Fiji plugins for segmentation, object classification, ratiometric analysis and intensiometric analysis. Includes a live updating GUI interface, a headless mode, a batch mode, object labelling, object classifier training and application and Trackmate 7 integration. This is under active development (as of 2026).
New FRETENATOR2.5 implemented features:
- Improved user interface
- Background subtraction (Global mean or Local label based)
- Pixel by pixel processing
- Segmentation settings saving/loading and importing from files
- Headless mode for batch
- 2D image processing
- WEKA based label classifier training and classification, using graph features for context
- Re-editing and annotation through ROI labeller
- Help button linking to tutorials (this github readme)
- Replaced default 'Max projection' with 'Average projection'
- 32 bit emission ratio image outputs (no 1000X multiplied 16 bit images) with background now assigned as NaN allowing manual ROI quantification
- Improved pixel by pixel processing - Z projections will be an average projection of the ratioed 3D image rather than Z sum projections ratioed
- Quickload segmentation settings buttons
- Bug fixes
- Separate 'Segment and intensity' plugin implementation to quantify the intensity of each channel for every ROI.
Planned features:
- Video tutorials of new interface/features
Installation requires Fiji and takes between 30s and 15 minutes, depending on whether the user's copy of Fiji is up to date. FRETENATOR has been tested on Windows 10, 11, Ubuntu 20, 22, and MacOS. Typical hardware requirements are up to date video card drivers including OpenCL and enough graphics RAM (or shared RAM) to handle the image you wish to process (typically 5x the image size).
Install FRETENATOR and dependencies by activating their update sites:
- Click Help>Update...
- Click Manage update sites
- Scroll down to C and make sure Clij and Clij2 are selected
- Scroll down to F and make sure FRETENATOR is selected
- Scroll down to I and make sure the IJPB is selected
- Accept the dialogs, allow installation, then restart ImageJ
FRETENATOR2 Segment and ratio is a powerful plugin to quickly perform ratiometric analysis of 2D, 3D or 4D microscopy images, with an new user interface and a live updating preview. The plugin performs full 3D segmentation of images, which means you don't analyse background, and does all the analysis, ready for you to plot and interpret. The algorithm can be used to analyse punctate sensors (e.g. nuclear localised) on a per object basis, or diffuse sensors (e.g. cytoplasmic) on a pixel by pixel basis, with quickload settings buttons. Saturated pixels are automatically removed. Settings can be saved and used for headless processing, or even batch (alpha).
- Results Table: ◦ Includes the ratiometric calculation (emission ratio) your channel quantifications, and x, y, z positions. This can be saved as a .csv and then analysed in python, R or excel.
- Label map: ◦ An image in which every nucleus is given a value that corresponds to the “label” in the results table.
- Emission ratio map: ◦ An image in which every nucleus is given the value of it’s emission ratio
- Average Z projected emission ratio map: ◦ An average Z projection of the emission ratio map
- Nearest point emission ratio map: ◦ A nearest point projection of the emission ratio map, with outlines added between the nuclei NB: the scale of this image is different to the original image and other images, allowing thin outlines to be drawn.
- Log: ◦ Details of the image file and exact analysis settings used to keep with your metadata. Savable as a .txt file
https://www.youtube.com/watch?v=OdPR_2kKuzg
https://www.youtube.com/watch?v=rTH1vWirORI
The segmentation tool works by a DoG or Gaussian filter, then Otsu to generate a binary map. An optional watershed can then be used to split objects, but a 3D watershed it a little too severe and causes the loss of many nuclei and many shrink down much smaller than their original size. By comparing my watershed to non watersheded binary maps I can create a map of the 'lost nuclei' to add them back in later. A connected components analysis is used to generate a label map of the watersheded nuclei, and then dilated the labelmap on zero pixels only to fill all the space. I then multiply this by the orginal threshold image to get a a good segmentation with good enough split objects. But this will give incorrect labelling to the 'lost nuclei' present in the image. To correct this, I run a connected components on the 'lost nuclei' map, to generate labels, and add on the max value of the OTHER label map. Then I use maximumImage to superimpose these labels on the other label map to get my FINAL label map.
The software will then use the segmentation to quantify statistics (postion, intesnity etc) for each nucleus for the chosen channel. Built upon the nuclear segmentation tool. Gives a dialog with segmentation settings, which can be adjusted in real time with a live labelmap max projection preview of frame 1. Pleasse note that the DoG filter and tophat background subtraction are only used to segment the image and are not applied to the channels to be quantified.
The chosen settings will then be applied to the time series and the data for emission ratio calculation etc are output to a results table. This is useful for ratiometric biosensors. Voxels saturated in the Donor or Emission (FRET) channels are excluded from analysis.
The "nearest point Z projection" option has outline drawing between segmented objects. This will make pretty Z projections where the different objects are discernable and overlayed properly.
There are two background subtraction methods. Global mean subtraction, subtracts the average intensity of the are excluded from segmentation in each channel from each pixel before performing calculation - this is good for the global background signal that is present in many camera/detector types. Local label based subtraction will process each ROI object individually, subtracting the average intensity of nearby pixels in the excluded area surrounding it, which is good for global background as well as local background such as light scattering/autofluorescence.
