Author note (2008). This 1994 paper is very much out of date
today. It is made available because it contains some interesting algorithms
for nuclear measurements that are still useful for, particularly for
precancers.
The name of the described software, IMAPATH, has been subsequently withdrawn
because it came to our attention that there is a software company with that name.
Our software was re-named ISAP.
The software has not been maintained or updated over the past decade or
so. So it is impossible now to receive free copies of the ISAP software.
Jules J. Berman, Jan 2, 2008.
Berman JJ, Moore GW. Image analysis software for the detection of preneoplastic lesions. Cancer Letters,
77(2-3):103-109, 1994
IMAGE ANALYSIS SOFTWARE FOR THE DETECTION OF PRENEOPLASTIC AND EARLY NEOPLASTIC LESIONS
Jules J. Berman, Ph.D., M.D. and G. William Moore, M.D., Ph.D.
Departments of Pathology, Baltimore Veterans Affairs Medical Center, University of Maryland School of Medicine and The Johns Hopkins Medical Institutions, Baltimore, Maryland (U.S.A.)
SUMMARY
Preneoplastic lesions are usually small, and often appear as foci of atypical cells that blend into the surrounding normal tissue without producing a detectable tumor mass. Since these lesions seldom provide adequate tissue for biochemical studies, their detection often depends upon subtle distinctions in cytologic features. Image analysis permits pathologists to obtain quantitative measurements on cytologic and histologic preparations, so that visual impressions can be augmented by quantitative morphometry. Preneoplastic lesions have well-described morphometric features relating to nuclear area, texture, or shape. It is now feasible for every pathology department to capture images of pathologic material with equipment costing less than the price of a microscope. Captured image files can be analyzed using commercial software or software developed in several U.S. government agencies and made freely available to the public. Image analysis has been shown to improve the detection of preneoplastic cells. Recent improvements in the resolution of captured images, in the algorithms that measure preneoplastic descriptors, and in the ease and speed of transmission of images between laboratories, should increase our ability to detect and treat preneoplastic lesions.
Keywords: image analysis, preneoplasia, software, image processing, early detection
INTRODUCTION
Preneoplasia (also called precancer and premalignancy) is arguably the most important disease entity of modern man. In theory, the identification and elimination of all cancer precursors should lead to the total eradication of cancer [39]. Despite the enormous investment made in cancer research, the overall cancer death rate of most malignancies has shown no decrease in the past several decades [3]. A notable exception is the dramatic decrease in the cancer death rate from uterine cervical cancer, which derives almost entirely from the detection and elimination of preneoplastic precursors [12,26]. In fact, the projected death rate from cervical carcinoma is reduced about 90% for female populations with Pap smears obtained at least every three years [14,15]. Success has been achieved for cytologic screening of esophageal and stomach cancer in high risk populations [7,40,43].
Over the past two decades, there have been several important advances in our understanding of preneoplastic lesions. Preneoplastic lesions are biologically distinct entities, often with characteristic morphologic features, and should not be confused with small cancers. The properties of preneoplastic lesions include: 1) Small size, often blending into the surrounding tissue without producing a tumor mass; 2) Multiple lesions, often spread widely within an organ or tissue; 3) Regression occurring more frequently than progression to a fully-developed cancerous state; 4) No metastatic or invasive capability; 5) Appearance soon after a carcinogenic event (months for preneoplastic lesions, compared to years for cancers) until a ); 6) Restricted and slow growth (not autonomous); and 7) Expression of some but not all the molecular markers usually associated with the malignancies into which they occasionally develop [39].
Pathologists detect preneoplastic lesions primarily by the recognition of changes in nuclear morphology. These changes include increased nuclear size; irregular distribution of chromatin; irregularly contoured nuclear rims; variation in size and shape of nuclei among a population of similar cells; and multiple enlarged nucleoli. These features differ fundamentally from the features that pathologists use to identify fully-developed cancers. Cancers are identified by architectural features in the growth pattern of the tumor cell population, and by the type of cells that compose the tumor, namely features that are basically non-quantitative. An art connoisseur distinguishes a Rembrandt from a Van Gogh using the same sort of non-quantitative feature comparisons that pathologists use to distinguish a squamous cell carcinoma from an adenocarcinoma. Whereas humans excel at pattern recognition, instruments are better at measuring size, changes in pixel values, and summed pixel densities within defined object boundaries [37]. Many features that characterize preneoplasia, such as nuclear size, nuclear optical density and variation in nuclear size, are measurable quantities that can only be reliably assessed by analyzing images of sampled cells [16,34].
The Baltimore Veterans Administration Medical Center has recently introduced a networked system for capturing and retrieving pathologic images [31]. Captured images are maintained in a permanent database and indexed to the patient from whom the specimen was obtained. Hospital staff can examine clinical laboratory data, surgical pathology reports, cytology reports and all associated image files captured on a particular patient. An interactive image analysis system was developed to measure cytologic parameters corresponding to preneoplastic features [32].
MATERIALS AND METHODS
Equipment.
