perspective

FOCUS ON BIOIMAGE INFORMATICS

Fiji: an open-source platform for

biological-image analysis
Johannes Schindelin1,10, Ignacio Arganda-Carreras2, Erwin Frise3, Verena Kaynig4,
Mark Longair5, Tobias Pietzsch1, Stephan Preibisch1,10, Curtis Rueden6, Stephan Saalfeld1,
Benjamin Schmid7,10, Jean-Yves Tinevez8, Daniel James White1, Volker Hartenstein9,
Kevin Eliceiri6, Pavel Tomancak1 & Albert Cardona5,10

Fiji is a distribution of the popular open-source numbers and varieties of biological samples under a
software ImageJ focused on biological-image variety of conditions4–6. Additionally, imaging of
© 2012 Nature America, Inc. All rights reserved.

analysis. Fiji uses modern software engineering single intact small organisms is now feasible with
practices to combine powerful software libraries high resolution in two dimensions, three dimensions
with a broad range of scripting languages to and across time7, resulting in massive image data sets
enable rapid prototyping of image-processing that are well beyond the scale accessible by subjective
algorithms. Fiji facilitates the transformation of observation8,9. The analysis of both types of image
new algorithms into ImageJ plugins that can be data requires computer-vision techniques to recon-
shared with end users through an integrated update struct image volumes from many overlapping parts
system. We propose Fiji as a platform for productive (registration)10,11, to search for biologically relevant
collaboration between computer science and biology
features (segmentation)12, to follow relevant objects
research communities.
across space and time (tracking)13,14 and to compare
Much primary biological data are acquired as images. specimens through mapping to anatomical or cellular
In recent years, with the adoption of automated digital atlases15,16.
microscopy technologies, the volume and complexity Algorithms to achieve these tasks have been devel-
of image data has increased to the point that it is no oped for natural images such as an urban street view
longer feasible to extract information without using but must be adapted for biological-image data and
computers. Thus biologists increasingly rely on com- transferred into a software platform that is accessible
npg

puter scientists to come up with new solutions and on to biologists. Such software should provide intuitive
software to apply those solutions. Starting with Alan and comprehensive mechanisms to apply an algo-
Turing’s seminal paper on morphogenesis1, through rithm to biological data and visualize the results in
the development of efficient algorithms for sequence a user-friendly fashion. At the same time, the sheer
searching2, to computational whole-genome assem- size of many image data sets demands that algorithms
bly3, computer scientists have had a profound impact run fast. These needs are tackled by the emerging
on biological research. In its simplest form, compu- interdisciplinary field of bioimage informatics that
terized analysis of images overcomes the limitations applies computer-vision and other image-processing
and bias of a human observer. More importantly, com- approaches to biological research questions17,18.
puters have been essential for processing the images There are several commercial (for example, Imaris,
generated during high-throughput microscopy of large Volocity and Amira) and open-source (for instance,
1Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany. 2Department of Brain and Cognitive Sciences,

Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 3Department of Genome Dynamics, Berkeley Drosophila Genome
Project, Lawrence Berkeley National Laboratory, Berkeley, California, USA. 4Department of Computer Science of the Swiss Federal
Institute of Technology Zurich, Zurich, Switzerland. 5Institute of Neuroinformatics of the University of Zurich and Swiss Federal Institute of
Technology Zurich, Zurich, Switzerland. 6Laboratory for Optical and Computational Instrumentation, University of Wisconsin at Madison,
Madison, Wisconsin, USA. 7Department of Neurobiology and Genetics, University of Wurzburg, Wurzburg, Germany. 8Institut Pasteur,
Imagopole, La plate-forme d’imagerie dynamique, Paris, France. 9Department of Molecular, Cell and Developmental Biology, University of
California, Los Angeles, California, USA. 10Present addresses: Laboratory for Optical and Computational Instrumentation, University of
Wisconsin at Madison, Madison, Wisconsin, USA (J.S.), Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany
(B.S.) and Janelia Farm Research Campus, Howard Hughes Medical Institute, Ashburn, Virginia, USA (S.P. and A.C.). Correspondence
should be addressed to P.T. ([email protected]) or A.C. ([email protected]).
published online 28 June 2012; doi:10.1038/nmeth.2019

