www.elsevier.com/locate/yabio

ANALYTICAL

BIOCHEMISTRY

Analytical Biochemistry 357 (2006) 137–143

Semiautomated computer-assisted image analysis to quantify3,3 0-diaminobenzidine tetrahydrochloride-immunostained small tissues

Sandra Leal a, Carmen Diniz a,*, Carlos Sa b, Jorge Goncalves a, Ana Sofia Soares a,Carolina Rocha-Pereira a, Paula Fresco a

a Servico de Farmacologia, CEQOFFUP, Faculdade de Farmacia, Universidade do Porto, 4050-047 Porto, Portugalb Centro de Materiais da Universidade do Porto (CEMUP), Universidade do Porto, 4150-180 Porto, Portugal

Received 23 May 2006Available online 1 August 2006

Abstract

This work aimed to develop a technique to measure stained areas in images from sample tissue sections, namely when the structure ofinterest does not fill the entire image field of the microscope. We propose a semiautomated computer-assisted image analysis (SACAIA)method in which brightfield color images of 3,3 0-diaminobenzidene tetrahydrochloride (DAB)-stained antigens are converted to theirblue component and boundaries are delineated to extract the object of interest. The number of pixels of a defined color (elicited byDAB) is counted and used to measure the stained area relative to the total area of the tissue under study. The percentages of area stainedwith adenosine A1 receptor were 40.76 ± 2.08 and 42.44 ± 2.26% for manual analysis and SACAIA, respectively (P = 0.582). A stronglinear correlation of A1 receptor quantification was found (r = 0.98, P < 0.001, and 95% CI = 0.97 to 0.99 for manual method; r = 0.99,P < 0.001, and 95% CI = 0.98 to 0.99 for SACAIA method). The extent to which misclassification affected staining quantification wasevaluated by Bland–Altman analysis, indicating that this method can be applied accurately to quantify the immunohistochemical stain-ing area (occupied by a specific antigen) in small sample tissues that do not fill the entire image field of the microscope.� 2006 Elsevier Inc. All rights reserved.

Keywords: Quantification; Immunostains; Small sample tissues; Semiautomated computer-assisted analysis

1 0

Immunohistochemistry is one of the techniques current-ly used to identify and map the distribution of a specificantigen. It is easy to standardize, but the interpretationrelies solely on subjective visual estimation and yields onlyqualitative results. The precise description of differences instaining patterns among sets of tissues and comparison ofstaining among experimental groups requires some formof quantification. Indeed, analysis using a quantitativemethodology of immunostained specimens provides moreinformation and leads to a more accurate elucidation ofputative physiological and pathophysiological dataobtained in functional studies.

During the past few years, several methods based onadvanced image analysis using interactive [1–4] and auto-

0003-2697/$ - see front matter � 2006 Elsevier Inc. All rights reserved.

doi:10.1016/j.ab.2006.07.017

* Corresponding author. Fax: +351 222078969.E-mail address: cdiniz@ff.up.pt (C. Diniz).

mated [5–8] imaging techniques have been shown to reducethe variation in analysis of the staining levels [9]. Thesetechniques are best suited for counting pixel number [10]or measuring areas [11], relying on the ability to clearly sep-arate or segment a structure of interest from its back-ground using a physical difference such as color.

One of the most frequently used chromophores inimmunostaining procedures is 3,3 0-diaminobenzidine tetra-hydrochloride (DAB)1, which results in a brown color. Pre-vious studies developed a computer-assisted analysis ofsections stained with DAB [3,8]. These studies have been

Abbreviations used: DAB, 3,3 -diaminobenzidine tetrahydrochloride;PBS, phosphate-buffered saline; RT, room temperature; H2O2, hydrogenperoxide; RGB, red, green, and blue; SACAIA, semiautomated computer-assisted image analysis.

Fig. 1. Image from a representative tissue section stained with anti-A1

adenosine receptor antibody using DAB as chromogen. The dark brownstaining represents positive A1 receptor staining. Scale bar = 1.38 lm.

