computerised handling pathology

76 Clin Pathol 1995;48:796-802

Computerised image handling in pathology

S-K Cheong, K Micklem, D Y Mason

IntroductionThree of the major branches of clinicalpathology-haematology, cytopathology andhistopathology-depend on microscopy. Micro-scopic images are often recorded as trans-parencies or as colour or black and white prints,and then used for teaching, publication andresearch. The methods and equipment usedfor the preparation ofphotographic images havechanged little in the past 30 years and serve theirpurpose well for most applications. However, aradically different approach to the gathering ofmicroscopic images is now available to thepathologist, based on their conversion to digit-ised images in a microcomputer. Systems forprocessing, storage and retrieval ofthese imageshave been expensive in the past, but the steadilyfalling cost of microcomputers and of highcapacity storage devices, and the availability ofsophisticated image processing software, nowmake systems for high quality digital imageprocessing, storage and retrieval a much moreattractive proposition for the pathology labora-tory.

In this review, we outline the considerableadvantages of this approach to image handling,which is summarised schematically in fig 1. Inthe appendix, we give details of the hardwareand software options which we have used overthe past three years in our own department.Many digital imaging packages are availablefrom scientific equipment manufacturers (typ-ically as options for microscopes), but the pur-pose of this article is to describe a systembuilt up of non-specialist components that arereadily available and somewhat cheaper thanan integrated system. There are many ad-vantages in building a system in this way, themost important ofwhich (after cost) is that onejoins a much larger user base, ensuring betteravailability and maintenance of software. Webelieve that by the close of this century thisnew technology will be very widely used in allbranches of clinical pathology, and we hope inthis review to indicate that it is much moreaccessible than appears at first sight.

UniversityDepartment ofCellular Science,John RadcliffeHospital,Headington,Oxford OX3 9DUS-K CheongK MicklemD Y Mason

Correspondence to:Dr D Y Mason.

Accepted for publication20 December 1994

Advantages of electronic imagRAPIDITY

In conventional photography images capturedon film require chemical development beforethe results can be visualised. "Instant" imagescan be obtained (using Polaroid sheet or trans-parency film), but they are relatively expensivecompared with other options, and may not be ofcomparable quality. In contrast, electronicallygenerated images are seen instantly (allowingunsatisfactory ones to be discarded), and print-outs can be obtained within a matter ofsecondsif needed.

EDITINGConventional photomicroscopy is a relativelyinflexible medium. Prints which are unsatis-factory in terms of contrast or colour valuesrequire reprinting. If multi-image numbered"collages", adorned with elements such as ar-rows and numbers, are required for publicationor display, many hours of tedious work may berequired with glue, scalpel and dry transfers.In contrast, electronic images can be almostinfinitely manipulated. Images can be resized,cropped and rotated in the computer so thatassembly of complex images is greatly fa-cilitated. Lettering and arrows can be added(fig 2A), and these can be kept in separate"masks", to be added when desired for printing,so that the original image remains intact. Col-our values can be coarsely or subtly adjusted(fig 2A) and, if nothing else, this can be usedto correct instantly differences in adjacent im-ages, or to correct for colour, temperature andexposure variations, a very common fault inpublished photomicrographs (as is evident fromthe pages of most pathology journals).

STORAGEIt is the authors' experience that photo-micrographs, despite one's best intentions,are often stored (to put things optimistically)in a haphazard fashion. Retrieval of slides canbe frustrating and time-consuming, and theyare subject to attrition due to loan, loss orphysical damage. Even if careful indexing isperformed, the final archive often becomes in-creasingly cumbersome, and may need to bekept at a site away from the laboratory, addingfurther to the inconvenience of finding a par-ticular image.

Electronically stored images, in contrast, canbe kept together on a suitable storage format(for example, optical disk) and, even if thematerial is kept elsewhere, it can be accessed onthe user's own microcomputer. As we outlinebelow, linkage ofmicrocomputers via networksmeans that images can be accessed almost asrapidly from another country as from the user'sown microcomputer. Retrieval is easy throughan indexed name search, a visual search ofminiature representations of the stored images,or a combination of both.

