search areas, from nanotechnology to as-tronomy.

It is common practice for manufactur-ers of image acquisition devices to includededicated image processing software, butthese programs are usually not very flex-ible and/or do not allow more compleximage manipulations. Image processingprograms also are available by themselves.

ImageJ holds a unique position because

T he advances of the medical and bi-ological sciences over recent years,and the growing importance of de-

termining the relationships between struc-ture and function, have made imaging anincreasingly important discipline. Theubiquitousness of digital technology —from banal digital cameras to highly spe-cific micro-CT scanners — has made im-ages an essential part of a number of re-

As the popularity of the ImageJ open-source, Java-based imaging program grows, its capabilities increase,

too. It is now being used for imaging applications ranging from skin analysis to neuroscience.

by Dr. Michael D. Abràmoff, University of Iowa Hospitals and Clinics; Dr. Paulo J. Magalhães,University of Padua; and Dr. Sunanda J. Ram, Louisiana State University Health Sciences Center

Image Processing with ImageJ

it not only is in the public domain (mean-ing that its source code is openly avail-able and its use is license-free), but alsoruns on any operating system. It is at-tractive because it is easy to use, can per-form a full set of imaging manipulationsand has a huge and knowledgeable usercommunity.

The program is the brainchild of WayneRasband of the Research Services Branch,National Institute of Mental Health, inBethesda, Md. It is called ImageJ becauseit is written in the Java language. Its firstrelease, version 0.50, was Sept. 23, 1997,and it is now in version 1.31, released inFebruary.

According to the institute, it has beendownloaded from its Web site tens of

IMAGING SOFTWARE

Figure 1. A scintillation rendering ofthe 3-D motion field of a fat-musclephantom was combined with asurface rendering to show motionand anatomy of the phantom. Themotion data were obtained as 1.5-T MRI gradient echo T1weighted image sequences, whilethe phantom was rotated in place at5° per frame. The motion field is toodense for motion and surface to berendered in the same place. The axis of rotation was orientedperpendicular to the surface of thephantom. The hue of the color codesfor direction, and the saturation ofthe color codes for motionmagnitude. Obtained with ImageJrunning VolumeJ for volumerendering and FlowJ for 3-D opticalflow estimation. Provided byMichael Abràmoff.

Reprinted from the July 2004 issue of Biophotonics International © Laurin Publishing Co. Inc.

A Laurin Publication

Photonic Solutions for Biotechnology and Medicine

Obviously, with a freeprogram, there is no tele-phone hot line for support.However, a large user basecommunicates through amailing list. This way, anyuser is free to ask ques-tions, put forward sugges-tions or ideas for newimaging functions andpublish solutions. At thetime of this writing, thiscommunity consisted ofmore than 1000 users/scientists (the majority)and users/developers (insmaller numbers, but offundamental importance).

Indeed, the know-howof the community is re-markably high becausemany members possess adetailed knowledge of thesoftware and the imagingproblems that the programcan address efficiently. Arequest on the mailing listis usually all it takes forsomeone, somewhere in

the world, to provide the required help.For example, Dr. Rex Couture of the

department of radiology at WashingtonUniversity School of Medicine in St. Louisran into a problem with reading largemicro-CT images. These files were about5 GB, but ImageJ couldn’t read beyondthe 4-GB limit at that time. “I posted aquery about this,” Couture said. “Withinan hour or two, someone workingthrough the night in Europe had foundthe problem. By 9:30 the next morning,I had a solution to try from Wayne. Sincehe didn’t have an image file that large, hecouldn’t test it, and it didn’t quite work.Twenty-four hours later he had created atest image, and I had a fully debuggednew version that solved the problem.”Shortly thereafter, someone in the dis-cussion group also posted a way to openlarge tiff files.

There has always been a somewhat terseusage instruction on the ImageJ Web site,but getting advice on a specific functionrequired taking the quite intimidatingstep of asking the mailing list. Now onehas the option of using a manual recentlyproduced by Tony Collins at the Wright

thousands of times, withthe current rate being about24,000 downloads permonth. Rasband said that,after working for 10 yearson NIH Image, ImageJ’sprecursor, he didn’t see a bright future for it, eventhough Scion Corp. ofFrederick, Md., had portedit from a Macintosh-onlyversion to the PC/Windowsplatform. However, theScion version wasn’t opensource, and he is a strongproponent of open sourceand platform indepen-dence. So he began writingImageJ in early 1997 afterhe had become intriguedby the new Java program-ming language.