138 Semiautomated computer-assisted image analysis / S. Leal et al. / Anal. Biochem. 357 (2006) 137–143

developed for the analysis of features and proteins intumors [3–7].

Recently, several studies have been reported concerningquantitative analysis of angiogenesis in excised tumor tis-sues [4,6,12]. However, to our knowledge, no quantitativecharacterization of a feature or protein has been reportedin vessels from nonpathological tissues (arteries and veins).The aim of the current study was to quantitatively charac-terize a membrane-bound protein, the adenosine A1 recep-tor, in rat tail artery sections stained with DAB. To attainthis objective, a procedure was developed that allows thetissue adenosine A1 receptor antibody staining fractionarea (Sf) to be quantified using digital images of DAB-la-beled immunostains from tail arteries of the rat that donot fill the entire image field of the microscope.

Materials and methods

Immunohistochemical procedures

Male Wistar rats weighing 230–260 g (Charles River,Barcelona, Spain) were kept under standard laboratoryconditions (controlled, light/dark cycles of 12 h, 20–22 �C, and unlimited access to water and a standard dietof pellet food). Handling and care of all animals were inagreement with the European Union guidelines for animalresearch.

Immunohistochemistry was carried out as described pre-viously [13,14]. Briefly, animals were subjected to intracar-diac perfusion fixation using 4% formaldehyde dissolved inphosphate-buffered saline (PBS, pH 7.4). Tail arteries wereremoved and postfixed overnight in Bouin’s solution. Afterrinsing, tissues were embedded in paraffin for subsequentprocedures. Tail arteries from three animals in total wereinvestigated. From each tail artery, 15–18 sections wereanalyzed.

Sections 7–8 lm thick were dewaxed in toluene andrehydrated in graded ethanol solutions. Then slices wererinsed in PBS at room temperature (RT) and treated with3% hydrogen peroxide (H2O2) in the dark for 5 min. Sliceswere rinsed in PBT at RT and incubated in blocking solu-tion (10% normal horse serum albumin in PBT) at RT in ahumidified chamber for 30 min. Then sections were incu-bated with a rabbit polyclonal antibody (anti-A1, A1R11-A, Alpha Diagnostics International, San Antonio, TX,USA). Anti-A1 was applied in PBT at a dilution of 1/250overnight at 4 �C. Then slices were rinsed three times inPBT and incubated with biotinylated secondary antibodyproduced in rabbit (Vectastain Elite ABC kit universal,Vector Laboratories, Burlingame, CA, USA) at a dilutionof 1/125 (1 h at RT). Slices subsequently were rinsed threetimes in PBT and incubated with avidin–biotin complexreagents (ABC, 1 h at RT, Vectastain Elite ABC kit univer-sal). Then slices were rinsed three times in PBS, and DAB(activated by H2O2) was used as a chromophore. Afterrinsing slices with PBS for 10 min, tissues were rehydratedwith ethanol/xilol solutions and mounted with Eukitt

mounting medium. Control samples performed on adjacentsections were carried out using 10% normal horse serum orblocking solution instead of the primary antibody.

Image acquisition of DAB-labeled immunostains from small

tissues

DAB-labeled immunostain sections were imaged usingbrightfield optics on a Nikon Eclipse E400 microscope(Nikon, Tokyo, Japan) with a 40· objective (CFI Plan0.75, Nikon). Individual images were acquired using aCDD camera (Canon PowerShot S40, Canon Europe,Amsterdam, The Netherlands) mounted on the microscopeand connected via USB port to a computer using ZoomBrowser EX (Canon Europe). Images were focused andselected before digital acquisition. Because tissue sectionsof rat tail artery do not fill the entire image field, the micro-scope field that included the most extensive portion possi-ble of tissue section was selected. Illumination conditionsof the brightfield optics and camera exposure were main-tained constant throughout acquisition of all images. Imag-es from control tissue sections were acquired in the samemicroscope/CDD camera operating conditions as theDAB-immunostained tissue sections.

Acquired images (24 bit, 8 bits/color, Fig. 1) from tissuesections had a resolution of 992 · 695 pixels, correspondingto a 357 · 250-lm area on the original histological section(1 pixel = 0.3597 lm, with a calibration micrometer slidebeing used to convert pixels into micrometers).