FLEXIBILITY OF OUTPUTPhotomicroscopy conventionally yields colourreversal transparencies (for projection), orprints on black and white or colour paper.Electronic images can also be produced in theseformats (fig 1), but their digital nature meansthat they can also, without modification, bedisplayed on the computer itself or on a TV

796

on February 27, 2022 by guest. P

rotected by copyright.http://jcp.bm

j.com/

J Clin P

athol: first published as 10.1136/jcp.48.9.796 on 1 Septem

ber 1995. Dow

nloaded from

http://jcp.bmj.com/

Computenised image handling in pathology

Input -*

Output -*

Illustration,e.g. in journal

/ Negative film

Copies Other Transparency Conventional Overheadon disk computer print transparency

Colour Video TVprint projector monitor

Figure 1 Schematic illustration of the many options for the acquisition, manipulation and output of digitised images.

monitor, projected onto a screen via a videoprojector, or exported for recording on videotape. Copies can be distributed on disk andthe next few years will see the increasing use

of compact disks (CD-ROM) as a means ofdistributing images. These disks have very highcapacity and, provided they are of a suitableformat, such as PhotoCD, can be displayed,using inexpensive equipment, on domestic TVmonitors. Laser disks (designed originally as

an alternative to video tape for home viewingof feature films) represent another TV com-

patible, high capacity storage medium, but our

impression is that, at least in Europe, little usehas so far been made of this technology forstoring and viewing still images.

Digitised images also lend themselves to timebased display in sequence, an ideal mediumfor teaching. These "slide shows" can be in-teractive-that is, manipulated by selectingscreen buttons or a via a keypad. For thosewho like that sort of thing, they can even beanimated (for example, adorned with flyingarrows, etc.) to increase their instructive value(and revive flagging viewer interest).

RequirementsThe different ways in which digitised imagescan be fed to a computer and then output ondifferent media are summarised in fig 1.

INPUT

Images can be digitised directly from the mi-croscope, or scanned in digital format from an

image already prepared through conventionalphotography (for example, a print or trans-parency). There are two parameters of res-

olution of the digital image: the number ofpixels (the number of sampled points com-

prising the image) and the bit depth of eachpixel (the maximum number ofcolours or tonesavailable). In the systems we describe here thedimensions of the images are typically about700 pixels wide by 500 pixels high, this beingroughly the resolution of a 14" computer mon-itor. Each pixel is described by 8 bits for eachred, green and blue signal giving a total of some16 million colour variations. Whilst both theseresolutions are relatively low compared withsome integrated instruments and somewhat lessthan what is regarded as a minimum' they are

adequate for most applications.

Direct video captureWhen we began work in this field, we fearedthat video resolution images would be ofinsufficient quality as the resolution is so

much lower than photography (approximately4500 x 3000 pixels2). However, we have foundthat high quality video cameras are available(see appendix), which give images which are

fully acceptable for most applications.

K Microscopeslide

\a%\

Electro-phoretic

gel

Videocamera

Videotape

797



j.com/

J Clin P


ber 1995. Dow

nloaded from

http://jcp.bmj.com/

Cheong, Micklem, Mason

Peri heral Blood...3wv. ALs ..

ii

x .

N, XM.:_-..~

.:.ibS

"d w

.-M

Anti-B cell"'

T--NoenaT celf MR. -

Anti-T cell e~~~~~~~~~~~~~~~~~~~..

t An -)¢01'.

Figure 2 Illustrations prepared by digitised image manipulation compared with those created by conventional photography. (A) Transparencies wereconverted to digitised images (on PhotoCD disks) and then assembled into this collage. Minor changes were made (using Adobe Photoshop), includingmatching of colour values, cropping, and (in the picture of the patient) replacement of the background. Lettering and arrows were created in a separatemask, making it possible to add these to this print, while leaving the collage intact (rather as one can put lettering on a transparent sheet to overlay aconventional photograph). Conventionally applied lettering on photomicrographs may be difficult to read when it coincides with dark areas in the printand in this digitised collage we have placed "burnt out" boxes (with feathered edges) behind the lettering (most easily seen behind the words "PeripheralBlood" and "Tissue Biopsy").

