Special article

Analytic morphology in clinical and experimentaldermatologyJosef Smolle, MD,a Wilhelm Stolz, MD,b Friedrich A. Bahmer, MD,c Stefan el-Gammal,MD,d Georg Heinisch, MD,e Torsten Mattfeldt, MD,f Martin Nilles, MD,g Friedrich Otto,MD,h Ralf U. Peter, MD,b Hans P. Sayer, MD,a Norbert Sepp, MD,i and Thomas Vogt,MDb Graz and Innsbruck, Austria; and Munich, Homburg/Saar, Bochum, Dresden,Giessen, Hornheide, and Heidelberg, Germany

During the past several years, quantitative morphology has gained increasing attention in di­agnostic pathology and in certain research applications. In the field of dermatopathology,quantitative morph ology has been applied to numerous problems, ranging from the interac­tive measurement of nuclear contours to fully autom ated, high-resolution image analysis ofultrastructural micrographs. Dermatologic applications are reviewed,and potential develop­ments in the future are briefly outlined. (J AM A CAD DERMATOL [993;29:86-97.)

Analytic morphology denotes a set of techniquesthat providequantitative and statistically evaluabledata from morphologicimages. In most quantitativetechniques.measurements are interactively or auto­matically performed, and the results are expressedon a linear scale.

Depending on the method used, different hard­ware systems are required. For basic stereology, anocular grid or a ruler may be sufficient, but high­resolutionimage analysis and computer simulationsneed sophisticated computer technology with acomputational speed markedly exceeding the capa­bilities of a conventional personal computer.

The challenge of quantitative morphology is todescribe image features that are beyond the percep­tion of the human eye. In dermatology, with itssound morphologicbackground, there are numerouspotential applications for quantitative techniques. Abroad range of them is reviewed herein.

TECHNOLOGICAL ASPECTS

In analyticmorphology, many different terms areused. Some are synonymous, whereas others de­scribe well-defined specialized techniques. In gen-

From the Departments of Dermatolog y, Graz," Munlch." Homburg!Saar," Bochurn,'! Dresden," Giessen.s Homheide," and Inns bruck,'and th e Department of Pathol ogy, H eidelberg,"

Reprint requests: J. Smolle, Depart ment of Dermatology, Universit y ofGraz, Auenbruggcrplatz 8, A-8036 Graz, Austria.

Copyright (~) 1993 by the American Academy of Dermatology, Inc.

0190-9622/93$1.00+.10 16/1 /4565]

86

eral, analytic morphology and quantitative mor­phology can be regarded as synonyms and includeall techniques that provide quantitative (or semi­quantitative or binary) data from morphologic im­ages. The quantitativefeaturescan begainedbyob­jective measurements based on different hardwaredevices. Morphometry includes all techniques forthe measurement of two-dimensional features fromstructural elements or particles by geometric crite­ria, be it by pointcounting,measuringdistanceswitha ruler, or tracing contours on a digitizer board.Stereo logy is a body of methods that derive esti­mates referringto the three-dimensional spacebasedon measurements obtained in fewer dimensions.Image analysis includes all methods that deal withcompletely digitized images in computer systems;this includes digital processing and enhancement ofimagesbefore the measurements. These proceduresrequire a substantial degree of hardware and soft­ware. Three-dimensional reconstruction and imagemodelingby computer simulationsmay be regardedas specialized image analysis methods.

STEREOLOGIC METHODS FORDERMATOPATHOLOGY

Histologic slides provide flat images from spatialstructures. Particles (cells, nuclei) appear as flat ar­eas; spatial surfaces (basement membranes, vesselwalls)appear as flat curves;spatial curves (collagen,fibers) appear as points. Moreover, in orientedtissues like skin and muscle, the properties of the

Journal of the American Academy of DermatologyVolume 29, Number I Smolle et al. 87

Fig. 1. Estimation of tumor cell nuclei from vertical section through a nodular malignantmelanoma. Paraffin section. Linear intercepts through all those tumor cell nuclear profilesthat have been hit by the evenly spaced points are measured. (Hematoxylin-eosin stain;X555.)

microscopic image depend crucially on the directionof cutting. To draw meaningful conclusions about abiologic tissue from measurements on histologicslides, all these facts must be taken into consider­ation. Fortunately, many mathematical relationsbetween the original structure and its image areknown, and these are often surprisingly simple. Ste­reology is the method that allows extrapolation frommeasurements on sections to real three-dimensionalstructures.' In skin sections many important stereo­logic quantities can be determined from verticalsections.? Thus various planar parameters can bemeasured in routine skin specimens and unbiasedestimates of three-dimensional quantities obtainedeven in retrospective studies. Volume fractions (per­centage of tissue volume occupied by a particularstructure) can easily be determined by point count­ing. For the estimation of surface areas (e.g., totalsurface of all nuclei per tissue VOlume), so-called cy­cloid test systems are necessary as a special measur­ing device. For the calculation of the size of cell nu­clei, the method of the "volume-weighted mean nu­clear volume" can be applied. This procedurerequires a set of parallel lines with evenly spacedreference points. The parallel lines must yield ran­domly selected, weighted angles with the verticalaxis. The measurements can then be obtained by

rulers, or the whole procedure can be implementedin an image analysis system. In the context of der­matopathology, the volume-weighted average ofnuclear volume (vv) has been studied extensively inpigmented tumors of the skin, particularly in mel­anocytic nevi, lentigo malign a, and malignant mel­anomal" (Fig. 1). The results of multivariate anal­ysishave shown that vvis in fact an independent in­dicator of prognosis.