Semiautomated computer-assisted image analysis of DAB-

labeled immunostains from small tissues

To quantify DAB-labeled immunostains of tissues thatdo not fill the entire image field, the flowchart (Fig. 2) forthe image analysis software (PAQI, CEMUP, Porto, Por-tugal) was used. This flowchart can easily be adapted to

Fig. 2. Flow chart of the semiautomated computer-assisted image analysis(SACAIA) method for immunohistochemically stained tissue sections thatdo not fill the entire microscope field: (a) the ‘‘user modify image’’ loopallows the operator to define boundaries because the tissue does not fill theentire image, discarding any areas that were selected incorrectly; (b) the‘‘quantify image’’ step measures the area of highlighted DAB-stainedtissue. Regions of intensity levels greater than 123 (maximum thresholdingobtained for the control images) were excluded and not considered asDAB-stained. temp., temporary; perm., permanent.

Fig. 3. Upper panel: Image of the blue component after split of originalRGB image from a cross section of rat tail artery stained with anti-A1

adenosine receptor antibody. Lower panel: The boundary of the tissuesection was delineated, and the external area was cleaned (white area).

Semiautomated computer-assisted image analysis / S. Leal et al. / Anal. Biochem. 357 (2006) 137–143 139

software other than PAQI such as NIH Images (http://rsb.info.nih.gov) and Adobe Photoshop (Adobe Systems,San Jose, CA, USA). Conventional brightfield color images(24 bit, red, green, and blue [RGB]) were used, and thephysical property ‘‘color’’ allowed to subtract the objectof interest from its background. On preliminary experi-ments, several images containing stained and nonstainedareas were studied. The histogram of RGB componentsshowed larger contrast on the blue component than onthe red and green components. Therefore, from RGB colorimages, the blue component (after the splitting of RGBimages) was selected for further analysis.

Although the observer can easily identify the tunicaadventitia of the artery, this layer is difficult to segmentcorrectly using an automated procedure. Therefore, bound-aries were delineated on a case-by-case basis to extract theobject of interest. Because tissue sections did not fill theentire image field (Fig. 3), only the part of the microscope

field that contained tissue was considered for further anal-ysis. Then image analysis software was used to set thresh-olds for automated DAB selection, and image noise wasremoved for quantification of the staining parameters(Fig. 4).

To determine differences between stained and non-stained tissues, control samples were imaged (with the samemicroscope illumination and camera operating conditionsas stated previously) and the average intensity level wasdetermined (a value of 123 for a maximum of 255). Thisintensity was used for threshold segmentation of thestained areas of the objects of interest (regions of intensitygreater than 123 were excluded and not considered as beingDAB stained).

A total of 50 random and representative images belong-ing to different staining batches were selected. On two sep-arate days, DAB-stained tissue area (Si for the field i) andtotal tissue area (Ti = DAB-stained tissue area + non-DAB-stained tissue area) in each image field of area (Fi)were quantified using semiautomated computer-assistedimage analysis (SACAIA 1 and SACAIA 2). Each imagewas also analyzed on two different days by manual quanti-fication (considered as one of the most accurate techniques

Fig. 4. Result of threshold segmentation of the image shown in Fig. 1.Red color highlights intensity levels less than 123 pixels. Green colorhighlights intensity levels greater than or equal to 123 pixels.

140 Semiautomated computer-assisted image analysis / S. Leal et al. / Anal. Biochem. 357 (2006) 137–143

for classifying an area as stained positive or negative forDAB chromogen) using the Cavalieri principle. Briefly, agrid with an area of 12 mm2 was used, and positivelystained areas were selected by manual quantification (Man-ual 1 and Manual 2).

Presentation of data and statistics

To measure the level of immunohistochemical staining(Sf), the fraction of the tissue stained with DAB was quan-tified. This parameter was computed based on the measure-ments of areas of the stained tissue and of the total tissue asimaged on our samples:

Sf ¼X

Si

.XT i;

where Sf is the staining fraction area, Si is the stained tissuearea in each image field, and Ti is the total tissue area ineach image field. Ti typically is not equal to the area ofthe image field because tissue samples do not fill the entireimage field of the microscope.