676  |  VOL.9  NO.7  |  JULY 2012  |  nature methods

 FOCUS ON BIOIMAGE INFORMATICS perspective
ImageJ19, CellProfiler20, Vaa3D21, BioImageXD, Icy22, Konstanz Fiji maintains compatibility with ImageJ and supplements it with
Information Miner (KNIME)23 and others) software platforms additional core functionality. The Fiji project was created to sup-
for the analysis of biological images. Commercial platforms often port the installation and maintenance of one of the more complex
focus on ease of use and broad coverage of image-processing tasks, ImageJ plugins, TrakEM2, which provides comprehensive solutions
targeting relatively inexperienced users. Almost invariably the for management, registration, segmentation and annotation of large
principal details of the image-processing algorithms are hidden, electron microscopy data sets24. TrakEM2 outgrew, in its complex-
which is undesirable for use in biological research. Conversely, ity and software-infrastructure demands, the facilities offered by
these details are transparent in open-source platforms such as classical ImageJ. Subsequently, several other advanced plugins (4D
ImageJ, whose long existence, wide adoption and extensible Viewer25, selective plane illumination microscopy (SPIM) registra-
plugin architecture has made it a tool of choice for scientists from tion26, Trainable Segmentation and many more; Supplementary
a broad range of disciplines. But ImageJ was developed primarily Table 1) came to rely on Fiji’s modern software-engineering prac-
by biologists for biologists, and its architecture does not follow tices such as a software versioning system, inclusion of third-party
modern software-engineering principles. This makes the platform libraries, automatic updates and straightforward compilation.
less attractive for computer scientists to use for delivering new The combination of advanced image-analysis solutions and the
solutions to biologists. simplicity and familiarity of ImageJ’s user interface has attracted
To address this deficiency in ImageJ, we started a new open- many users to the platform. Fiji is effectively an open-source
source software project Fiji (Fiji is just ImageJ) that updates the distribution of ImageJ that includes a great variety of organized
underlying architecture of ImageJ and allows researchers to focus libraries, plugins relevant for biological research (Supplementary
on the process of developing innovative, cutting-edge solutions for Table 1), scripting languages, extensive tutorials and documenta-
biological-image analysis. Fiji introduces powerful software libraries tion. The overproliferation and redundancy of plugins in ImageJ
© 2012 Nature America, Inc. All rights reserved.

for rapid transfer from new algorithms to practical image-analysis can make it difficult to identify solutions suitable for a particular
tools. Core algorithms available in Fiji can be exploited through biological problem. Fiji addresses this issue by offering a curated
a broad range of scripting languages that are familiar to bioinfor- selection of plugins that are organized into categories in the plugin
maticians and simplify the prototyping of new bioimage solutions. menu (Supplementary Fig. 1).
Finally, Fiji provides a robust distribution system, which ensures The Fiji project aims to provide useful functionality for a broad
that new algorithms reach its broad user base as soon as possible, range of researchers, from programming-agnostic bio­logists to
initiating an iterative refinement based on communication between bioinformaticians and software engineers to professional com-
developers and users. In summary, Fiji is designed to serve as a soft- puter science researchers (Fig. 1a). Fiji enhances ImageJ’s ease
ware-engineering ecosystem in which computer science and biology of installation by bundling all required components into a self-
research communities can collaborate to turn algorithms into usable contained package that will run on any computer platform.
programs for solving biological research questions. In addition, biologists skilled in programming can use the many
familiar scripting languages available in Fiji to build custom
How Fiji enhances ImageJ image-processing pipelines. For professional software engineers,
ImageJ, created by Wayne Rasband at the US National Institutes Fiji offers the ability to manage source-code, high-performance
of Health19, provides easy installation on arbitrary platforms and implementation of algorithms and simple worldwide deploy-
a simple user interface. ImageJ is primarily targeted at researchers ment. Finally, Fiji could become useful to computer scientists as
with minimal computer skills, but because ImageJ’s functional- it bundles standard libraries, builds bridges to other platforms (for
npg

ity can be easily extended with plugins (software components instance, Matlab through the Miji plugin) and provides facilities
that can be separately installed to add functionality), it has also for rapid prototyping of data type–agnostic, generic algorithms.
been attractive to researchers with training in software develop- Fiji’s source code is hosted in a version controlled source code
ment. Over the years an impressive assortment of ImageJ plugins repository (Git) for people to download and tinker with, and
has been developed, touching on most areas of biological-image compiles in one step (http://fiji.sc/cgi-bin/gitweb.cgi/). A special
analysis (http://imagej.nih.gov/ij/). ‘contrib’ branch is accessible for anybody to publish their plugin,

a b Figure 1 | Fiji as a high-powered distribution of ImageJ. (a) Screenshot

of Fiji’s main window, which is identical to that of ImageJ (top). Biology
Fiji updater researchers can interact with multidimensional image data in Fiji’s point-
and-click interface, which is identical to that of ImageJ. Bioinformaticians
Secondary User-
update customized
can construct image-processing pipelines with scripting languages using
Biology
researchers site Fiji the Script Editor plugin (screenshot is shown). Software engineers can use
powerful software libraries such as ImgLib (schematic diagram of ImgLib
design is shown) to transform mathematical formulations of computer
Bioinformatics science algorithms to functional programs (mathematical formulation
researchers Primary
update of difference of Gaussian blob detector). (b) The Fiji updater workflow.
Type NumericType RGBALegacyType site Researchers who created a new plugin can contribute their code to the
Software
LabelType
ComplexType ComplexFloatType

FloatType
primary update server in Dresden, Germany, where their contribution
engineers ComparableType RealType
UnsignedByteType
will be commented on, enhanced and maintained for the long term by
IntegerType ShortType

1 (x – �)2 Code
a group of core Fiji developers. Alternatively, anyone can offer their
Computer science f(x;�, �1, �2) =
�1 2π
exp –
2�21
review plugins directly through their own secondary update site. Users can freely
researchers
2
1 (x – �)
–
�2 2π
exp –
2�22 customize their Fiji installation by choosing what to download.