798

AUM,!W'4ceV.-II.I.. .:- qw...

i.



j.com/

J Clin P


ber 1995. Dow

nloaded from

http://jcp.bmj.com/


** *¢SZ^:9,0~~~~~~O"

AP,0 sS*** **i

0' 04

Figure 2 (B) Three of the transparencies used to createthe collage opposite are reproduced here withoutmodification, as they would normally be handled by thisjournal. Comparison with the collage shows that some ofthese (notably the illustration ofB and T cell staining inthe blood smear) were greatly improved by digitised imagemanipulation.

A video camera mounted on the microscopefeeds a video signal to the computer. In thecomputer a video digitiser "grabber" boardconverts the signal into digitised information.There are various types of board available thatcan be slotted into either a Macintosh (Apple)or an IBM PC compatible computer. Videoimages generated by video cassette recordersor video disk players can also be grabbed anddigitised.

ScanningPrinted images-for example, a photographicprint or an illustration from a journal, can

Mass storage devices and their approximate cost

Cost (in pounds sterling)

Device Capacity Drive Media

Removable hard disk 44 megabytes 100 30CD-ROM 650 megabytes 230 120PhotoCD 100 images 230 80Magneto-optical disk 128-2000 megabytes 600-2000 20-200Digital audio tape 1200 megabytes 1000 50

be acquired using a flat bed scanner. Thisresembles a conventional photocopier and ver-sions are also available which will "read" trans-parencies. The Kodak PhotoCD format is alsoofinterest here, because it allows transparenciesto be digitised at very high resolution and storedon CDs. These are suitable for viewing on amicrocomputer equipped with a CD drive, butcan also be displayed on a TV screen from asuitable CD player.

IMAGE HANDLINGHardwareA typical microscopic colour image will requireabout 1-2 megabytes when digitised. Manip-ulation of such information obviously re-quires a high speed processor (if the impatientuser is not to be left waiting for long periodsin front of a motionless screen). Details aregiven in the appendix of recommended micro-computer specifications.

SoftwareThere are many scientific image processing andanalysis programs available, but we prefer thewidely used Adobe Photoshop program (seeappendix). We do not attempt any analysis ofthe images.

IMAGE STORAGEImage data files contain much more in-formation than other types-for example, textfiles used for word processing, and so presenta problem for storage. A single colour image isapproximately equivalent to one high density,double sided floppy disk, and a high capacityhard disk within the computer for initial storageis mandatory. However, space on this in-builtdisk is rapidly consumed and other high cap-acity mass storage devices, preferably portable,are essential for backup and archiving purposes.The different options are summarised in thetable and discussed in the appendix.The problems of establishing an effective

system for storing and retrieving images arenot trivial, but a detailed discussion is outsidethe scope of this review. There are, however,essentially two options. One involves using oneofthe many general purpose personal computerdatabases, entering information on each imageunder categories such as tissue, disease andkeywords, and incorporating details of whereeach image is located (for example, onPhotoCD, on hard disk, etc.). Some data-bases-for example, Filemaker Pro and Hyper-card permit the creation of reduced "thumb-nail" copies of images, which can helpduring retrieval.An alternative is to use a dedicated image

archiving program. The demands of marketssuch as the graphic art world mean that heretoo there is a substantial choice, including theFetch, Pick Bank, Image Browser, and Cu-mulus programs. We currently use the Shoeboxprogram, which was developed by Kodak spe-cifically for PhotoCD archiving.

799



j.com/

J Clin P


ber 1995. Dow

nloaded from

http://jcp.bmj.com/


OUTPUTOnce an image has been captured, it is tem-porarily stored on the computer hard disk, andmodification or enhancement may be per-formed at this stage. Multiple images may becombined to form a single composite illus-tration. The image may now be output in a

number of formats (see appendix and fig 1),the choice depending on the final use for theimage.