DNA ANALYSIS OF MICROSCOPIC SLIDES

Nuclear DNA measurements can be performedwith a cytophotometer (DNA cytophotometry), animage analysis system (DNA image cytometry), ora flow cytometer. Although image cytometry andimage cytophotometry are more time-consumingthan flow cytometry, these methods have the ad­vantage that each individual measurement can beinteractively controlled and that the information onDNA content can be combined with morphologicdata of the particular cell. Furthermore, DNAmeasurements on histologic sections do not requirethe preparation of single cell suspensions, which isoften difficult to perform.

In contrast to pure morphometric methods, DNAanalysis directly reflects the functional status oftheparticular cell with regard to ploidy and cell cycle

88 Smolle et al.

phase. Whereas most normal tissues are composedof a majority of diploid cells in the Go or G] phaseof the cell cycle, along with a small proportion of te­traploid cells in the G2 phase and some cells betweenthe diploid and tetraploid values (S phase), hyper­proliferative tissues show an increase of Sand G2

phase cells. Particularly in malignant tumors, cellclones with DNA values other than diploid or tetra­ploid may occur (aneuploidy).

Image cytometry has repeatedly been applied tofacilitate a differential diagnosis between benign andmalignant melanocytic lesions. Coefficient of varia­tion of DNA contentf mean optical density, nucleararea,? and differences in these features between thesuperficial and deep parts of the lesions'' proved toprovide sensitive discrimination between commonmelanocytic nevi and Spitz nevi, on the one hand,and of malignant melanoma on the other. Stolz etaI.9 found the mean value of ONA content, standarddeviation of nuclear area, and 95th percentile ofDNA distribution as the most efficient criteria. Aneffective discrimination between benign and malig­nant melanocytic lesions could also be obtained byquantifying not only nuclear size and density, butalso chromatin distribution and architecture byhigh-resolution image and multivariate analysis,even when only intraepidermal melanocytic nucleiwere taken into account.l? Efficiency was found tobe lOO% at the ultrastructural level!' and 96% inlight microscopic semithin sections.'?

The results concerning prognosis of malignantmelanomas have been less encouraging because tu­mor thickness usually turned out to be more signif­icant than DNA content.P However, significantcorrelations between tumor thickness, on the onehand, and mean nuclear area, 80th percentile ofDNA distribution and the number of nuclei with arelative DNA value higher than 5c (relative DNAvalue compared with chicken erythrocytes as DNAstandard), on the other, could be demonstrated in 34cases of malignant melanoma. 14

In premalignant melanocytic lesions, a relationbetween morphologic features of atypia and of ab­normalities in the DNA histograms could be dern­onstrated.P

QUANTITATIVE NUCLEAR FEATURES INMALIGNANT CUTANEOUS LYMPHOMASAND PSEUDOLYMPHOMAS

The differential diagnosis between malignant cu­taneous lymphomas and pseudolymphomas (e.g.,lymphocytic infiltration of Jessner-Kanof, different

Journal of the American Academy of DermatologyJuly 1993

types of cutaneous lymphoid hyperplasia, actinic re­ticuloid, and lymphomatoid papulosis) can be diffi­cult. Early stages of cutaneous T-ceIllymphomas(mycosis fungoides, Sezary syndrome) often resem­ble benign inflammatory dermatoses such as chroniccontact dermatitis, atopic dermatitis, and psoriasis,both clinically and histologically. There is evidencethat quantitative morphology may be helpful fordifferential diagnosis.

Cutaneous malignant T-cell lymphomas are char­acterized by lymphoid cells with deep cerebriformindentations in their nucleus. Qualitatively similarcells, however, may also occur in chronic eczema, li­chen planus, contact dermatitis, and peripheralblood of healthy donors. Nevertheless, detailedultrastructural analysis of the nuclear contour index(NCI; defined as the ratio of the nuclear perimeterand the square root of the nuclear area) by a graphictablet interfaced with a small computer revealedstriking differences between the cerebriform nucleiof malignant lymphoid infiltrates and reactive in­flammatory dermatoses.l'' with higher NCI valuesin the malignant cases. These differences were alsofound in blood cells.17 The assessment of nuclear in­dentations revealed a sensitivity of 62% with a spec­ificity of 100%.18

Because most neoplasms are characterized by anincreased number of polyploid and aneuploid cells,the DNA content of cutaneous lymphomas andpseudolymphomas was investigated by image anal­ysis. It was shown that the majority of the cells inpseudolymphoma are diploid and have relative DNAvalues of 2c. In contrast, malignant lymphomasshowed a variable increase of cells with DNA valuesin the tetraploid region. In addition, an increasedrate of cells in the region between 2.25c and 3.5c andcellswith DNA values higher than 5c were observed.With the 2c deviation index (2cDI), which reflectsthe increased variation of the DNA values aroundthe normal value of 2c in malignant lymphomas, adifferentiation between pseudolymphomas and ma­lignant lymphomas was achieved with a sensitivityand specificity of 95%. Further insights into the cy­tology ofcutaneous lymphoproliferative diseases canbe expected from analysis of the chromatin struc­ture'? and the combined application of DNA cy­tometryand immunolabcling.P

NUCLEOLAR ORGANIZER REGIONS

An interesting, simple, and more common addi­tional method for the histologic evaluation of tumorsis the determination of nucleolar organizer regions

Journal of the Ame rican Academy of DermatologyVolume 29, Number I Smolle et al. 89

Fig. 2. N ucleolar organizer regions (NORs) in melanocytic nevus. Most cells contain oneor two small NORs per nucleus. (Silver staining techniq ue; X400.)