nature methods  |  VOL.9  NO.7  |  JULY 2012  |  677

 perspective FOCUS ON BIOIMAGE INFORMATICS
library or script. Upon successful review, the code is included in multidimensional images and apply it to a large collection of
the Fiji package system and immediately released to thousands images in a directory using scripting commands for file manipula-
of Fiji users worldwide via an integrated Fiji Updater (Fig. 1b). tion (Supplementary Fig. 2). The details of the code are unimpor-
This mechanism provides a formalized way to release new image- tant (extensive tutorials are available at http://fiji.sc/wiki/index.
analysis solutions to biological researchers while ensuring quality php/Category:Scripting), but it is crucial to note that all the script-
control of the contributed code and its long-term maintenance. ing languages enable users to access Fiji’s extensive algorithm
Every time Fiji is launched it offers to download available plugin libraries that implement advanced image analysis techniques
updates from the main server in Dresden, Germany or from alter- in Java. Thus researchers do not need to become fluent in Java
nate update sites. Research groups can also set up their own update programming and can use their existing scripting skills to apply
sites and offer new plugins to the community by following simple complex algorithms to their data. Scripts are seamlessly included
instructions (http://fiji.sc/Adding_Update_Sites). Consequently, in Fiji’s menu structure and can be publicly distributed through
users can select which set of plugins they obtain from the various the Fiji updater. These publishing mechanisms ensure validity
update sites, forming their own customized Fiji installation that and long-term viability, and are more convenient for biologists
suits their image-analysis needs (Fig. 1b). compared to closed scripting environments such as Matlab.
The Fiji project emphasizes communication among users and A key feature of the Fiji project is ImgLib (Fig. 2b–j)27, a library
developers. Fiji plugins are extensively documented via an active for type-, dimension and storage-independent representation of
wiki, which is the definitive guide to all aspects of Fiji’s installation, image data that enables writing generic algorithms in a high-
maintenance, programming and usage (http://fiji.sc/). Feedback and performance manner. In ImgLib, image-processing algorithms
feature requests are posted to a dedicated mailing list. Fiji develop- are implemented just once and can be applied without change to
ers engage in active discussions online, and the core Fiji developers most kinds of images. In other words, ImgLib enables users to
© 2012 Nature America, Inc. All rights reserved.

meet regularly for coding sprees called ′hackathons′ at which they apply a complex image-processing algorithm to any image regard-
work collaboratively on the Fiji source code. These events typically less of how many dimensions the image has (one dimension, two
result in dramatic improvement and extension of the Fiji code base, dimensions, three dimensions, two dimensions plus time, three
the ′hackathon effect′ (Supplementary Video 1). dimensions plus time and so on), what the underlying data type
In short, Fiji focuses on the process of making new solutions is (8-bit, 16-bit and so on) or how the image is stored (in memory,
practical and bringing them immediately to the users. Next we dis- paged to disc or distributed over the net).
cuss how scripting languages in Fiji empower biologists to do com- We demonstrate the dimensionality independence of ImgLib
plex image analysis by themselves and how the Fiji library ImgLib when applying segmentation algorithms to confocal scans of
simplifies transfer of abstract algorithms into usable programs. Caenorhabditis elegans larvae expressing fluorescent markers of
cell nuclei (Fig. 2b–j). Difference of Gaussian (DoG) and maxi-
Scripting and ImgLib for implementing algorithms mally stable extremal regions (MSER)28 are two examples of blob-
Analysis of biological image data typically requires applying mul- detection algorithms inspired by computer-vision literature that
tiple algorithms in sequence to many images. Scripting uses a are suitable for detecting cells in biological images. The DoG is
series of simple programming commands (the ‘script’) to define computed by subtracting two consecutive Gaussian convolu-
a sequence of algorithmic operations and then apply them to an tions of the image. Intensity maxima and minima in the DoG
image collection. map represent bright and dark blob detections, respectively. The
The simple macro language that allows recording of commands MSER tree provides a nested hierarchy of blob-segmentation hypo­
npg

and construction of basic programs made scripting popular theses, which is particularly relevant in a densely packed cellular
among ImageJ users. Fiji builds on that functionality by sup- field. Both algorithms are applicable for processing of 1D curves,
porting a broad range of scripting languages (Jython, Clojure, 2D image slices and 3D image volumes (Fig. 2b–j). With ImgLib
Javascript, JRuby and Beanshell), providing the most compre- a single, simple piece of computer code (Supplementary Fig. 3)
hensive selection of programming tools on both open-source and can run the algorithms on images of arbitrary dimensions without
commercial bioimaging platforms. These languages can be used any modification (Fig. 2b–j).
without knowledge of Java, and compared to the macro language ImgLib is and should be invisible to casual users. It is an ‘under
offer more advanced programming syntax while retaining relative the hood’ advanced software engineering concept that makes pro-
simplicity for the casual programmer. gramming easier. ImgLib empowers users and computer science
In Fiji, individual scripting commands can be quickly tested by researchers with the abstraction necessary to seamlessly translate
applying them interactively to opened images using the appropri- the mathematical description of an advanced idea or algorithm
ate interpreter plugin (for example, Jython Interpreter). Beyond into concise code that will run efficiently on large bioimages.
interactive execution of commands on current images, Fiji also Similar libraries that separate the algorithm essence from the
provides a script editor that enables writing, debugging, testing actual implementation are available on other platforms such as
and running of arbitrarily complex scripts in all the above lan- Insight Segmentation and Registration Toolkit (ITK)29 and Vision
guages and additionally Java itself. With the script editor, users can with Generic Algorithms (VIGRA)30.
perform complex tasks with a few lines of code, requiring minimal Both examples of scripts mentioned above use ImgLib to do
programming knowledge. It becomes relatively simple (16 lines the processing steered by simple scripting commands. Equivalent
of code) to load a multichannel three-dimensional (3D) image programs in ImageJ’s macro language or Java would be much
stack, identify all cells in one channel and visualize the results more complex or not possible. This is due to the combination
in the 4D Viewer plugin (Fig. 2a). Similarly users can imple- of scripting-language simplicity and the dimension-, type- and
ment a simple task such as swapping fluorescence channels in storage strategy–independence of ImgLib. Next we showcase how