To provide some insight into the accuracy of dataobtained by SACAIA, a comparison between the measure-ments made by the SACAIA method and those made bythe manual method was performed on a field-by-field mea-surement basis (si = Si/Fi, fractional area field as percent-age of stained tissue; ti = TiFi, fractional area fieldpercentage of total tissue).

Correlation coefficients (rs) and the slope of regressionlines for the mean percentage of the observer (manualmethod: Manual 1/Manual 2) and computer (SACAIAmethod: SACAIA 1/SACAIA 2) were calculated, and datawere plotted against the line of perfect agreement (45� line).Therefore, linear regression provided a graphical represen-tation of agreement and correlation linearity of data. Thismethod gives some insight into accuracy of the data butcannot be used as an independent measure of agreement[3,15].

Results of percentage area measurements by theSACAIA were compared with mean manual values usingsimilar analysis (Bland–Altman analysis). When two tech-niques are compared (SACAIA method vs. manualmethod), automated percentage area measurements areexpected, with 95% confidence, to differ from the meanmanual value (limit of agreement = mean differ-ence + 1.96 · standard deviation) [3,8,16,17]. This rangeis the confidence limit for the techniques under study whenthe percentage area is measured; the narrower this range,the more precise the method. A fundamental assumptionfor Bland–Altman analysis is that the mean difference mustbe independent of the measurement magnitude [18], andthis was verified by calculating correlation coefficients forplots of the mean difference against mean observer values(data not shown).

Results are presented as means ± SEM of 50 images,and Student’s t test was used to compare individual read-ings of both the manual and SACAIA methods.

Results

A primary antibody (anti-A1 antibody) against adeno-sine A1 receptor, a membrane protein widely expressed inrat tail artery [14], was immunostained with DAB to devel-op a brown color (Fig. 1) in tissue sections of rat tail arter-ies that do not fill the entire image field.

The SACAIA method is able to discriminate tissue frombackground when sample tissues do not fill the entire imagefield of the microscope and provided results that are shownin Table 1. The fractional area field (percentage of total tis-sue) measured by the manual method was not significantlydifferent from the value obtained with the SACAIAmethod (P > 0.05, Table 1). Separation between DAB-stained tissue and non-DAB-stained tissue (Fig. 5) was car-ried out using the semiautomated technique describedabove. Therefore, the SACAIA method provided a quanti-tative measurement of the staining corresponding to aden-osine A1 receptors in tissue sections labeled with a specificadenosine anti-A1 receptor antibody plus DAB. Mean andSEM values of the stained area for each group are present-ed in Table 2. Images (n = 50) were analyzed twice usingboth the manual and SACAIA methods. As shown inTable 2, mean readings of both methods were not statisti-cally different (P = 0.582).

Validation of the SACAIA and manual methods todetermine the area of staining was obtained by their respec-tive correlation coefficients (r = 0.99, P < 0.001, 95%CI = 0.98–0.99 for SACAIA method; r = 0.98, P < 0.001,95% CI = 0.97–0.99 for manual method). Confirmationwas made by plotting, which showed that replicate datapoints for both methods fell near the 45� line, indicatinghigh precision and accuracy.

The average differences using the manual and SACAIAmethods were 0.22 (95% CI = �0.53 to 0.97) and 0.78 (95%CI = �0.18 to 1.39), respectively. For the manual method,the lower and upper limits of agreement were �4.98 and

Table 1Comparison between manual and SACAIA methods in their ability to discriminate tissue from background

Fractional area field (percentage of total tissue)(mean ± SEM) (n)

r (95% CI) Mean difference(95% CI)

Upper limit Lower limit

Manual 47.65 ± 1.93 (50) 0.98 (0.97–0.99) 0.81 (0.10–1.53) 5.75 �4.12SACAIA 47.82 ± 2.05 (50) 0.98 (0.97–0.99) �0.33 (�1.08–0.42) 4.84 �5.51Manual/SACAIA 0.95 (0.91–0.97) �0.15 (�1.49–1.20) 9.12 �9.41

Note. Tissue sections were incubated with a polyclonal primary antibody against adenosine A1 receptors. The resulting immunocomplexes were detectedwith a biotinylated secondary antibody and amplified by ABC complex using DAB as chromogen. Results were obtained on images acquired in the samemicroscope/CDD camera operating conditions. Tissue samples that do not fill the entire image field of the microscope were extracted and calculated asfractional area field percentage of total (stained and unstained) tissue (ti = Ti/Fi). Significant differences from manual method: P < 0.05 (Student’s t test).