Fig. 3. Nucl eolar organizer regions (N ORs) in malignant melanoma. Most cells showmultiple large NORs. (Silver staining techniq ue; X400.)

(N ORs). NORs are segments of DNA that containcoding genes for ribosomal RNA and contribute tothe regulation of cellular protein synthesis. Theirassociation with argyrophilic proteins makes it pos­sible to visualize NORs in conventional histologicsections by a modification of a silver staining tech­nique.2l-23 Morphologic changes in NORs are signsof rapidly dividing cells with high metabolic activity.Usually NORs are larger a nd more numerous inmalignant than in benign cells, as has been demon-

str ated in tumors of the breast , endometrium, brain,kidney, prostate, colon and rectum, lymphomas andothers.21,23,24

In derm atopathology NORs have been investi­gated mainly in melanocytic lesions.22, 23, 25-28 Sig­nificantly higher NOR rates have been found inmelanomas than in melanocytic nevi-? (Figs. 2 and3). Because melanomas and melanocytic nevi alsodiffer with respect to the Immunohistologic assess­ment of actively cycling cells,30 the NOR results in

I

90 Smolfe et al.

Number of Cells per Charlnel

DNA Index: 1.00

5 Phase Fraction: 1.1-'_

..'Jet 14M I

Channel Number(Relative DNA Conlenl)

Number of Cells per Channel._-------------,

DNA Index; 1.76

S Phase Fraclion: 15.1-'.

III

....I.

I" .. '- I_

I G1 S G2·t.1

, Tumor Cell'PopulO-tion I

Diploid C.lIs

Fig. 4. Top, Flow cytometric DNA histogram of nor­malhuman skin.TheSand G2+M phase region hasbeenenlarged 8-fold (dotted line) to become visible. Bottom,Flow cytometric DNA histogram of melanoma metasta­sisshows smallpeakofnormal diploid cells at channell00and a population of aneuploid tumor cells (DNA in­dex » 1.76) with increased S phase fraction (15.1%).

melanocytic lesionsmay, in fact, indicate differencesin proliferative activity.Recent studies indicate thatmorphometric analysis of NOR areas (e.g., NORarea/nucleus) may be of greater value than simpleNOR counting.'?

Besidesmelanocytic skin tumors, NOR determi­nation also shows significant differences betweenkeratoacanthoma and squamous cell carcinoma,"although the diagnostic accuracy remains to be de­termined. Remarkable NOR findings have beendescribed in basal cell carcinomas. These tumors

Journal of the American Academy of DermatologyJuly 1993

seem to have more NORs than other (benign ormalignant) epidermal tumors.F This may be be­cause basal cells of the normal epidermis possessmore NORs than cells of the stratum spinosum.P

FLOW CYTOMETRY

Likeother cytometricmethods,flowcytometry indermatology is mainly aimed at the diagnostic andprognostic characterization of skin tumors. Flowcytometricmeasurementsofthe cellularDNA con­tent provide three types of information (Fig. 4):- The ploidy state, i.e., the existence or not of

abnormal (aneuploid) cell lines- The DNA index(DI), i.e.,the mean cellularDNA

content of a cell populationrelative to the diploid(euploid) DNA value

_ The portionsof cells in the cell cycle phases,par­ticularly the S phase fraction as an indirect mea­sure of proliferative activityThe flowcytometric histogram of normal human

skin (Fig. 4, top) shows a predominant narrow peakrepresenting the cells in the G, phase of the cellcy­cle. The portions of cells in the Sand 02+M phaseare small, reflectingthe relatively slowbut continu­ous proliferation of epidermal cells. The S phasefraction is normally 1% to 2%. In contrast, Fig. 4(bottom) showsthe histogramof a tumor sample (amelanoma metastasis). In addition to a peak ofdip­loid cells from normal epidermal cells or lympho­cytes in the specimen that can be usedas an internalstandard to define the position of the diploid value,the histogram exhibits a population of aneuploid(DNA index == 1.76), fast-proliferating (8 phasefraction == 15.1%) tumor cells. The clinical useful­ness of these flow cytornetric data has been provedin many malignancies.P: 34 The main advantage offlow cytometry is the ability to measure large quan­tities ofcellsin a short time. However, problemsmayarise by preparing singlecell suspensions from solidtumor tissues.

In primary malignant melanoma the existence ofaneuploid tumor cells is correlated with tumorthickness and is associated with an unfavorableprognosis. Aneuploid tumors with hypertetraploidcell lines and/or more than one aneuploid cel1line(so-called muIticlonal tumors) exhibit an increasedrisk of recurrence and metastasis.P S-phase frac­tions above 15% indicate a lowsurvival probability.In the metastatic stage of the disease sequentialmeasurements showthat an increasedgenetic insta-

Journal of the American Academy of Derm atologyVolume 29, Number 1

bility of the tumor as detected by changes in theploidy pattern is associated with short patient sur­vival time.36

In squamous cellcarcinomas of the skin aneuploidcell lines and particularly multiclonal cell popula­tions are found more frequently in advanced stages.The proliferative activity is higher in aneuploid tu­mors than in diploid ones. Thus ploidy state and Sphase fraction can be used as prognostic criteria inaddition to clinical and histopathologic findings.

Measurements of the cellular DNA and proteincontent'? or immunocytologic surface markers mayalso yield additional information in certain tumorsystems.

QUANTITATIVE IMMUNOHISTOLOGY

Immunohistology facilitates the selectivestainingof defined molecules or antigenic epitopes in histo­logic sections.Although the intensity of the stainingreaction is usually not stoichiomet rically correlatedwith the amount of antigen present, the relative dif­ferences in staining intensity are considered to rep­resent differences in the amount of antigen, if thestaining procedures are kept identical. Immunohis­tology has been used to characterize certain celltypes in histologicslides, but also for the detection ofparticular cellular activities (e.g., the expression ofoncogene products).