678  |  VOL.9  NO.7  |  JULY 2012  |  nature methods

 FOCUS ON BIOIMAGE INFORMATICS perspective

a 1 from ij3d import Image3DUniverse

2 from javax.vecmath import Color3f, Point3f
Figure 2 | Scripting and ImgLib. (a) Screenshot of 4D Viewer plugin with
3 cell = 5.0 # diameter in microns orthogonal view of the 3D image of Drosophila melanogaster first instar
4 minPeak = 40 # The minimum intensity for a peak to be considered
5 imp = IJ.openImage (‘http://fiji.sc/samples/first-instar-brain.zip’) brain (left), in which cortex and neuropile glia are green, labeled by
6 cal = imp.getCalibration ()
7 sigmaLarge = [cell / cal.pixelWidth, cell / cal.pixelHeight, cell / cal.pixelDepth]
Nirvana-Gal4 and UASmcd8GFP, surface glial cells are red, labeled with
8 sigmaSmall = [a / 2 for a in sigmaLarge] anti-repo antibody and all nuclei are blue, labeled with Sytox. Red spheres
9 peaks = DoGPeaks(Red(ImgLib.wrap(imp)), sigmaLarge, sigmaSmall, minPeak, 1)
10 print ‘Found’, len(peaks), ‘peaks’ mark the surface glial cells detected using a simple Jython script shown
11 points = peaks.asPoints ([cal.pixelWidth, cal.pixelHeight, cal.pixelDepth])
12 # Show the peaks as spheres in 3D, along with orthoslices on the right. The script opens a 3D RGB image (line 5) and automatically
13 univ = Image3DUniverse (512, 512)
14 univ.addIcospheres (points, Color3f (1, 0, 0), 2, cell / 2, ‘Cells’).setLocked (True)
counts red surface glial cells using the DoG detector (line 9), applying
15 univ.addOrthoslice (imp).setLocked (True) constraints for cell size and labeling intensity (lines 3 and 4). These
16 univ.show ()
constraints are expressed as DoG sigma parameters (lines 7 and 8) by
b extracting image dimensions from metadata (line 6). Cell count is printed
in the dialog box (line 10), and cells are subsequently displayed in the
1D
4D Viewer as red spheres of fixed diameter (lines 13–16). (b–j) Output of
c ImgLib algorithms MSER (d,g,j) and DoG (c,f,i) for one (c,d), two (f,g)
and three (i,j) dimensions. The 3D input image was a confocal stack of
1D
d C. elegans expressing a nuclear marker (h); a slice from the stack was
used as the 2D input image (e) (scale bar, 10 µm), and a line segment
1D from the slice was used as the 1D input image (b). MSER and DoG were
e run on all input images without changing the code. Nested MSER regions
representing competing segmentation hypotheses for the nuclei are
2D colored green, red, blue and magenta.
f
© 2012 Nature America, Inc. All rights reserved.

2D
pairs of overlapping multidimensional image tiles (Fig. 3b).
g Subsequently, the plugin uses a global optimization procedure
2D that evenly spreads the error of the registration steps, avoiding
h the introduction of artifactual deformations and delivering an
undistorted representation of the specimen31. The reconstructed
3D
volume is visualized in the 4D Viewer25 (Fig. 3c), structures are
i
segmented using manual segmentation tools such as Segmentation
3D Editor (Fig. 3d), allowing measurements of lengths, areas, vol-
j umes of segmented elements (Fig. 3e) and quantification of their
3D
overlap by colocalization analysis.

the synergy of Fiji, ImgLib and other libraries results in trans- ssTEM reconstruction. ssTEM is experiencing a renaissance
formation of abstract algorithms into usable applications for the as a tool of choice for mapping brain micro-circuitry32,33. Large
analysis of biological images. sections are typically imaged as sets of overlapping image tiles
(Fig. 3f) and data sets consisting of hundreds or even thousands
Advanced image-processing pipelines in Fiji of such sections are becoming available. Fiji users can first apply a
The growing collection of modern image analysis algorithms in lens deformation model to pre-process the ssTEM images before
npg