Fig. 5. Overlap of images shown in Figs. 3 and 4. The area delineated bythe red line shows the area immunostained with anti-A1 adenosinereceptor antibody. The area delineated by the green line represents thearea with intensity levels similar to those obtained in control images (notimmunostained). (For interpretation of the references to color in thisfigure legend, the reader is referred to the Web version of this article.)

Table 2Percentages of area stained with adenosine A1 receptor

Minimum Maximum Mean ± SEM (n)

Manual1 7.50 70.00 40.97 ± 2.07 (50)2 7.50 70.00 40.65 ± 2.11 (50)(1 + 2) / 2 40.76 ± 2.08 (50)

SACAIA1 11.07 77.55 42.86 ± 2.28 (50)2 10.99 73.87 42.01 ± 2.24 (50)(1 + 2) / 2 42.44 ± 2.26 (50)

Note. Tissue sections were incubated with a polyclonal primary antibodyagainst adenosine A1 receptors. The resulting immunocomplexes weredetected with a biotinylated secondary antibody and amplified by ABCcomplex using DAB as chromogen. Results were obtained on imagesacquired in the same microscope/CDD camera operating conditions.Results are presented as fractional area of field (percentage of stainedtissue: si = Si/Fi). Significant differences from manual method: P < 0.05(Student’s t test).

Fig. 6. Upper panel: Data with line of perfect concordance. All datapoints fall relatively near and on both sides of the 45� line. Lower panel:difference between mean manual and mean SACAIA measurementsagainst the overall mean of the image percentage area (average of manualand SACAIA measurements). The differences are scattered on either sideof the mean difference line, and there is no tendency for those differencesto increase with increased measurements.

Semiautomated computer-assisted image analysis / S. Leal et al. / Anal. Biochem. 357 (2006) 137–143 141

5.42, respectively; for the SACAIA method, they were�3.38 and 4.95, respectively. There was no evidence ofincreasing variability with increasing the number ofmeasurements.

The manual and SACAIA methods had a concor-dance coefficient value of 0.96 (95% CI = 0.93 to 0.97),with data points falling relatively near the 45� line, indi-cating high accuracy and agreement (Fig. 6, upper pan-el). The bias between the two methods was �1.66 (95%CI = �2.95 to �0.37), and the lower and upper limitsof agreement were �10.56 and 7.25, respectively(Fig. 6, lower panel).

142 Semiautomated computer-assisted image analysis / S. Leal et al. / Anal. Biochem. 357 (2006) 137–143

Discussion

Internet-based measurement systems have become avail-able during recent years, and SACAIA can be carried outwith most conventional image processing software pro-grams and is not dependent on platforms software. In fact,this technique can easily be implemented using Image J(http://rsb.info.nih.gov/ij/) or Adobe Photoshop on eithera Windows or Macintosh platform.

The current study has presented a technique using anSACAIA that allows the measurement of immunohisto-chemical DAB-positive staining areas (e.g., those obtainedwith the anti-adenosine A1 receptor) in vascular tissues(e.g., arteries, veins) that do not fill the entire image fieldof the microscope. By segregating immunostained areafrom non-immunostained areas, it is possible to establisha proportion of the stained tissue area that represents a rel-ative measure of the antibody binding that can be assumedto reflect the extension of tissue where adenosine A1 recep-tors are present. Information on the dispersion of receptorsgiven by this method can be integrated with functionaldata, allowing a better understanding of receptors’ physio-logical roles. Furthermore, because of its simplicity, thismethod may stimulate and extend the use of quantitativeimage analysis of immunostained tissues. We havedescribed a cellular membrane receptor in a tissue prepara-tion (the rat tail artery) and a simple technique that accu-rately isolates the tissue area from the rest of the field ofthe microscope (Table 1). In addition, this technique allowsDAB-stained tissue areas to be discriminated from non-DAB-stained tissue areas.