There are two basic approaches for quantifyingimmunohistologic slides by image analysis. Thenumber, size, and distribution of stained structurescan be assessedeither by manual, interactive, or au­tomated image analysis. Second, the intensity of thereaction products can be quantified by densitometrywith automated image analysis systems.

The analysis of natural killer (NK) cells inmalignant skin tumors may serve as an example forquantifying positively stained structures. NK cellsare cytotoxic to certain tumor cell clones in vitro andare considered to playa role in host-tumor interac­tion in vivo. However, most studies implicating hostresistance to tumor growth by NK cells have beenperformed with NK cells isolated from peripheralblood mononuclear cells. In vitro, the experimentshave been carried out with an NK cell/tumor cellratio of 1:I up to 100:1. However, stereologicestimation ofthe amount ofNK cellsand tumor cellsin melanomas-" revealed that only a minority of in­filtrating cells belonged to the NK cell population,whereas most of the inflammatory cells were lym-

Smolle et at. 91

phocytes and macrophages. The NK cell/tumor cellratio in vivo was found to be about 103 times lowerthan that used experimentally in vitro. On the basisof these quantitative immunohistologic results it be­came obviousthat the results obtained in vitro maybe highly artificial and cannot be related to the ac­tual situation in vivo.

Immunoelectron microscopy has been used toenhance the accuracy of morphometric measure­ments in the diagnosis of cutaneous T-cell lym­phoma. Limiting the assessment of the nuclear con­tour index to infiltrating cells expressing a pan-T cellmarker excludes potential errors arising from histi­ocyte-like cells with a cerebriform nucleus.P

The other approach to quantitativ.e immunohis­tology is to measure the staining intensity of the im­munohistologic reaction product. 8-100 protein isknown to be expressed in melanocytic skin lesions.However, there is no qualitative difference betweenbenign and malignant tumors with regard to S-100protein expression. With automated image analysis,the staining intensity of the immunohistologic reac­tion to S-100 protein was assessed in benign mel­anocytic neviand in malignant rnelanomas.t '' It wasfound that the overall staining intensity did not dif­fer between benign and malignant lesions. However,malignant lesions were characterized by an in­creased staining intensity in the depth of the lesion,which was not the case in benign nevi.

PROLIFERATION IN MELANOCYTIC SKINTUMORS

A useful candidate for quantitative immunohis­tology in tumors is the monoclonal antibody Ki-67that enables the determination of the proliferativepool (G}, S, G2, M phase) of a given tissue compo­nent on frozen sections in situ .3D

In a recent study in 145 melanocytic skin tumorsmean numeric density (NY mean) and maximumnumeric density (NY max) of Ki-67-positive nuclei(number per cubic millimeter) were quantitativelyevaluated by interactive image analysis. There wasa strong correlation between mean and maximumnumeric density (r =0.815; P < 0.05) indicatingthat each of the two factors may be regarded as rep­resentative of proliferative activity. Maximum nu­meric densities in melanocytic nevi, primary malig­nant melanoma, and metastatic melanoma revealedhighly significant differences. In primary malignantmelanoma mean and maximum numeric density

92 Smolle et (/1.Journal of the American Academy of Dermatology

July 1993

Maximum Ki 67 density (1000 nuclel/mms)300,--------------------------,

260

200

150

100

50

....

...:.:....-.. .: ..'. . -. .. . ... --..... ..'

::C .. - ."::IiIiI :: . ..," : I •

nevus mm rom mm mm metapT1 pT2 pT3 pT4

Fig. 5. Maximum numeric density of Ki-67-positive nuclei in melanocytic skin lesions(N == 145). T he d ifferent lanes refer 10 be nign nevi, primar y malignant melanoma with tu­mer thickness <0.75 mm (pT l), primary malignant melanom a with tum or thicknessbetween0.76 an d 1.50 mm (pT2), prim ary malignant melanom a with tumor thickness between 1.51and 3.50 mm (pT3), primary malignant melanoma with tumor thi ckness ;:0: 3.51 mm (pT4),and metastatic malignant melanoma (meta).

values showed a noticeable variation ranging fromlow values as observed in benign nevi to high valuesas observed in metastatic melanoma with a consid­erable overlap between the diagnostic groups (F ig.S). Within the group of primary malignant melano­mas, there was a significant correlation betweenproliferative activity (NY max) and maximal tumorthickness according to the Breslow or Clark level ofinvasion and mitotic rate.

A prospective short-term evaluation of the pa­t ients with primary malignant melanoma showed asignificantly shorter disease-free interval in patientswith high NV max. Furthermore, all five patientswho died of malignant melanoma were in the groupwith a high NV max.

The use of quantitative microscopic systems ob­viously enables more precise data from Ki-67­immunostained preparations to be derived.t! Fur­ther studies will reveal whether the estimation of themaximum numeric density of Ki-67-positive nucle irepresents a superior prognostic indicator comparedwith the well-est ablished Breslow index.

IN SITU HYBRIDIZATION

Similar to immunohistology, image analysis mayalso be of value in objectively defining the reactionproduct in in situ DNA or RNA hybridization.