Fiji is a product of interactions between Fiji projects that use com- registration using the Distortion Correction plugin34. Relatively
puter-vision algorithms to support ongoing biological research. simple ssTEM data sets can be reconstructed using standalone
We briefly outline three advanced image-processing pipelines plugins such as Register Virtual Stack Slices, Elastic Alignment
for stitching multidimensional images from tiles (Fig. 3a–e), and Montage35 or bUnwarpJ36. Arbitrarily large ssTEM data sets
reconstruction of massive serial section transmission electron consisting of tens of thousands images are best assembled using the
microscopy (ssTEM) mosaics (Fig. 3f–j) and reconstruction integrative TrakEM2 plugin24 for registration, segmentation and
of long-term, time-lapse, multiview recordings of development analysis of electron-microscopy data. One of the many avenues for
with selective-plane illumination microscopy (SPIM) (Fig. 3k–o). reconstructing ssTEM data in TrakEM2 relies on the scale-invari-
Many more examples of projects and synergies can be found on ant feature transform (SIFT)37 to detect corresponding image con-
the Fiji wiki (http://fiji.sc/), and we invite biologists with pro- tent and then solves the globally optimal configuration of massive
gramming skills, professional developers with interest in biology tile systems11 (Fig. 3g,h and Supplementary Video 2).
and computer-vision researchers with a new algorithm looking Reconstructed volumes can be segmented into a series of
for applications to join the Fiji movement and contribute new profiles following individual neurons either using the powerful
approaches, ideas and plugins. manual segmentation tools (Fig. 3i) or automatically with Active
Contours38 and Level Sets39 algorithms integrated into TrakEM2.
Stitching of confocal stacks. Large biological samples are fre- Alternatively, to extract the neuronal tracks from electron micro-
quently imaged with high-resolution microscopy techniques as scopy data, users can employ the Trainable Segmentation plugin,
sets of overlapping images or image stacks (Fig. 3a). The recon- which leverages the Waikato Environment for Knowledge Analysis
struction of the entire specimen involves pairwise registration (WEKA) library for machine learning40 to extract statistical prop-
of the overlapping images, with a strategy to minimize the dis- erties from user-provided examples and apply them to classify
placement of corresponding image content. The Stitching plugin the rest of the image or other similar images41. The analysis of
uses phase correlation to compute the optimal translation among the reconstructed ssTEM scans of parts of the nervous system

nature methods  |  VOL.9  NO.7  |  JULY 2012  |  679

 perspective FOCUS ON BIOIMAGE INFORMATICS

a Microscopy data b Registration c Visualization d Segmentation e Measurements

4,000

3,000

Volume (µm3)
2,000

1,000

0
Phase correlation Type

f g h i j
2,000

presynaptic sites
1,500

Number of
1,000

500

0
1 5 10 15
Number of postsynaptic
SIFT Error
partners

k l m n o 5,000
© 2012 Nature America, Inc. All rights reserved.

4,000

Number of nuclei
3,000

2,000

1,000

0
Geometric 12 13
<1 px

10 px

>100 px

local Nuclear division

descriptor cycle

Figure 3 | Fiji projects. (a–e) Stitching plugin for globally optimal registration of tiled 3D confocal images. The 3D confocal stacks of D. melanogaster
first instar larval nervous system (a) were registered using phase correlation with global optimization (b) and visualized in the Fiji 4D Viewer (c).
Four labeled neurons (color coded in 4D Viewer (d)) were segmented using a manual segmentation plugin (segmentation editor) and their volumes were
measured (e). (f–j) Globally optimal reconstruction of large ssTEM mosaics using TrakEM2 plugin. Schematic of the ssTEM mosaic; each square is an
individual image tile, and independent sections are color-coded (f). Screenshot of a video visualizing the progress of global optimization for a single
section. SIFT features (SIFT), and residual error signifying displacement of corresponding SIFT features at the current iteration (error) are shown (g).
Dual-color overlay of two registered consecutive sections showing the entire hemisphere of the larval brain (h). Scale bar, 10 µm. Axonal profiles in
a small part of a single section in the ssTEM data set were manually segmented using TrakEM2. Each profile is labeled with a different color (i). Scale
bar, 0.5 µm. Relationship between numbers of presynaptic partners and postsynaptic sites extracted manually with TrakEM2 from a micro-cube of the
registered data (j). (k–o) Plugin suite for processing of multiview SPIM data. Schematic representation of multiview (4D) SPIM imaging showing 3D
npg

stacks of the same specimen acquired from different angles (k). Progress of the global optimization of multiview SPIM acquisition of a D. melanogaster
embryo (l). Corresponding geometric bead descriptors are colored according to their residual displacement at the current iteration of the optimizer.
The resulting reconstructed embryo at the 12 th (top) and 13th (bottom) nuclear division cycle shown as a 3D rendering in Fiji’s 3D viewer (m). Scale bar,
50 µm. Results of the DoG segmentation of the nuclei marked with His-YFP (n); same stages as in m. Each nucleus is marked with a different color.
Quantification of the nuclear counts at the 12th and 13th nuclear division in the embryo shown in m and n (o).

requires quantitative comparisons across different samples42 such throughout development. SPIM is an emerging technique to
as counting individual synaptic connections between segmented achieve these goals, and Fiji provides a comprehensive pipeline for
neurons (Fig. 3j). Ultimately high-resolution TEM scans need to acquisition, reconstruction, segmentation, tracking and analysis of
be registered to low-resolution light microscopy data for correla- terabyte-size long-term, time-lapse, multiview SPIM data sets.
tive analysis and tools such as the Simple Neurite Tracer plugin43, For image acquisition with SPIM, Fiji synergizes with another
the Virtual Insect Brain (VIB) Protocol and Interactive Image large ImageJ offshoot project, Micro-Manager44, to provide open-
Transformations using various landmark correpondences are source steering software for custom OpenSPIM microscopes (P.T.;
ready to be adapted for this purpose. The flexibility and integrative unpublished results) to deliver 3D stacks of the same specimen
nature of TrakEM2 with its unique ability to reconstruct very large imaged from different angles directly into the Fiji environment
electron microscopy data sets on commodity hardware cannot be (Fig. 3k). After acquisition, the multiview data can be recon-
found on any other bioimaging platform to our knowledge. structed into a continuous 3D volume using fluorescent beads
in the rigid agarose medium as fiduciary markers (bead-based
Processing of multiview SPIM data. Modern developmental SPIM Registration plugin26). The constellations of fluorescent
biologists want to image developing embryos in toto with resolu- beads form local geometric descriptors that are matched across
tion sufficient to distinguish and track individual labeled cells views and used to create a 3D image by globally optimizing