Imaging capabilitiesImageJ can read most

of the common and im-portant formats used inthe field of biomedicalimaging (see table). If afile format is not currently supported,someone from the international user/developer community usually developssupport within days. In addition,ImageJ can be used to acquire imagesdirectly from scanners, cameras andother video sources, including camerasthat are compatible with TWAIN andFireWire, and frame grabber boardsfrom Cooke, National Instruments andPixelSmart.

The program supports all commonimage manipulations, including readingand writing of image files, and operationson individual pixels, image regions, wholeimages and volumes. Volumes, calledstacks in ImageJ, are ordered sequencesof images that can be operated upon as awhole. It can perform basic operationssuch as convolution, edge detection,Fourier transform, histogram and parti-cle analyses (including sophisticated sta-tistical processing of groups of particles),editing and color manipulation; and moresophisticated operations such as dilation,erosion and closing of structures, andmathematical operations on sets of im-ages, such as multiplication, exclusive or,

and division. In addition, visualizationoperations, including color space con-versions — for example, converting fromRGB to Hue Saturation Intensity colorspace, two- and three-dimensional plot-ting — and surface and volume render-ing, are supported. It also offers core sup-port for analyzing electrophoretic gels.

Cross-platformOne of the strong points of ImageJ is

its ability to run on different platforms.Statistics covering the last three months,with more than 80,000 downloads, in-dicate that it is being used mostly withMicrosoft operating systems (80 percent),followed by Macintosh platforms (16 per-cent) and Linux (4 percent). Althoughthese numbers are estimates and can bemisleading because someone could down-load it to one platform and use it on an-other, it is interesting to note the rela-tively high number of Macintosh users(the worldwide market share was less than2 percent in 2003, according to marketresearch company IDC), supporting thelong-held view that this type of computerattracts large sections of academia.

IMAGING SOFTWARE

Figure 2. This screen shot shows thresholding, edge detection,particle analysis with particles indicated, and histogram analysis onthe gray-scale diatoms image (upper left).

Cell Imaging Facility at Toronto WesternResearch Institute. This free, very thor-ough manual has many examples and il-lustrations. It emphasizes microscopy andaccompanies a collection of microscopy-related plug-ins. Although many open-source programs lack an extensive usermanual, even novices can find most ofthe information for their imaging needsin this manual.

Extensions: Macros and plug-insThe program is virtually limitless be-

cause of the availability of user-writtenmacros and plug-ins.

Macros are meant to make it easier toautomate oft-repeated tasks, which wouldbe tedious to implement manually.ImageJ has an easy-to-use macro-languagethat means that knowledge of Java isn’trequired for writing simple scripts. Forexample, a macro can be written that ac-quires an image every 10 seconds andstores it in a sequence (Figure 3).

Plug-ins are external programs, mostlywritten in the Java language, that offerimage processing capabilities that do not

exist in core capabilities of ImageJ. Onceimplemented, they cannot be distin-guished from the program itself. A smallcottage industry has sprouted from users/developers who are designing plug-insfor their own use and sharing them withall users. Plug-ins have brought ImageJfrom an image processing program to aframework that scientists can use to de-velop their own imaging solutions.

Plug-ins range from very small andstraightforward, such as the Grid plug-in— which simply draws a grid on an image— to complex, practically stand-aloneimage programs, such as the dendrite-tracing tool NeuronJ1 or the surface andvolume-rendering plug-in VolumeJ.2 Infact, the standard imaging capabilitiesthat ImageJ comes with are implementedas plug-ins as well. Developing new plug-ins requires knowing Java language and isthus not for everyone.

However, after using ImageJ for a while,users often need a specific solution fortheir problem that goes beyond its core ca-pabilities. The Medientechnik und designgroup in Austria has written a tutorialspecifically as an introduction to writingplug-ins for ImageJ.

Most plug-ins that are judged by theirdevelopers to be of general use can bepublished on the ImageJ Web site, whichnow holds more than 150 plug-ins, writ-ten by 98 developers. Some developersmay instead make their plug-ins andmacros available through personal Websites; some of these are listed on theImageJ site.