Global differences inoptical density between labeledand control regions after in situ hybridization wereevaluated by Uhl et al.42 More precisely, imageanalysis was applied at the cellular level to quantifythe grains either interactively" or automaticallytaking background labeling also into account." Re­liability and objecti vity of this method were evalu­ated by comparing the values of grains per cell ob­tained by conventional and automatic techniquesafter in situ hybridization with a -collagen DNAprobes in various specimens.v Correlation coeffi­cients were 0.97 for fibroblasts embedded into athree-dimensional collagen gel, 0.94 for dermalfibroblasts in skin obtained from a patient with pro­gressive systemic scleroderma, and 0.90 for fibro­blasts in a 2-week-old scar." Because the automaticanal ysis technique allows a more rapid and reliablequantitative evaluation of in situ hybridization, sim­ilar procedures are also being applied in variousnondermatologic research projects.46, 47

MORPHOMETRY IN EXPERIMENTALDERMATOLOGY IN VIVO

All aforementioned studies are largely observa­tional. However, quantitative morphology can alsobe of particular value in experimental investigations .Besides in vitro application of morphometric meth-

Journal of the American Academy of DermatologyVolume 29, Number I

ods, many studies have been performed in vivowithimage processing to investigate the morphology ofcells48 and tissues (above all vascular network49.52).

As in other applications, adequate sampling, testingof reproducibility, and critical interpretation aremandatory.

Quantification of cells or their morphologicchanges is one of the great advantages of imageanalysis procedures in comparison to visual assess­ment. For example, morphometric methods demon­strated by image processing that only contact sensi­tizers, but not tolerogens (DNTB) or nonsensitizers(DCNB), induced migration of Langerhans cells todraining lymph nodes in mice (manuscript in prep­aration). Therefore important questions not only inclinical dermatology but also in experimental der­matology concerning immunology and oncologythat need quantitative assessment might be an­swered by morphometry.

Quantitative morphology may also serve as a bi­ologic alternative to classical physical dosimetry inthe study of radiation damage. 53 The skin seems tobe a preferential organ for the development of bio­logic indicators because it is constituted of prolifer­ating cells and is always involved if a patient is ex­posed to ionizing radiation. Short-term organ cul­tures of human skin were irradiated in vitro withvarying doses (1 to 6 Gy, x-rays 240 kV) and keptin culture for different time periods after irradiation(6 to 72 hours). At different time intervals, the spec­imens were fixed, processed, and analyzed with afluorescent microscope connected to an image-ana­lyzing device. Whereas no significant change in nu­clear area of the basal cells was found after irradi­ation with 1 Gy, a remarkable reduction of the nu­clear area was detected after single-dose irradiationwith 6 Gy. In addition, the individual response toidentical radiation doses at defined time intervalsappeared to vary considerably. Thus quantitativemorphology may be a valuable adjunct in assessingthe individual biologiceffect of ionizing radiation onhuman skin.

QUANTITATIVE MORPHOLOGY IN VITRO

Numerous aspects of dermatologic research arecarried out in vitro. In many cases, nonmorphologicquantification (biochemical, radiochemical, immu­nochemical assays) is sufficient. In some instances(e.g., tumor cell invasion) morphologic quantifica­tion has to be performed.

Smolie et al. 93

Fig. 6. Confrontation culture of multicellular mela­noma spheroid (right) with fragment of embryonic chickheart (left) after 72 hours. Heart fragment becomesinvaded and partially disintegrated by melanoma cells.(Anti-chick heart antiserum; Xl 00.)

To elucidate the mechanisms involved in tumorcell invasion and to search for antiinvasivestrategies,the embryonic chick heart model has been intro­duced.P" Tumor cells are cultured as multicellularspheroids of about 200~m in diameter and are sub­sequently confronted with fragments of embryonicchick heart, which serve as a stroma analog. After1 to 7 days, confrontation pairs are harvested, sec­tioned, and stained immunohistochemically for dis­tinction of the tumor and the stroma component. Forquantification, immunohistologically stained sec­tions are fed into an image analyzing computer (Fig.6). A program has been developed that facilitates themeasurement of nine quantitative features simulta­neously referring to tumor and stroma proliferationand to several aspects of invasion.55, 56 These fea­tures are created by algorithms of image operationsbased on mathematical morphology. The statisticalevaluation program finally compares various exper­iments and provides a full text interpretation of theresults, thereby providing an objective base for theevaluation of potentially antiinvasive strategies.

THREE-DIMENSIONAL RECONSTRUCTIONSIN DERMATOLOGY

Besides measuring specified features insingle im­ages, comparing several different images of the same

94 Smolle et al.Journal of the American Academy of Dermatology

July 1993

Fig. 7. Three-dimensional reconstruction. Examples of reconstruction of mitochondrionfrom electron micrographs (A). hair follicle structure in keratosis pilaris derived from lightmicroscopy (B), senile angioma studied with 20 MHzsonography (0, and normal human hairfollicle reconstructed from high-frequency ultrasound B-scansections of skin (D).

structure may provide qualitative and quantitativenew information beyond the capability of immediatevisual examination. This is particularly true whenthe three-dimensional morphology of particularstructures has to be derived from two-dimensionalsections. In the past, the structures had to be copiedsection by section on wax or polystyrol plates, which

were aligned to form a section pile. Finally this pilewas trimmed to the structure contour, a task thatwas very labor-intensive.

Computer-aided techniques can reduce consider­ably the time necessary [or such a three-dimensionalreconstruction, especially when the interface be­tween the user and the computer has been well de-

Journal of the American Academy of DermatologyVolume29, Number I

..-.

,~ . ..

•,

i9. .;-

• e ..~

;'_.. .- a " , /1\,) ~... ~ ~ ,.,. .

i!~ 1I~.. ~.{..

e ... ~

!~ .. <l •.. .. ~ (,0.

~ . .. ~ . &

Smolle et (If. 95

·1l) ..