It is generally accepted that the nature of tissue process-ing, including the fixative used, the duration of fixation,and the size of the tissue originally fixed, can potentiallyalter the signal of immunohistochemistry. This drawbackcan be circumvented by SACAIA because this method pro-vides information about the fraction of the image occupiedby a particular chromogen and not its absolute amount.

Conversion of brightfield (24-bit) images to blue (8-bit)images is simple and standard, and the setting of a thresh-old on the blue channel allows accurate selection of DAB-stained pixels and, therefore, extraction of the stained areain the tissues. In our study, threshold was not adjusted ineach individual image by interactive evaluation of theobserver as it has been in previous studies [3]; rather, itwas defined after a previous analysis of all control sampletissues (unstained tissue reference images) acquired in thesame standard measurement conditions as the stained sam-ple images. This analysis revealed that values greater than123 correspond to pixels associated with nonstained imageareas (tissue and field of microscope). This value was usedas the set threshold, and once a threshold was selected(from control sample tissues), all subsequent blue channelimages (obtained after splitting the 24-bit images) wereanalyzed. Selection of the blue channel values used in thismethod has been shown previously to be the best optionof the RGB channels [3,7,19]. The 8-bit blue image kept

a clearer picture of the overall tissue morphology (Fig. 4)conserving the primary images and allowed the tissueboundaries to be easily identified and drawn, a critical stepin defining the area of the image field filled with the tissuesample area. An advantage of the proposed method is thatit was designed to reduce the reported difficulties in distin-guishing the counterstained areas from low-intensity areasof DAB staining [3,7,8,19]. Omitting counterstainingreduces errors, particularly those caused by pixel misclassi-fication. This approach warrants that all DAB-stainedareas correspond to positive-stained areas of the proteinunder study and discards positive-stained areas locatedoutside the tissue (artifacts). Therefore, this method cir-cumvents the difficulties presented by other computer-as-sisted and fully automated image analysis methods toseparate or segment the staining areas of the structure ofinterest from background [3,7,19] and from the rest ofthe image field. In this method, the operator evaluation isrequired only to delineate tissue boundaries to preventinclusion of misclassified pixels from artifacts.

To evaluate the consistency of a new method, the mostappropriate approach was to compare the values of mea-surement obtained using the new method with thoseobtained using an established one [20]. Using the Bland–Altman analytic approach, agreement between theSACAIA and manual methods was demonstrated. In sum-mary, this work showed that the SACAIA method can beeasily applied to accurately quantify membrane receptorantibody (e.g., adenosine receptors) staining fraction areain sample tissues that do not fill the entire image field ofthe microscope.

Acknowledgments

The authors are grateful to Centro de Materiais da Uni-versidade do Porto (CEMUP) for technical assistance. Thiswork was supported by FCT:I & Dn� 226/94, POCTI–QCAIII, POCTI/SAU–FCF/60782/2004, FEDER, andFundacao Ilıdio Pinho.

References

[1] H.-A. Lehr, D.A. Mankoff, D. Corwin, G. Santeusanio, A.M. Gown,Application of Photoshop-based image analysis to quantification ofhormone receptor expression in breast cancer, J. Histochem. Cyto-chem. 45 (1997) 1559–1565.

[2] J. Vilaplana, M. Lavialle, A method to quantify glial fibrillary acidicprotein immunoreactivity on the suprachiasmatic nucleus, J. Neuro-sci. Methods 88 (1999) 181–187.

[3] T.W. King, E.M. Brey, A.A. Youssef, C. Johnston, C.W. Patrick Jr.,Quantification of vascular density using a semiautomated techniquefor immunostained specimens, Anal. Quant. Cytol. Histol. 24 (2002)39–48.