Imaging libraryA third way of extending ImageJ, which

is used by only a small number of moretechnically advanced developers, is to useits imaging capabilities and plug-ins fromtheir own programs, so that, in technicalterms, they are using it as a library of imag-ing methods. This is called ImageJ’s ap-plication programmer’s interface (API),which has been extensively documentedso that it is clear how to use and carry out these methods. This way, several online image database servers have been

IMAGING SOFTWARE

Figure 3. This simple ImageJ macroacquires an image every 10 secondsand stores it in sequence.

TABLE Image formats supported by ImageJ as of June 2004

Format Read and write

Analyze (Mayo Clinic’s format) (plug-in)AVI uncompressed movies �Bio-Rad Z-series (plug-in)BMP �DICOM (uncompressed) read �DICOM (uncompressed) write (plug-in)DM3 (Gatan Corp. transmission electron microscopy format) (plug-in)FDF, VFF (Varian MRI imaging system/EVS 900 micro-CT system (plug-in) formats)

FITS (NASA format) readGIF (including animated) �ICO, CUR (MS Windows) (plug-in)IPLab (Scanalytics Corp. microscopy format) (plug-in)Jpeg �Jpeg EXIF digital camera header (plug-in)Leica SP multichannel (plug-in)LSM (Zeiss LSM 510 confocal microscopes) (plug-in)PCX (Zsoft Corp.) (plug-in)PDS (Planetary Data System) (plug-in)PGM (P2 and P5) readPICT (Apple Mac) (plug-in)PNG �PSD (Adobe Photoshop) (plug-in)QuickTime movies �Raw (any uncompressed file format not directly supported, �8-,16-, 24-bit RGB, 32-bit, small- and big-endian)

SIF (Andor Technology spectroscopy format) (plug-in)SPE format (Roper Scientific cameras) (plug-in)Sunraster (plug-in)Targa (plug-in)Text images (tab or comma separated) �Tiff uncompressed and zip compressed, both as �1-, 8-, 12-, 16-bit unsigned, 32-bit (real) or RGB color, single and multiple images (stacks)

XBM, XPM (X-Windows formats) (plug-in)XLS (Microsoft Excel) (plug-in)

for (i = 0; i < 5; i++) //5 frames�

next = getTime ()+ 10 * 1000; //10s between frames//Run plug-in that grabs from a Quicktime framegrabberrun (“QT Capture”, “grab”);while (getTime) () < next) wait (1); //until 10s are over.

�

Figure 5. This 3-D reconstruction of DAPI-stained (blue) nuclei in a rat spinalcord section after intraventricular DiI injection results in red-staining of cells in the ependymal cell layer of the central cord. Scale bar = 10 µm. Imageprocessed with ImageJ running VolumeJ and provided by Andrea Mothe andTony Collins, Toronto Western Research Institute.

developed, including one for ophthal-mologic telediagnosis.

The applications to which ImageJ hasbeen applied are astounding. Space lim-itations dictate that only a few salient ex-amples can be given here: Paulo Magal-hães and co-workers in the departmentof biomedical sciences at CNR Instituteof Neuroscience, University of Padua,Italy, are using it to study the dynamics ofintracellular calcium,3,4 while researchersat the Laboratory for Cellular Neu-robiology of the Swiss Federal Instituteof Technology in Lausanne and theBiomedical Imaging Group at ErasmusMC-University Medical Center Rotterdamin the Netherlands, are using the NeuronJplug-in, for the automated quantitativeanalysis of neurons and dendrites (Figure4).1 The NeuronJ plug-in is based on re-cently developed and validated algorithmsfor detecting and linking elongated imagestructures specifically for this purpose.

Andrea Mothe and co-workers at thedepartment of zoology at the Universityof Toronto at Scarborough, in collabora-tion with Collins from Toronto WesternResearch Institute, use ImageJ and theVolumeJ plug-in for the 3-D reconstruc-tion of the differential localization ofnerve cell gene expression (Figure 5).

Procter & Gamble in Cincinnati is usingImageJ as a framework for Visia, a systemthat carries out complexion analysis bycomparing skin texture analysis of a sub-ject’s facial skin against the texture analy-ses of other subjects stored in a database.The program uses the ImageJ applicationprogrammers interface (API) as a library(Figure 6).