Film recordersThe image may be sent to a film recorder toyield conventional 35 mm transparencies forprojection. These can also be used to producecolour prints on Cibachrome paper. Al-ternatively, the images can be recorded on

colour negative film and prints then madecheaply on colour printing paper. The qualityof such prints is lower than that of dye-sub-limation prints so that they could not be usedin preparing large collages for publication butthey are adequate for many other applications-for example, posters.

Printers

The image can be sent to a conventional laserprinter but the result is rarely of an adequatestandard for purposes other than checking lay-out and obtaining a rough indication ofthe finalappearance. Colour prints can be produced byprinters which rely (in increasing order of cost)on inkjet, thermal transfer or dye-sublimationtechnology. Colour photocopiers are availablewhich can also accept digital input and printimages. Dye-sublimation printers yield printsessentially indistinguishable from conventionalphotographic prints. Their high cost (1C7-12 000) means that normally one would use a

print bureau. Images can also be printed as

colour separation films from an image setter tomake plates for offset printing and, althoughwe have no experience of this, it is likely to beused increasingly by publishers in the future.

Applications-now and the futureDigital image processing finds many ap-

plications in laboratory practice.

TEACHING

High quality digital images can be integratedinto computer assisted teaching programs. Fur-thermore, images saved in FilmStrip formatcan be used to make animated sequences.

These, and automated slide shows, make teach-ing/learning sessions more stimulating.

In the future, it is hoped that digitised equi-valents of atlases of pathology will appear, as

these contain pictures of great value in teachingand training. It is clear that publishers will

increasingly publish in electronic form, and itis probable that much more copious illus-tration-for example, multiple colour pic-tures rather than one or two back and whiteillustrations for each condition, will be possible.It is also likely that many pathology laboratories

with responsibility for teaching will becometheir own publishers via the creation of digitalimage databases which can be accessed locallyor globally through computer networks such asthe Internet. Many computers around theworld are already connected via the Internet,and some readers of this article may alreadyhave found how easily one can down-load im-ages (for example, of a satellite land imageshowing the next weather system to reach theseisles) from distant databases. The ability todistribute images in this way, based on clinicalpathology photomicrographs, would be of ob-vious value for teaching.

RESEARCHWriting grant proposals, preparing reports, giv-ing presentations and writing papers for pub-lication are part and parcel of the researcher'slife today. The approach described in this re-view finds application in all these areas. Illus-trations to accompany written progress reports,slides for presentations, and records ofthe results of experiments (for example,microscopic images, electrophoretic gels) arebut some of the examples. We believe that inmany instances the quality of this material canmake a crucial difference between success andfailure.

ROUTINE SERVICESPathologists daily translate pictures into words,whether for reporting haematology, cyto-pathology or histopathology samples, for de-scribing an electrophoresis result, or for re-cording a stab wound. In practice it is im-practical to archive digitally every image of thissort (because of cost), but with the increasingtrend towards keeping medical records in mul-timedia format, representative pathology im-ages may appear more widely in patients'records in the future.Once digital images are accessible on com-

puter networks, one can hold image basedconferences for consultation on difficult cases,and this can be done on a global scale.

ProblemsAny pathologist embarking on digitised imagehandling is immediately faced by the difficultquestion of what computer to use. The choiceis essentially between a Macintosh and one ofthe many IBM PC compatible instruments.Our own preference, based on factors such asspeed and the wide choice of off the shelfgraphics programs, is for Macintoshes, but itshould be noted that the disk formats used byMacintosh and IBM clones for high capacityremovable storage media (for example, mag-neto-optical disks) are not readily compatible.However, small images can be transferred be-tween the two types of computer on floppydisks and all images are exchangeable via net-works. For example, the Photoshop programwhen run on a Macintosh offers the user morethan a dozen different formats for reading andsaving files, covering both standard formats