[4] J.N. McGinley, K.K. Knott, H.J. Thompson, Semi-automatedmethod of quantifying vasculature of 1-methyl-1-nitrosourea-inducedrat mammary carcinomas using immunohistochemical detection, J.Histochem. Cytochem. 50 (2002) 213–222.

[5] M. Goto, Y. Nagatomo, K. Hasui, H. Yamanaka, S. Murashima, E.Sato, Chromaticity analysis of immunostained tumor specimens,Pathol. Res. Pract. 188 (1992) 433–437.

Semiautomated computer-assisted image analysis / S. Leal et al. / Anal. Biochem. 357 (2006) 137–143 143

[6] R. Montironi, L. Diamanti, D. Thompson, H.G. Bartels, P.H. Bartels,Analysis of the capillary architecture in the precursors of prostatecancer: recent findings and new concepts, Eur. Urol. 30 (1996) 191–200.

[7] A.C. Ruifrok, Quantification of immunohistochemical staining bycolor translation and automated thresholding, Anal. Quant. Cytol.Histol. 19 (1997) 107–113.

[8] E.M. Brey, Z. Lalani, C. Johnston, M. Wong, L.V. McIntire, P.J.Duke, C.W. Patrick Jr., Automated selection of DAB-labeled tissuefor immunohistochemical quantification, J. Histochem. Cytochem. 51(2003) 575–584.

[9] T. Seidal, A.J. Balaton, H. Battifora, Interpretation and quantifica-tion of immunostains, Am. J. Surg. Pathol. 25 (2001) 1204–1207.

[10] P.D. Kohlberger, F. Breitenecker, A. Kaider, A. Losch, G. Gitsch, G.Breitenecker, D.G. Kieback, Modified true-color computer-assistedimage analysis versus subjective scoring of estrogen receptor expressionin breast cancer: a comparison, Anticancer Res. 19 (1999) 2189–2193.

[11] E.J. Goldlust, R.P. Paczynski, Y.Y. He, C.Y. Hsu, M.P. Goldberg,Automated measurement of infarct size with scanned images oftriphenyltetrazolium chloride-stained rat brains, Stroke 27 (1996)1657–1662.

[12] T. Szoke, K. Kayser, J.D. Baumhakel, I. Trojan, J. Furak, L.Tiszlavicz, A. Horvath, K. Szluha, H.J. Gabius, S. Andre, Prognosticsignificance of endogenous adhesion/growth-regulatory lectins in lungcancer, Oncology 69 (2005) 167–174.

[13] C. Diniz, S. Leal, J. Goncalves, Regional differences in the adenosineA2 receptor-mediated modulation of contractions in rat vas deferens,Eur. J. Pharmacol. 460 (2003) 191–199.

[14] C. Diniz, P. Fresco, S. Leal, J. Goncalves, Adenosine receptorsinvolved in modulation of noradrenaline release in isolated rat tailartery, Eur. J. Pharmacol. 504 (2004) 17–25.

[15] S. Hansen, D.A. Grabau, C. Rose, M. Bak, F.B. Sorensen,Angiogenesis in breast cancer: a comparative study of the observervariability of methods for determining microvessel density, Lab.Invest. 78 (1998) 1563–1573.

[16] J.M. Bland, D.G. Altman, Measurement error and correlationcoefficients, Br. Med. J. 313 (1996) 41–42.

[17] L.E. Daly, G.J. Bourke, in: G.J. Bourke (Ed.), Interpretation andUses of Medical Statistics, Blackwell Science, Oxford, UK, 2000, pp.381–420.

[18] J.M. Bland, D.G. Altman, Comparing methods of measurement: whyplotting difference against standard method is misleading, Lancet 346(1995) 1085–1087.

[19] B.L. Kuyatt, C.A. Reidy, K.Y. Hui, W.H. Jordan, Quantification ofsmooth muscle cell proliferation in cultured aorta, Anal. Quant.Cytol. Histol. 15 (1993) 83–87.

[20] J.M. Bland, D.G. Altman, Statistical methods for assessing agree-ment between two methods of clinical measurement, Lancet 1 (1986)307–310.