The Biomedical Imaging Group of theSwiss Federal Institute of Technology is

IMAGING SOFTWARE

Figure 4. The main window in theback shows a fluorescencemicroscopy image (compiled fromseveral scans) of a neuronal cell andits outgrowth, together with theprimary (red), secondary (blue) andtertiary (magenta) tracings.Superimposed are the ImageJwindow (top left) with the NeuronJtoolbar (replacing the ImageJ toolbarduring operation) and two NeuronJresult windows (bottom right), givingthe length of individual tracings andstatistics on selections of tracings.Courtesy of Erik Meijering, ErasmusMC-University Medical CenterRotterdam.

using the program to teach image pro-cessing to engineering students,5,6 and theCenter for Image Processing in Education

of Tucson, Ariz., is using it for teachingbasic imaging to high school students.7

Dr. Michael D. Abràmoff and co-work-

IMAGING SOFTWARE

Figure 6. The Visia complexion analysis system uses ImageJ to analyze how theclient compares with a database of women in her age and skin-type groups forthe given categories — in this example, spots. It includes scores that indicate thepercentage of women who scored below the client’s evaluation for the category.Provided by Procter & Gamble.

Figure 7. The EyeCheck Web site provides online diabetic retinopathy screeningfrom retinal color photographs. The left area of the image shows the Web pagewith patient information. When this page is accessed, an ImageJ appletautomatically displays the retinal photographs (in this case, of a normal subject),at right. This allows for user interaction so that, for example, the intensity imagecan be displayed, as in the bottom part, or specific abnormalities can be marked.Courtesy of EyeCheck.

Figure 8. ImageJ is being used toexamine the second to fourth digitratio (index to ring finger) in theanole lizard, which is thought toreflect the relative concentration ofor sensitivity to androgens duringdevelopment. Provided by PeterHurd, University of Alberta.

ers at the departments of ophthalmologyand radiology at University HospitalUtrecht in the Netherlands, have usedImageJ and the FlowJ and VolumeJ plug-ins for the differentiation of orbital tu-mors and for measuring the motion ofsoft tissues in patients.2,8

Abràmoff also worked with other clin-icians to develop the EyeCheck Web site,an online diabetic retinopathy screeningproject in the Netherlands. It uses ImageJ’scapabilities for storing and displaying reti-nal images in a telediagnosis environ-ment. The program can accept and storea great variety of image formats, whichcan then be used for manual grading byophthalmologists or for semiautomatedor automated detection of diabeticretinopathy (Figure 7).

Finally, Dr. Peter Hurd at the depart-ment of psychology, University of Albertain Edmonton, Canada, is using ImageJ todetermine the second to fourth digit ratio(index to ring finger) in Anolis carolinen-sis, or anole lizard, which is thought toreflect the relative concentration of (orsensitivity to) androgens during devel-opment (Figure 8).

The program illustrates that imaging ison the boundary between being a fieldof science and a field of engineering. Thesolutions being proposed by and imple-mented in conjunction with users and de-velopers are sometimes engineering so-lutions (in that they are derived straight

from textbook material or a publication)and sometimes scientific solutions (whennecessity dictates a scientifically new ap-proach to solve a real-world problem).

A critical and pragmatic reader may ex-pect some drawbacks to ImageJ, and thereare some issues. The program requiresminimal computer knowledge for instal-lation and first steps, while commercialvendors may offer on-site installation andtraining. Also, because of the continuousstate of development, bugs and “undoc-umented features” can creep into the dis-tributed version. This can be a problem forthe unaware researcher who comparesdata acquired with the old and new ver-sions. Users usually spot these problems,and corrected versions of the program aremade available immediately.

There is a misconception that an imag-ing program written in Java cannot befast, and this may divert some potentialusers from ImageJ. Abràmoff used to sharethis worry and tested its validity by rewrit-ing some of his convolution routines(which he had originally implementedin Java) in C++ and calling these routines

from a plug-in using the ImageJ JavaNative Interface API. Although there was a decrease in processing time of about 30 percent, this did not weigh upto the increased development time for the routines.