Fig. 8. Tissue modeling . Example of tumor pattern simu lated by computer shows morpho­logic result of moderate degrees of tumor cell proliferation and motility, expansive growthproperty and strong tumo r cell cohesion.

signed.Such reconstruction programs may facilitatedifferent viewing modes for every structure, real­time rotation, scaling, realistic perspective surfacemodeling with a virtual light source to improve in­depth illusion, and volume and surface calcula­tions.57-59

For data input, light or electron micrographs,traced on a digitizer board, automated discrimi­nated structures inon-lineimagesorevenultrasoundimages can be used. For display, a line-contourmodel, a wire-frame model,or a surface-area modelcan be chosen. Combining these different presenta­tion modes, structures withinstructures can easilybestudied.

Fig. 7 shows examples of the reconstruction of amitochondrion from electron micrographs (Fig. 7,A), the hair follicle structure in keratosis pilaris de­rived from light microscopy (Fig. 7, B), a senilean­gioma studied with 20 MHz sonography (F ig. 7, C),and a normalhuman hair follicle reconstructedfromhigh frequency ultrasound B-scan sections of theskin (Fig. 7, D).

In additionto three-dimensionalreconstructionasjust described, the measurement of features corre­spondingto the real three-dimensional tissuemay beof substantial value. Tumor volume measurementsobtained from serial sections of cutaneous melano-

mas proved superior to thickness as a prognostic in­dicator.f

TISSUE MODELING AND TISSUEINTERPRETATION

Although the descriptive analysis of tumor mor­phology provides useful hints for diagnostic andprognostic assessment.v'- 62 the relation betweenbi­ologic propertiesof the tumor cells and the evolvingmorphologic patterns is not yet clear. To elucidatethis relation, tissue interpretation through tissuemodeling has been Introduced.f'

The method is based on the development of amathematical model of the dynamic process under­lying a particular morphologic situation, for exam­ple,the process of tumor cell growth, motility,decay,and stroma interaction in the case of morphologictumor patterns. After formulation and implemen­tation of the model, a large reference set of simula­tions is created, with virtually all possible variationsof the simulation factors. From a comparisonof thesimulation factors, which are set at the beginning ofeach simulation, with the resulting morphologicpattern, it can be learned in which way specific bi­ologicproperties influence the morphologic appear­ance64-66 (Fig. 8). Thus, for example, the importanceof both tumor cell proliferation and motility for

96 Smolle et al.

morphologic patterning can be demonstrated. Theassumptions of the simulation model are supportedby observations of melanocytic skin tumors'< 66, 6i

and of animal experimental systems.P As soon asthe relation between biologic properties (i.e., simu­lation factors) and morphology has been quantita­tively established in computer simulations, this in­formation can be used to estimate these biologicproperties from static histologic images of realtumors. In melanocytic skin tumors, biologic tumorcell properties estimated by this procedure proved tohave prognostic significance.v' Thus tissue modelingby computer simulation may lead to tissue interpre­tation beyond the capabilities of pure heuristicdescription.

REFERENCES

1. Mattfeldt T. Stereologische Methoden in der Pathologie.Stuttgart: Thieme, 1990.

2. Baddeley AJ, Gundersen HJG, Cruz-Orive LM. Estima­tion of surface area from vertical sections. J Microsc1986;142:259-76.

3. Howard V. Rapid nuclearvolume estimation inmalignantmelanoma using point-sampled intercepts in vertical sec­tions:a reviewof the unbiased quantitative analysis of par­ticles of arbitrary shape. In: Mary JY, Rigaut JP, eds.Quantitative image analysis in cancer cytologyand histol­ogy. Amsterdam: Elsevier, 1986:245-56.

4. Bruengger A, Cruz-Orive LM. Nuclear morphometry ofnodular malignant melanomas and benign nevocytic nevi.Arch Dermatol Res 1987;279:412-4.

5. Soerensen FB. Objective histopathologic grading of cuta­neous malignant melanomas by stereologic estimation ofnuclear volume. Cancer 1989;63:1784-98.

6. Schmiegelow P, Schroiff R, Breitbart E, et al. Malignantmelanoma-its precursorsand its topography of prolifera­tion: DNA-Feulgen cytophotometry and mitosis index.VirchowsArch [AJ Pathol Anat 1986;409:47-59.

7. Lindholm C, Bjelkenkrantz K, Hofer P. DNA-cytopho­tometry of benign compound and intradermal nevi,Spitzepithelioidnevi and malignant melanomas. Virchows Ar­chiv [B] Cell Pathol 1987;53:257-8.

8. LeBoitPE, VanFletcher H. A comparativestudy of Spitznevus and nodular malignant melanomausing imageanal­ysis cytometry. J Invest Dermatol 1987;88:753-7.

9. Hempfer S, Stolz W, BinglerP, et al. Prognosticanddiag­nosticvalue of DNA cytometry in imprintsof melanocyticlesions [Abstract]. Analyt Cell Pathol 1989;9:319.

10. Abmayr W, Stolz W, Korherr S, et al.Chromatin textureof melanocytic nuclei: correlation between light and elec­tron microscopy. Appl Optics 1987;26:3343-8.

II. Stolz W, Abmayr W, SchmoeckelC, et a1. Ultrastructuraldiscriminationbetweenmalignant melanomas and benignnevocytic neviusinghigh-resolution imageand multivariateanalyses. J Invest Dermatol 1991 ;97:903-1 O.

12. Abmayr W, StolzW, PilletL, eta1.Differentiation betweenmalignant melanomaand melanocyticnevi using high-res­olution image analysis [Abstract]. Analyt Quant CytolHistoI14:245.