680  |  VOL.9  NO.7  |  JULY 2012  |  nature methods

 FOCUS ON BIOIMAGE INFORMATICS perspective
Figure 4 | Fiji usage. (a) A chart showing the a b
number of unique visitors to the Fiji wiki per 25,000
month over the last three years. (b) World map
with locations of computers that updated Fiji 20,000

Unique visitors
between 20 March 2012 and 27 March 2012. 15,000

10,000

local affine transformation models (Fig. 3l 5,000

and Supplementary Video 3). The recon-
0
structed images can be optionally fused 2009 2010 2011 2012
into a single output volume (Fig. 3m) using Year

the ‘multiview fusion’ plugin. Interactive

DoG segmentation that is part of the bead-based SPIM registra- projects and new talented researchers join, expanding the
tion plugin can be used to segment and count the nuclei in the ability of biologists to cope with image data.
reconstructed specimen (Fig. 3n). The registration process, the Many image-analysis solutions available for Fiji have been
segmentation results (Fig. 3o) and the tracking of nuclei across developed uniquely under Fiji and do not have a comparable alter-
time (Supplementary Video 4) can be visualized using the Fiji native in other platforms. Some of the standard image-processing
4D Viewer25. All this is happening with truly massive long-term, or visualization tasks implemented in Fiji plugins are addressed,
time-lapse, multiview SPIM data sets in the terabyte range and sometimes more comprehensively, by other platforms (for
to our knowledge makes Fiji currently the only widely available instance, visualization in three dimensions (Vaa3D21) or train-
solution for dealing with this type of data. able segmentation (CellProfiler20)). However, in Fiji, these tools
© 2012 Nature America, Inc. All rights reserved.

are embedded in a large ecosystem of plugins that can interact

Fiji-plugin integration. The various Fiji plugins interact on through a common infrastructure, for example, by assembling
two principal levels. Trivially, the output of one plugin can diverse image-processing steps into complex pipelines with the
serve as input for another because they all run in a common use of scripting languages.
platform. More elaborate integration is possible because most The reliance on Java invites the question of whether Fiji
Fiji methods are increasingly implemented as software libraries can deal efficiently with computing-intensive tasks on top of
using ImgLib for data representation. For instance, the confo- large biological-image data. It is true that careful algorithm
cal stack, ssTEM mosaic and multiview SPIM registration plu- implementations in the C/C++ programming language will
gins are using a common global optimizer capable of accepting sometimes execute faster than the equivalent Java program. In
different representations of image content (phase correlation, other cases, Java outperforms C/C++ (for example, http://fiji.
SIFT and local geometric descriptors) and minimizing the dis- sc/Why_Java). It is equally true that specialized implemen-
placement of correspondences using the transformation model tations will typically outperform generic ones. Many of the
appropriate for the given imaging modality (translation only, flagship Fiji projects described above deal with massive tera-
rigid and affine). The integration of standalone plugins into byte-sized data sets and yet deliver interactive performance on
tightly cooperating suites such as TrakEM2 is an ongoing trend standard hardware. Important advantages of Java are the ease
in all Fiji projects. Casual users benefit from such integration by of installation on a variety of platforms, portability, stability
being able to seamlessly combine solutions from diverse areas and backwards compatibility that are harder to achieve with
npg

of bioimage analysis and create efficient analysis pipelines to packages built with C. Thus, Fiji strikes a balance between
interpret their images. usability, performance and flexibility, and as such it occupies a
unique position in the landscape of biological-image analysis
Conclusion where it synergistically complements other tools as well as
ImageJ has for a long time been the tool of choice for biologists provides unique and new capabilities.
who need basic as well as advanced image analysis. Over the life- Fiji is open not only with respect to its source code but also
time of ImageJ, a revolution in microscopy brought about an order in its relationship with other platforms. It aims to communicate
of magnitude increase in typical image sizes and two or more and integrate with other bioimage analysis software that outper-
orders of magnitude increase in throughput. The Fiji project pro- form Fiji on a particular task. An example of such integration is
vides biologists with powerful tools for generation of advanced the interface to Matlab and ITK, the availability of ImgLib for
image-processing pipelines, via scripting languages and feature- adoption in any Java-based platform (for instance, KNIME and
rich libraries for processing large quantities of large images, while ImageJ2) and the participation of developers from diverse plat-
building on the simplicity of ImageJ. forms in Fiji-centered hackathons (representatives of ImageJ2,
Over the past two years the Fiji project has achieved CellProfiler, Micro-Manager, ITK, KNIME, Icy, BioImageXD and
remarkable international recognition (Fig. 4a). Analysis of OMERO attended the last two hackathons). Most importantly, in
Fiji updater records reveals that Fiji is used in every major the future, Fiji will reconnect to its roots by becoming the outer,
academic research center throughout the world (Fig. 4b). application-oriented layer of the ImageJ2 (http://developer.imagej.
The number of Fiji downloads and the access to the Fiji wiki net/) project, the aim of which is to redesign the very core of
(Fig. 4a) has grown steadily over the past years currently with ImageJ to meet the demands of next-generation biology research.
20,000 estimated users. Several prominent scientific institu- Fiji will focus on developing state-of-the-art plugins inspired by
tions use Fiji as an open-source solution for image-processing the needs of biological research projects, and ImageJ2 will put an
needs of their researchers. As the community grows, new emphasis on creating the infrastructure necessary to prototype,