Furthermore, there is no inherent rea-son why an algorithm coded in C++ willrun significantly faster than the same al-gorithm coded in Java. The executionspeed depends on how good a compileris at optimizing the generated code. Manypeople will even argue that the compil-ers that translate Java byte code into ma-chine code can do a better job of opti-mization because more information isavailable to them about the programbeing compiled and about the machinethat it is running on. So the reason whysome commercial image processing pro-

grams are appreciably faster than ImageJat convolutions or similar processor-in-tense operations is related to increasedsophistication of their algorithms. Theopen nature of ImageJ, however, may en-able end-users with inspiring insights todevelop better and more sophisticated al-gorithms, without the constraints of pro-prietary code.

In summary, ImageJ has attracted a var-ied and dedicated group of users becauseit is free and expandable, and can operateon any platform. It is especially remarkablehow robustly the framework, designed sixyears ago, has withstood the test of time.Though it is difficult to predict where theprogram will be five years from now, theevolution will probably be a very inter-esting and rewarding experience for bothusers and developers. G

IMAGING SOFTWARE

Figure 9. The SmartRoot system, aninteractive plant root image analysisplug-in, was developed by XavierDraye and co-workers at the Unitéd’Ecologie des Grandes Cultures,Université Catholique de Louvain,Louvain-la-Neuve, Belgium. The toolhelps the user track root objects,displays them in different “layers,”saves them in Extend MarkupLanguage documents and sendsmeasurements to any StandardizedQuery Language-compliantdatabases. Courtesy of Xavier Draye.

Chem 279, pp. 11521-9.4. L. Filippin et al (Oct. 3, 2003). J Biol Chem

278, pp. 39224-34.5. D. Sage and M. Unser (2001). Proc. of the

2001 IEEE International Conference on ImageProcessing, Thessaloniki, Greece 3, pp. 298-301.

6. D. Sage and M. Unser (2004). IEEE SignalProcessing Magazine 20, pp. 43-52.

7. J.V. Ekstrom (2000). The Science Teacher 67,pp. 53-55.

8. M.D. Abràmoff and M.A. Viergever (2002).IEEE Trans. Med. Imaging 21, pp. 296-304.

Meet the authorsDr. Michael D. Abràmoff, ophthalmologist

and image science engineer, is at the depart-ment of ophthalmology and visual sciences,University of Iowa Hospitals and Clinics, IowaCity, Iowa. He is the author of VolumeJ andmany other popular plug-ins; e-mail: michaelabramoff@uiowa.edu.

Paulo J. Magalhães is at the department ofbiomedical sciences, University of Padua inItaly. He has been a contributor of the popu-lar Ratio Plus plug-in.

Sunanda J. Ram is assistant professor ofmedicine at the Louisiana State UniversityHealth Sciences Center in Shreveport.

AcknowledgmentThe authors wish to thank Wayne Rasband,without whom neither ImageJ nor its com-munity would have existed.

Please send requests for additional copies ofthis article to biophotonics@laurin.com.

References1. E. Meijering et al (April 2004). Cytometry

58A, pp. 167-176.2. M.D. Abràmoff et al (February 2002). Invest

Ophthalmol. Vis. Sci. 43, pp. 300-307.3. M.F. Abad et al (March 19, 2004). J Biol

IMAGING SOFTWARE

Function and dynamics of cells and organelles• M. Ghosh et al (April 30, 2004). Cofilin promotes actinpolymerization and defines the direction of cell motility. SCI-ENCE, pp. 743-746.

Imaging calcium and signal transduction• T.J. Collins et al (April 2, 2002). Mitochondria are mor-phologically and functionally heterogeneous within cells. EMBOJ., pp. 1616-1627.• T.J. Collins and M.D. Bootman (June 2003). Mitochondriaare morphologically heterogeneous within cells. J. EXP. BIOL.,pp. 1993-2000. • J. Bruton et al (Aug. 15, 2003). Mitochondrial and my-oplasmic [Ca2+] in single fibres from mouse limb muscles dur-ing repeated tetanic contractions. J. PHYSIOL., pp. 179-190.• J.D. Bruton et al (Oct. 15, 2003). Mitochondrial functionin intact skeletal muscle fibres of creatine kinase deficient mice.J. PHYSIOL., pp. 393-402.