A specialised version of FRETENATOR2_Segment_and_ratio, developed for Tang et al 2025, which measures the fluorescence of an additional channel, in a dilated area surrounding and including the original ROI. This allows nearby fluorescence to be quantified and is included on the results table. Please select 'Local Label Based' from the "Background Subtraction Method" to use this functionality. This will be added as a new column to the results table.
Uses the last saved settings of FRETENATOR2_Segment_and_ratio, and performs analysis without opening a dialog box (faster).
Currently only reliable when run from the script editor Uses the last saved settings of FRETENATOR2_Segment_and_ratio, and performs analysis on all images in a user defined folder, then exports the analysis into another user defined folder.
A follow on tool for after segmentations where users can categorise the ROI in their segmented images. As a work in progress, it currently works on single timepoint 3D label images, allowing users to visually assign labels to one of 10 categories. Results are either output to an existing results table or can be used to measure a chosen image. Alpha functionality: In the latest version, time course analysis can be performed, but usage asumes the same label usage through time (making it compatable with Trackmate exported files - see below).
Usage: FRETENATOR ROI Labeller tutorial https://www.youtube.com/watch?v=EKXR4z5g8Pg
Training object classifiers in FN2 ROI Labeller
An object classifier can be trained directly from an individual label map and label preview image from the FN2 ROI Labeller. To train a classifier on more than one image, please use the separate FN2 Label map to training data and FN2 Train label classifier plugins.
- Fully annotate your image using manual methods.
- Click the 'Train classifier from this image' button. Wait a few seconds - generating initial training data can take up to a minute.
- Select the folder where you would like the training data and classifier saved.
- Choose all the parameters you would like the classifier trained on - if you're not sure what this means, just try the defaults!
- Click OK - your classifier should be trained within seconds. Three files will be generated and saved in the specified folder :
- A copy of the training data in csv format. This can be reused to train classifiers from this image again in the future
- A date_classifier.model that contains the classifier which must be reloaded to apply the classifier to other images.
- A date_classifier_fileheaders.json that contains the chosen parameter headings which must be reloaded to apply the classifier to other images.
Applying pretrained classifiers in FN2 ROI Labeller
Existing object classifiers can be applied directly to individual label maps and FRETENATOR results tables from the FN2 ROI Labeller. To apply a classifier:
- Click the Use pretrained classifer button.
- Leave the Label map file location field blank.
- Select the date_classifier.model file.
- Select the date_classifier_fileheaders.json file.
- Click OK, applying a classifier typically takes 15s-1min.
- Check the classifier output and correct any mistakes.
To use pretrained classifiers on timeseries images, use the dedicated plugins instead of the FN2 ROI Labeller.
Training
Training a classifier on a single image can be done from the FN2 ROI Labeller but training a classifier on a multiple annotated images is a two stage process:
- Generate training data using FN2 Label map to training data
- Train the classifier on a folder of training data using the FN2 Train label classifier plugin.
FN2 Label map to training data
To train a model, the training data from Label maps and the annotated output of the ROI Labeller must be generated and saved in the folder for model training. Training data from mutliple images can be saved to the same folder and will all be used in classifer training.
- Fully annotate your image using manual methods and have both images open in Fiji.
- Use the FN2 Label map to training data plugin.
- Select the label map and ROI map.
- Training data will be generated. This can take up to a minute.
- Click the file>save options on the training data window. Save this training data in your training folder as a csv.
FN2 Train label classifier
- Select the folder where your training data csv files are saved, your classifier will be saved to the same folder.
- Choose all the parameters you would like the classifier trained on - if you're not sure what this means, just try the defaults!
- Click OK - your classifier should be trained within seconds. Two files will be generated and saved in the specified folder :
- A date_classifier.model that contains the classifier which must be reloaded to apply the classifier to other images.
- A date_classifier_fileheaders.json that contains the chosen parameter headings which must be reloaded to apply the classifier to other images.
Applying pretrained classifiers (alpha functionality)
Existing object classifiers can be applied directly to individual label maps and FRETENATOR results tables using the FN2 Apply Weka to labels. This version is compatible with timecourse images. To apply a classifier:
- Use the FN2 Apply Weka to labels plugin.
- Leave the Label map file location field blank to apply to the open selected image, or pick one from your harddrive.
- Select the date_classifier.model file.
- Select the date_classifier_fileheaders.json file.
- Click OK, applying a classifier typically takes 15s-1min.
Note: Output will be a ROI image. An auto generated results table is coming but not yet implemented.
A simple plugin to allow Trackmate 7 analysed label images (Analyse the FRETENATOR label map for tracking then export the tracked label map as dots) to be combined with FRETENATOR_Segment_and_ratio output. This adds TrackIDs to the results table and creats a new TrackID labelmap that can be analysed with the ROI manager.
All these plugins use CLIJ/CLIJ2 to process images on the graphics card. This means image processing is lightning fast, but also means there are sometimes errors/crashes.
The majority of these crashes are due to one of two reasons: i. the image stack being too large to process on the graphics card this can be solved by using a computer with more video memory, or scaling/cropping the images to be smaller. Normally 4-5x the image size is required in video memory. Running Plugins>ImageJ on GPU (CLIJ2)>Macro tools>CLIJ2 Clinfo will allow you to select GPU and provide info on the hardware’s maximum image size.
or ii. out of date graphics card drivers. This often presents with black/blank images. This can often be solved by downloading the latest drivers from the manufacturer website (usually AMD, Nvidia or intel).