800



j.com/

J Clin P


ber 1995. Dow

nloaded from

http://jcp.bmj.com/


(for example, GIG, JIF, TIFF, and bitmap)and proprietary formats (for example, AmigaIFF, PIXAR, Kodak PhotoCD, and Targa).With versatile software that permits easy

image editing, the question arises as to howextensively one is justified in modifying anillustration. A number of our colleagues, onseeing this technology for the first time (and inparticular the remarkable manipulation cap-abilities of the Photoshop program) smile dis-approvingly (or furtively) and say "So you couldmove the band on the gel" or "You don't needantibodies- you can just paint the cells in thecomputer". These fears (or hopes), which werethe subject of a recent article in Science,3 implythat conventional photographic reproduction isnear perfect, whereas computer based imagingtechniques belong in the realm of the coun-terfeiter. However, it is evident on a moment'sreflection that a photograph, whether it showsan electrophoretic gel or a tissue section, is ahighly artificial rendering of the original image.It is dependent on the histological or proteinstain used, and on the photographer's choiceof film paper and developing method. It maybe non-representative either because an expertphotographer has obediently enhanced or"lost" elements in the picture, or because anunskilled photographer has failed to capturethe image well. In our experience computerprocessed images usually provide a more ac-curate rendition ofthe original than the averagephotograph, and it is often the old technologyrather than the new that is producing deceptiveresults.Another potentially controversial topic con-

cerns the use of images prepared in anotherlaboratory. This becomes increasingly relevantas remote image transmission and exchangeis used more widely. Plagiarism may causeproblems, particularly if some kind of un-justified modification or enhancement is made,and ethical and legal issues are posed whichrequire careful consideration if users are toavoid infringing copyright. However, it is pos-sible to exaggerate problems of this sort andthey should not become an obstacle to the wideuse of the new technology.

ConclusionDesktop digital image processing has now comeofage and is within the grasp ofmany pathologylaboratories in terms of affordability and sim-plicity of operation. The authors have foundthe facility immensely useful in the setting ofour research and teaching department and be-lieve that its wider use will, in addition tobenefiting the laboratory itself, also improveoutside perceptions of clinical pathology andraise the profile of this medical specialty.

This work was supported by the Leukaemia Research Fund.

1 Jeffreys RA. Film and electronics-a marriage ofconvenience.J7Audiov Media Med 1988;11:49-51.

2 Kilbourne S. A primer on digital imaging-post productionfor still photography. J Biol Photogr 1991;59:43-8.

3 Anderson C. Easy-to-alter digital images raise fears of tam-pering. Science 1994;263:317-18.

4 Brown S. Digital imaging in clinical photography. J AudiovMedia Med 1994;17:53-66.

AppendixIn this section we give technical data on thehardware and software options of which wehave personal experience. However, this is arapidly growing field and we cannot pretendto have carried out a "Best Buy" survey of allavailable options. The one rule to follow is toevaluate any system at first hand, ideally inone's own laboratory, before making a finalchoice.

INPUTMicroscopyWe have used a JVC KY-F30 colour camera(which contains three CCD chips) for threeyears and obtained excellent results. Single chipcameras, however, are substantially cheaperalthough they produce slightly lower resolutionimages. A new generation of cameras providingdigital signal processing within the camera hasappeared and they provide even better results.Any high quality video digitiser board ac-

cepting RGB signals is suitable. We have usedthe Kingfisher and Falcon boards (GraphicsUnlimited, Cambridge, UK) because themanufacturer, the designer and the softwareauthors are locally available.

Flat bed scannersMany scanners currently available in the marketyield high quality images, and selection is verymuch a matter of personal preference. Colourscanners are desirable, although a grey scalescanner may be used for monochrome images(electron micrographs, electrophoretic gels).For best results images on x ray film must beacquired with a scanner fitted with a trans-illuminator, or by video grabbing from acamera.

Kodak PhotoCDIn 1990 Kodak introduced a new approach tothe storage and display of 35 mm photographicimages, based on scanning (at a range of differ-ent resolutions) and writing them to CDs.We are in the process of transferring many ofthe transparencies which we have acquired inthe past to this format. The scanner and CDwriter which records these images are expensiveso that film has to be despatched to a processingcentre, which will charge approximately C50for a disk holding 100 images. Images storedin this format can then be viewed on a con-ventional television screen using a PhotoCDplayer, or on a computer monitor from a CD-ROM drive. Each CD has a catalogue facility,including miniature "thumbnail" images tofacilitate viewing and retrieval.