Dental imaging• R.A. Couture and C. Hildebolt (April 2000). Quantitativedental radiography with a new photostimulable phosphor sys-tem. ORAL SURG. ORAL MED. ORAL PATHOL. ORAL RA-DIOL. ENDOD., pp. 498-508.• R.A. Couture and C.F. Hildebolt (January 2002). Preciseimage-receptor calibration and monitoring of beam qualitywith a step wedge. DENTOMAXILLOFAC. RADIOL., pp. 56-62. • U. Meyer et al (April 2002). First experience with a public do-main computer-aided surgical system. BR. J. ORAL MAXILLO-FAC. SURG., pp. 96-104.

Tumor differentiation and soft-tissue motionmeasurement• M.D. Abràmoff et al (October 2000). MRI dynamic colormapping: a new quantitative technique for imaging soft tissuemotion in the orbit. INVEST. OPHTHALMOL. VIS. SCI., pp.3256-3260.• M.D. Abràmoff, W.J. Niessen and M.A. Viergever (October2000). Objective quantification of the motion of soft tissues inthe orbit. IEEE TRANS. MED. IMAGING, pp. 986-995.• M.D. Abràmoff et al (September 2001). Patients with per-sistent pain after enucleation studied by MRI dynamic colormapping and histopathology. INVEST. OPHTHALMOL. VIS.SCI., pp. 2188-2192.• M.D. Abràmoff and M.A. Viergever (April 2002).Computation and visualization of three-dimensional soft tis-sue motion in the orbit. IEEE TRANS. MED. IMAGING, pp.296-304.

Retinal image analysis• J. Staal et al (April 2004). Ridge-based vessel segmentationin color images of the retina. IEEE TRANS. MED. IMAGING, pp.501-509.• M.D. Abràmoff et al (2004). A spatial truncation approachto the analysis of optical imaging of the retina in humans andcats. Proc. IEEE International Symposium on Biomedical Imaging2004 2, pp. 1115-1118.

Brain and fat tissue imaging• B. Haelewyn et al (September 2003). Desflurane affordsgreater protection than halothane against focal cerebral is-chaemia in the rat. BR. J. ANAESTH., pp. 390-396.• B. Haelewyn et al (April 2004). Cardioprotective effects ofdesflurane: effect of timing and duration of administration inrat myocardium. BR. J. ANAESTH., pp. 552-557.• J.P. Sacha et al (May 2003). Quantification of regional fat vol-ume in rat MRI. Proc. SPIE Medical Imaging, pp. 289-297.

Neuroscience• G. Aakalu et al (May 2001). Dynamic visualization of localprotein synthesis in hippocampal neurons. NEURON, pp. 489-502.• S.J. Tang et al (Nov. 8, 2001). A role for a rat homolog ofstaufen in the transport of RNA to neuronal dendrites. NEURON,pp. 463-475.• M. Zonta et al (2003). Neuron-to-astrocyte signaling is cen-tral to the dynamic control of brain microcirculation. NAT.NEUROSCI., pp. 43-50.

Craniofacial surgery simulation • T. Stamm et al (December 2003). Validity of a three-di-mensional public-domain system for contemporary endodonticresearch. J. ENDOD., pp. 801-805.

Simulation of cell growth patterns • G. Landini and P.M. Iannaccone (April 2000). Modeling ofmosaic patterns in chimeric liver and adrenal cortex: algorith-mic organogenesis? FASEB J., pp. 823-827.

Telediagnosis and image servers• J.D. Parker and J.W. Wallis (October 2003). Systems forRemote Interpretation of Emergency Studies. SEMINARS INNUCLEAR MEDICINE XXXIII, pp. 324-330.• M.D. Abràmoff (2003). Impact of Security on Data Transferin Teleophthalmology for Diabetic Retinopathy Screening overthe Internet. Proc. Computer Aided Fundus Imaging and Analysis(CAFIA).

Publications in which ImageJ was used, listed by application

• rsb.info.nih.gov/ij — ImageJ and most of its plug-ins and macros are available for download.• www.uhnres.utoronto.ca/wcif/download.html — ImageJ manual by Tony Collins.• mtd.fh-hagenberg.at/depot/imaging/imagej — The Medientechnik und design group tutorial on writing plug-ins.• imagescience.bigr.nl/meijering/software/neuronj — NeuronJ is available for download.• bij.isi.uu.nl — Web site for VolumeJ and many other biomedical imaging plug-ins.

Web Resources