13. Heenan PJ, Jarvis LR, Armstrong BK.Nuclear indices and

Journal of the American Academy of DermatologyJuly 1993

survival in cutaneous melanoma. Am J Dermatopathol1989; II :308-12.

14. HempferS, StolzW, Bingler P, et a1. Prognostic and diag­nosticvalueof DNA-cytometry in imprintsofmelanocyticlesions. Analyt Cell PathoI1989;1:319.

15. Bergman W, Ruiter DJ, Scheffer E, et al. Melanocyticatypia in dysplastic nevi: immunohistochemical and cyto­photometrical analysis. Cancer 1988;61:1660-6.

16. Meijer CJLM, VanderLoo EM, Scheffer E, et al. Rele­vance of morphometry in the diagnosis and prognosis ofcutaneousT celllymphomas. In: GoosM, Christophers E,eds. Lymphoproliferative diseases of the skin. Berlin:Springer, 1982:117-27.

17. Stolz W, Schmoeckel C, Burg G, et al. CirculatingSezarycellsin the diagnosis ofSezary syndrome (quantitative andmorphometric analysis). J InvestDermatoI1983;81:314-9.

18. WillemzeR, Cornelisse CJ, Hermans J, et al.Quantitativeelectron microscopy in the early diagnosis of cutaneous Tcelllymphomas: a longtermfollow-up studyof 77patients.Am J Patho! 1986;123:166-73.

19. Vonderheid EC, Fang SM, Helfrich MK, et al. Biophysi­cal characterization of normal T-Iymphocytes and Sezarycells. J Invest Dermatol 1981 ;76:28-37.

20. Noronha A, Richman DP.Simultaneouscell surfacephe­notype and cell cycle analysis of lymphocytes by flowcytometry. J Histochem Cytochem 1984;32:821-6.

21. Crocker J, Nar P. Nucleolarorganizerregions in lympho­mas. J PathoI1987;151:1I1-8.

22. CrockerJ, SkilbeckN. Nucleolarorganizerregion-associ­ated proteinsincutaneous melanotic lesions: a quantitativestudy. J Clin Pathol 1987;40:885-9.

23. Crocker 1. Nucleolar organizer regions. In: UnderwoodJCE, ed. Pathologyofnucleus. Current topics in pathology.Heidelberg: Springer, 1990:91-150.

24. RueschoffJ, PlateK, Bittinger A,et al.Nucleolarorganizerregions: basic concepts and practical application in tumorpathology. Pathol Res Pract 1989;185:878-85.

25. Fallowfield ME, Dodson AR, Cook MG. Nucleolarorga­nizerregionsin melanocytic dysplasiaand melanoma. His­topathology 1988;13:95-9.

26. Fallowfield ME, Cook MG. The value of nucleolar orga­nizerregionstaininginthe differential diagnosis of border­line melanocytic lesions. Histopathology 1989;14:299-304.

27. Howat AJ, Wright AL, Cotton DWK, et al. AgNORs inbenign, dysplastic, and malignantmelanocytic skin lesions.Am J Dermatopathol 1990;12:156-61.

28. Howat AJ, Giri DD, Wright AL, et al. Silver-stained nu­cleoli and nucleolar organizer region countsare of no prog­nostic value in thick cutaneous malignant melanoma. JPathol 1988;156:227·32.

29. HeinischG, Kuehn E, DimmerV. Nucleolar organizerre­gions bei melanozytaen Hautvernderungen. Z Hautkr1991;66:321-4.

30. SmolleJ, SoyerHP, KerlH. Proliferative activity of cuta­neous rnelanocytic tumors defined by Ki 67 monoclonalantibody-a quantitative immunohistochemical study.AmJ DermatopathoI1989;11:301-7.

31. Denham PL, Salisbury JR. Nucleolar organiser regioncountsinkeratoacanthomas and squamous cell carcinomas.J Pathol 1988;154:64-6.

32. Egan MJ, Crocker J. Nucleolarorganizer regions in cuta­neous tumours. .J Pathol 1988;154:247-53.

33. Buechner T, Bloomfield CD, Hiddemann W, et a1. Tumoraneuploidy. Berlin: Springer, 1985.

34. Zerntsov A, Koss W, Dixon L, et al. Anal verrucous carci-

Journal of the American Academy of DermatologyVolume29, Number I

nomaassociated with human papillomavirus type II: mag­netic resonance imaging and flow cytometry evaluation.Arch Dermatol 1992;128:564-5.

35. Bartkowiak D, Schumann J, Otto FJ, et al. DNA flow cy­tometry in the prognosis of primary malignant melanoma.Oncology 1991;48:39-43.

36. Bartkowiak D, Otto F,Schumann J, et al.SequentialDNAflow cytometryin metastatic malignant melanoma.Oncol­ogy 1991;48:154-7.

37. Heiden T, Goehde W, Tribukhit B. Two wave-lengthsmercury arc lamp excitation for flow cytometric DNA­protein analyses. Anticancer Res 1990;10:l555-62.

38. SmolleJ, Soyer HP, Ehall R, et al. NK celldensityin ma­lignant skin tumors: a stereological study. Br J Dermatol1987;116:823-9.

39. Iwahara K, Hashimoto K. T-cell subsetsand nuclear con­tour index of skin-infiltrating T-cells in cutaneous T-celllymphoma. Cancer 1984;54:440-6.

40. Williams RA, Rode J, Dhillon AP, et al. MeasuringSIOOprotein and neurone specific enolase in melanocytic tu­mours using video image analysis. J Clin Pathol 1986;39:1096-8.