IMAGE HANDLINGWe prefer to use Apple Macintosh computers,because they are very well suited to imageprocessing and have a user friendly operating

801



j.com/

J Clin P


ber 1995. Dow

nloaded from

http://jcp.bmj.com/


system. We typically use a Macintosh Quadra700 with 20 megabytes ofRAM (a figure whichrepresents a minimal requirement).For processing, we use Adobe Photoshop

(Adobe Systems Inc., Mountain View, Cali-fornia, USA). This is a pixel based image ed-iting program which endows, among manyfeatures, the ability to crop images, to rotatethem, to adjust their intensity, contrast andcolour values, and to make any alteration re-quired in their size. Most importantly the pro-gram is modular, allowing the addition of "PlugIns" to acquire images directly from input de-vices such as video grabbers and scanners. Thisallows us to use just one program for all imageprocessing rather than the software providedwith the different devices. Modules are alsoavailable to read a variety of image formats,including compressed images from KodakPhotoCDs; to write in non-Apple formats-forexample, CompuServe GIF (used widely on theInternet); and for specific "filtering" operationswhich modify an image.

IMAGE STORAGERemovable hard diskSyQuest 44 disks are relatively fast and can

store 44 megabytes. Whilst these now representtechnology developed some years ago, they are

still the most widely used method of trans-ferring data, especially to desktop publishingbureaux.

Magneto-optical diskThese are available with capacities ranging from128 to 1200 megabytes. As they can store a

high volume of data and are small and robust,they represent our choice at present for mediumterm storage.

Compact disk-read only memory (CD-ROM)These disks are identical to the compact diskswidely used for musical recording. With a stor-age capacity of 650 megabytes, each disk can

store 100 to 6000 images (depending on res-

olution). CD-ROM drives are relatively in-expensive and provide the best option for longterm storage of image data. CD recorders are

becoming cheaper, making this a very goodmethod of archiving.

Digital audio tapeThis is an inexpensive approach to image ar-

chiving. Images are stored sequentially andretrieved through an indexed name list. Theprocess is considerably slower than disk basedforms of storage, so that it is not recommendedother than for archiving.

Image compressionThough mass storage media such as magneto-optical disks have come down in price, the everexpanding nature of any image database meansthat the cost of storage may pose problems.Some kind of file compression strategy is de-sirable to reduce images from their workingformat to a more manageable size. For a fullframe 24 bit colour image, the industrial stand-ard for image compression is JPEG (Joint Pho-tographic Expert Group). This method canreduce the file size by a factor of 20:1 withlittle effect on appearance.4 Not only does thissave space, but image transmission times overcomputer networks are also reduced. Com-pression should be used with care, as successivesteps of compression and decompression forviewing cause image degradation.

OUTPUTFile formatsThe image may be converted and saved in avariety of different formats depending on use.We routinely save images in the native ApplePICT format for transfer to other applications.TIFF format is used for transfer to IBM PCcompatible computers. Postscript formats arenot suited to large bitmapped images andshould be avoided except for desktop pub-lishing.

Film recordersThere are a wide range of 35 mm film recordersavailable and at present we use an Agfa Slide-writer. We have found that 4000 line resolutiongives distinctly better results than 2000 lineresolution, suggesting that the former may bea useful minimum standard. Higher resolutionrecorders will only provide better results whenrecording onto film formats larger than 35 mm.

Dye-sublimation printersThere are only a few printers available at thislevel. We have found that bureaux equippedwith the Kodak XLT7720 printer provide thebest results, even though its specified resolution(203 dots per inch) is lower than some otherprinters. This is the means by which fig 2Awas prepared for printing.

FURTHER ILLUSTRATIONSAdditional examples of digitised images fromthe authors' department can be accessed viathe World Wide Web Internet, site URL http://phoenix.jr2.ox.ac.uk. A set of computer gen-erated transparencies (illustrating the REALlymphoma classification) are available from theauthors (e-mail:[email protected]; fax: 01865 63272).

802



j.com/

J Clin P


ber 1995. Dow

nloaded from

http://jcp.bmj.com/

computerised handling pathology

Documents