nature methods  |  VOL.9  NO.7  |  JULY 2012  |  681

 perspective FOCUS ON BIOIMAGE INFORMATICS

realize, distribute and deploy the new algorithm solutions in a 15. Long, F., Peng, H., Liu, X., Kim, S.K. & Myers, E. A 3D digital atlas of C. elegans
sustainable manner. The two projects have a common starting and its application to single-cell analyses. Nat. Methods 6, 667–672 (2009).
16. Peng, H. et al. BrainAligner: 3D registration atlases of Drosophila brains.
point, the classical ImageJ environment, and are committed to Nat. Methods 8, 493–500 (2011).
work together, exchanging code and coordinating design deci- 17. Swedlow, J.R. & Eliceiri, K.W. Open source bioimage informatics for cell
sions. The process of creating a powerful synergy between Fiji biology. Trends Cell Biol. 19, 656–660 (2009).
and ImageJ2 has already begun by incorporating the flagship Fiji 18. Peng, H. Bioimage informatics: a new area of engineering biology.
Bioinformatics 24, 1827–1836 (2008).
plugins and libraries (such as the Fiji Updater and ImgLib) into 19. Abramoff, M.D., Magalhaes, P.J. & Ram, S.J. Image processing with
ImageJ2, forming a basis of a long-term partnership that will com- ImageJ. Biophotonics International 11, 36–42 (2004).
bine professional software engineering of the ImageJ2 core and 20. Carpenter, A.E. et al. CellProfiler: image analysis software for identifying
and quantifying cell phenotypes. Genome Biol. 7, R100 (2006).
biological application–driven plugin development in Fiji. 21. Peng, H., Ruan, Z., Long, F., Simpson, J.H. & Myers, E.W. V3D enables
real-time 3D visualization and quantitative analysis of large-scale
Note: Supplementary information is available in the online version of the paper.
biological image data sets. Nat. Biotechnol. 28, 348–353 (2010).
22. de Chaumont, F., Dallongeville, S. & Olivo-Marin, J.-C. ICY: a new open-
Acknowledgments
source community image processing software in. IEEE Int. Symp. on
We thank W. Rasband for developing ImageJ and helping thousands of scientists,
Biomedical Imaging 234–237 (2011).
those who contributed to the Fiji movement by financing and organizing the
23. Berthold, M.R. et al. KNIME: the Konstanz Information Miner. in Studies in
hackathons, namely G.M. Rubin for hackathons at Janelia Farm, I. Baines for
Classification, Data Analysis, and Knowledge Organization (GfKL 2007) 319–326
hackathons at Max Planck Institute of Molecular Cell Biology and Genetics in
(Springer, 2007).
Dresden, R. Douglas for a hackathon at Institute of Neuroinformatics in Zurich,
24. Cardona, A. et al. TrakEM2 software for neural circuit reconstruction.
F. Peri and K. Miura for the hackathon at European Molecular Biology Laboratory,
PLoS One (in the press).
and International Neuroinformatics Coordinating Facility for Fiji image-processing
25. Schmid, B., Schindelin, J., Cardona, A., Longair, M. & Heisenberg, M. A
school, W. Pereanu for the confocal image of the larval fly brain, M. Sarov for
high-level 3D visualization API for Java and ImageJ. BMC Bioinformatics 11,
© 2012 Nature America, Inc. All rights reserved.

the confocal scan of C. elegans larva, the scientists who released their code
274 (2010).
under open-source licenses and made the Fiji project possible. We want to thank
26. Preibisch, S., Saalfeld, S., Schindelin, J. & Tomancak, P. Software for
Carl Zeiss Microimaging for access to the SPIM demonstrator. K.E. and C.R. were
bead-based registration of selective plane illumination microscopy data.
supported by US National Institutes of Health grant RC2GM092519. J.S. and
Nat. Methods 7, 418–419 (2010).
P.T. were funded by Human Frontier Science Program Young Investigator grant
27. Preibisch, S., Tomancak, P. & Saalfeld, S. in Proc. ImageJ User and
RGY0083. P.T. was supported by The European Research Council Community′s
Developer Conf. 1, 72–76 (2010).
Seventh Framework Programme (FP7/2007-2013) grant agreement 260746.
28. Matas, J., Chum, O., Urban, M. & Pajdla, T. Robust wide baseline stereo from
maximally stable extremal regions. Image Vis. Comput. 22, 761–767 (2004).
COMPETING FINANCIAL INTERESTS 29. Ibanez, L., Schroeder, W., Ng, L. & Cates, J. The ITK Software Guide
The authors declare no competing financial interests. (Kitware Inc., 2005).
30. Köthe, U. Reusable software in computer vision. in Handbook of Computer
Published online at http://www.nature.com/doifinder/10.1038/nmeth.2019. Vision and Applications Vol. 3 (eds. Jähne, B., Haussecker, H. & Geissler P.)
Reprints and permissions information is available online at http://www.nature. 103–132 (San Diego: Academic Press, 1999).
com/reprints/index.html. 31. Preibisch, S., Saalfeld, S. & Tomancak, P. Globally optimal stitching of tiled
3D microscopic image acquisitions. Bioinformatics 25, 1463–1465 (2009).
32. Cardona, A. et al. An integrated micro- and macroarchitectural analysis of
1. Turing, A.M. The chemical basis of morphogenesis. 1953. Bull. Math. Biol. 52, the Drosophila brain by computer-assisted serial section electron
153–197, discussion 119–152 (1990). microscopy. PLoS Biol. 8, e1000502 (2010).
2. Altschul, S.F., Gish, W., Miller, W., Myers, E.W. & Lipman, D.J. Basic local 33. Bock, D.D. et al. Network anatomy and in vivo physiology of visual
alignment search tool. J. Mol. Biol. 215, 403–410 (1990). cortical neurons. Nature 471, 177–182 (2011).
3. Myers, E.W. et al. A whole-genome assembly of Drosophila. Science 287, 34. Kaynig, V., Fischer, B., Müller, E. & Buhmann, J.M. Fully automatic
npg

