProfiler (or similar) pipeline.
- Recovering contrast and resolution in widefield or confocal stacks before measurement: deconvolution with a measured PSF.
Compared with alternatives
Measures per-cell intensities in far larger numbers, without spatial information or segmentation errors.
Removes out-of-focus light optically; deconvolution does it computationally from widefield data.
When colocalisation must mean nanometre proximity rather than overlap within the diffraction limit.
Produces the image series that segmentation and tracking pipelines analyse.
Sub-techniques
Choosing one here switches the whole page, including the protocol and the lab, to it.
The open-source workhorse for scientific image analysis: open any format, correct, threshold, measure and script analyses as macros, with plugins for colocalisation (Coloc 2, JACoP), tracking and much more.
- Interactive exploration plus macros that apply identical steps to every image.
- Colocalisation plugins compute Pearson, Manders and Costes tests.
- Keep the original images; work on duplicates and record every step.