41. Soyer HP. Ki 67 immunostaining in melanocytic skintumors: correlation with histologic parameters. J CutanPathol 1991 ;18:264-72.

42. Uhl GR, Zingg HH , Habener JF. Vasopressin mRNA insitu hybridization: localizationand regulation studied witholigonucleotide cDNA probes in normal and Brattlebororat hypothalamus. Proc Natl Acad Sci USA 1989;821:5555-9.

43. Hoefler H, Ruhri C, Puetz B, et al. Simultaneouslocaliza­tion of calcitonin mRNA and peptide in a medullary thy­roid carcinoma. Virchows Archiv [B] Cell Pathol 1987;54:144-51.

44. Stolz W, Scharffetter K, Abmayr W, et al. An automaticanalysis methodfor in situ hybridization usinghigh-resolu­tion imageanalysis. Arch Oermatol Res 1989;281 :336-41.

45. ScharffetterK, Stolz W, Lankat-Buttgereit B,et al. In situhybridisation-an useful tool for studies on collagen geneexpression in cellculture as wellas in normal and alteredtissue. Virchows Arch [B] Cell PathoI1989;56:299-306.

46. WiemannIN, Clifton DK, Steiner RA. Gonadotropin-re­leasing hormonemessenger ribonucleic acid levels are un­altered withchanges in the gonadal hormonemilieuof theadult male rat. Endocrinology 1990;127:523-32.

47. Larsson LI, Hougaard OM. Optimization of non-radioac­tive in situ hybridization: image analysis of varying pre­treatment hybridization and probe labelling conditions.Histochemistry 1990;93:347-54.

48. Scheynius A, Lundahl P. Three-dimensional visualizationof human Langerhans' cellsusing confocal scanning lasermicroscopy. Arch Dermatol 1990;281 :521-5.

49 . Flotte TJ, Seddon JM , Zhang Y, et al, A computerizedimageanalysis methodfor measuringelastic tissue. J InvestDermatol 1989;93:358-62.

50. Kim NH, AggarwalSJ, BovikAC, et al. 3-Dmodel ofvas­cular network in rat skin obtained by stereo techniques. JMicrosc 1990;158:275-84.

51. Braverman 1M, Sibley J, Keh-Yen A. A study of the veil

Smolle et a/. 97

cells around normal, diabetic, and aged cutaneous mi­crovessels. J Invest Dermatol 1986;86:57-62.

52. Lindelof B, Forslind B, Hedb1ad MA, et a1. Human hairform: morphology revealed by light and scanningelectronmicroscopy and computer aided three-dimensional recon­struction. Arch Dermatol 1988;124:1359-63.

53. Peter RU, Maeler W, Mueller M. Morphometric assess­ment of radiationdamage in basal cellsof human skin:ev­idence for differences in individual reactivity. Proceedingsof the 22nd Annual Meet ing of the European Society forRadiation Biology (In press.)

54. Mareel MM. Invasion in vitro: methods of analysis. Can­cer Metast Rev 1983;2:201-18.

55. Smolle J, Helige C, Soyer HP, et a t. Quantitative evalua­tion of melanoma cell invasion in three-dimensional con­frontation cultures in vitro using automated image analy­sis. J Invest DermatoI1990;94:1I4-9.

56. Smolle J, Helige C, Tritthart HA. An image analysis andstatisticalevaluation program for the assessmentof tumourcell invasion in vitro. Analyt Cell Pathol 1992;4:49-57.

57. ElGammalS, Hoffmann K, Kenkmann J, et al. Principlesof three-dimensional reconstructions from hig-resolutionultrasoundin dermatology. In: Altmeyer P, E[Gammal S,Hoffmann K, eds, Ultrasound in dermatology. Berlin:Springer, 1992:355-84.

58. EIGammal S. Altmeyer P, Hinrichsen H. ANAT3D:shaded 3D surface reconstructions from serial sections.Acta Stereol 1989;suppl 8/2:543-50.

59. ElGammal S, Altmeyer P. The three-dimensional histo­logical architecture of pustulosis palma-plantaris. ActaDerrn Venereal Suppl (Stockh) 1989;146:184-91.

60. Friedman RJ. Rigel DS. Kopf AW , et al. Vo1umeofma­lignant melanoma is superior to thickness as a prognosticindicator: preliminaryobservation. Derrnatol Clin 1991;9:643-8.

61. Ackerman AB. Differentiation of benign from malignantneoplasmsbysilhouette.AmJ Dermatopathol1989;II :297­300.

62. Ackerman AB, Magana-Garcia M. Naming acquiredmelanocytic nevi:Unna's, Miescher's, Spitz's, Clark's. Am] DermatopathoI1990;12:193-209.

63. SmolleJ, Soyer HP, Smolle-Juettner FM, et al. Computersimulation of tumor cell motility and proliferation. PatholRes Pract 1990;186:467-72.

64. Smolle J, Soyer HP, Smolle-Juettner FM, et al. Computersimulation analysis of morphological patterns in humanmelanocytic skin tumours. Patho1 Res Pract 1991;187:986­92.

65. Smolle J, Grimstad lAo Tumor cell motility within tumorsdeterminedby applying computer simulation to histologicpatterns. Int J Cancer 1992;50:331-5.

66. SmolleJ. Smolle-Juettner FM, Stettner H, et al. Relation­ship of tumor cell motility and morphologicpatterns.Part1. Melanocytic skin tumors. Am J Dermatopathol 1992;14:231-7.

67. Smolle J, Hofmann-Wellenhof R, Kerf H. Prognostic sig­nificanceofproliferationand motility in primary malignantmelanoma of the skin. J Cutan Pathol 1992;19:[ 10-5.