2196–2204 (2000). stitching and distortion correction of transmission electron microscope

4. Neumann, B. et al. Phenotypic profiling of the human genome by time- images. J. Struct. Biol. 171, 163–173 (2010).
lapse microscopy reveals cell division genes. Nature 464, 721–727 35. Saalfeld, S., Fetter, R., Cardona, A. & Tomancak, P. Elastic volume
(2010). reconstruction from series of ultrathin microscopy sections. Nature Methods
5. Collinet, C. et al. Systems survey of endocytosis by multiparametric image advance online publication, doi:10.1038/nmeth.2072 (10 June 2012).
analysis. Nature 464, 243–249 (2010). 36. Arganda-Carreras, I. et al. Consistent and elastic registration of
6. Shariff, A., Kangas, J., Coelho, L.P., Quinn, S. & Murphy, R.F. Automated histological sections using vector-spline regularization. in Lecture Notes in
image analysis for high-content screening and analysis. J. Biomol. Screen. 15, Computer Science 4241, 85–95 (Springer, 2006).
726–734 (2010). 37. Lowe, D.G. Distinctive image features from scale-invariant keypoints.
7. Megason, S.G. & Fraser, S.E. Imaging in systems biology. Cell 130, Int. J. Comput. Vis. 60, 91–110 (2004).
784–795 (2007). 38. Sethian, J.A. Level Set Methods and Fast Marching Methods: Evolving
8. Keller, P.J., Schmidt, A.D., Wittbrodt, J. & Stelzer, E.H. Reconstruction of Interfaces in Computational Geometry, Fluid Mechanics, Computer Vision,
zebrafish early embryonic development by scanned light sheet microscopy. and Materials Science (Cambridge University Press, 1999).
Science 322, 1065–1069 (2008). 39. Kass, M., Witkin, A. & Terzopoulos, D. Snakes: active contour models.
9. Fowlkes, C.C. et al. A quantitative spatiotemporal atlas of gene expression Int. J. Comput. Vis. 1, 321–331 (1988).
in the Drosophila blastoderm. Cell 133, 364–374 (2008). 40. Hall, M. et al. The WEKA data mining software: an update. SIGKDD Explor. 11,
10. Anderson, J.R. et al. A computational framework for ultrastructural 10–18 (2009).
mapping of neural circuitry. PLoS Biol. 7, e1000074 (2009). 41. Kaynig, V., Fuchs, T.J. & Buhmann, J.M. Geometrical consistent 3D tracing
11. Saalfeld, S., Cardona, A., Hartenstein, V. & Tomancak, P. As-rigid-as- of neuronal processes in ssTEM data. Med. Image. Comput. Comput. Assist.
possible mosaicking and serial section registration of large ssTEM Interv. 13, 209–216 (2010).
datasets. Bioinformatics 26, i57–i63 (2010). 42. Cardona, A. et al. Identifying neuronal lineages of Drosophila by sequence
12. Murray, J.I. et al. Automated analysis of embryonic gene expression with analysis of axon tracts. J. Neurosci. 30, 7538–7553 (2010).
cellular resolution in C. elegans. Nat. Methods 5, 703–709 (2008). 43. Longair, M.H., Baker, D.A. & Armstrong, J.D. Simple Neurite Tracer: open
13. Fernandez, R. et al. Imaging plant growth in 4D: robust tissue source software for reconstruction, visualization and analysis of neuronal
reconstruction and lineaging at cell resolution. Nat. Methods 7, 547–553 processes. Bioinformatics 27, 2453–2454 (2011).
(2010). 44. Edelstein, A., Amodaj, N., Hoover, K., Vale, R. & Stuurman, N. Computer
14. Bao, Z. et al. Automated cell lineage tracing in Caenorhabditis elegans. control of microscopes using µManager. in Current Protocols in Molecular
Proc. Natl. Acad. Sci. USA 103, 2707–2712 (2006). Biology (John Wiley & Sons, Inc., 2010).

682  |  VOL.9  NO.7  |  JULY 2012  |  nature methods