Color images were captured using an RGB frame capture camera, model TK-1070U, JVC Professional Products, Elmwood Park, NJ. Black and white images were captured on a model CCD 200E VideoScope International, Ltd., Washington, DC. Captured images were digitized with an ATVista (TrueVision, Indianapolis) image grabber board. Image capture was integrated into the VA DHCP (Decentralized Hospital Computer Program). The DHCP is a public domain hospital information system written in ANSI standard MUMPS and used at 169 VA Medical Centers. Image workstations workstations are all IBM-PC compatible 80386 or 80486 computers with Ethernet DNI Cards from Cabletron Systems, Rochester, NH. The VA Integrated Data Communications Utility is a fully digital optical fiber network with 23,000 miles of fully digital optical fiber [13].
Images were formatted either as 8-bit uncompressed grayvalue images (756 by 486 pixels) in Targa format or as 16-bit uncompressed color (red-green-blue, 756 x 486) Targa images. The Targa image file format consists of an 18-bit header followed by the pixel values in row-by-row order. Targa images can easily be converted to any other array (raster) image file format with commercially available image conversion utilities. Captured images were downloaded from the DHCP as freestanding files (367,434 bytes for black and white images, 734,850 for color images), that were transferred to another computer (Everex 386/33) and used by IMAPATH [ed: now ISAP].
RESULTS
Software.
IMAPATH [ed: now ISAP] is a pathology image analysis system developed at the Baltimore VA Medical Center [32] and distributed in the public domain. IMAPATH [ed: now ISAP] can be obtained at no charge by contacting the corresponding author. It was written with Microsoft Visual Basic Version 2.0 (Microsoft Corporation, Redmond, WA) and operates in the Microsoft Windows environment (v. 3.1). IMAPATH [ed: now ISAP] converts uncompressed Targa image files into uncompressed BMP files, a preferred Windows environment format. All display and analysis algorithms in IMAPATH [ed: now ISAP] operate on files in the BMP format. IMAPATH [ed: now ISAP] interactively recognizes the nucleus of each cell in an image file, and then estimates various parameters of nuclear size, chromatin regularity, and nuclear rim regularity.
Interactive cell segmentation.
IMAPATH [ed: now ISAP] requires the user to select a block containing each cell of interest, using the mouse. The system then isolates the nuclear region as an object of edge-contiguous pixels within the selected-block by the following algorithm:
1. The pixel at the center of the selected-block is forced to be an `inside-pixel' (i.e., inside the nucleus).
2. For each inside-pixel (starting with the center of the selected-block), the system recruits the four edge-contiguous pixels as `candidate-pixels'.
3. For each candidate-pixel not already known to be an inside-pixel, the system determines the `average-height' of the candidate-pixel as follows. By default, the average-height is the arithmetic mean of the pixels in a 3x3 pixel window, with the candidate-pixel at the center.
4. If the average-height of the candidate-pixel is above a predetermined threshold, then the candidate-pixel becomes an inside-pixel. If additional candidate-pixels remain, then go to 2. If no additional candidate-pixels remain, then stop.
5. The user may adjust the window-size (3x3 pixels, 5x5 pixels, etc.), the threshold value (grayvalue between 0 and 255), or the average-height calculation (arithmetic mean, geometric mean, minimum pixel value, maximum pixel value).
6. A pixel is an `edge-pixel' if it has at least one edge-contiguous pixel inside the nucleus and at least one edge-contiguous pixel outside the nucleus.
Nuclear Measurements.
Once a nucleus has been outlined (segmented), several parameters are estimated and added to an image database (Figure 1). These parameters serve to quantitate features of nuclear morphology considered important in detecting preneoplasia, namely, nuclear size (AREA, GRAYSUM, DENSESUM, DENSEAVG, AVGDIAM, see below), chromatin regularity (DENSESTDV), and nuclear rim regularity (MAXDIAM, MINDIAM, ECCENT, PERIMETER, CONTOUR, FRACTAL). The parameters are calculated as follows: 1) AREA, the total number of pixels inside the nucleus; 2) GRAYSUM, the sum of the grayvalues of pixels inside the nucleus; 3) DENSESUM, the sum of optical densities of pixels inside the nucleus, where the optical density for each pixel is calculated by the formula: optical density =log10(255/(255-grayvalue)), for grayvalues between 0 and 254; 4) DENSEAVG, the average optical density inside the nucleus, calculated as DENSESUM/AREA; 5) DENSESTDV, the standard deviation of optical density values for pixels inside the nucleus; 6) AVGDIAM, the average diameter of the nucleus, assuming that it is a circle with area AREA, namely AVGDIAM=2*sqrt(AREA/pi); 7) MAXDIAM, the (Euclidean) distance between the two most distant edge-pixels of the nucleus; 8) MINDIAM, the minimum diameter of the nucleus, assuming that it is an ellipse with area AREA and maximum diameter MAXDIAM, namely MINDIAM=(2*AREA)/(pi*MAXDIAM); 9) ECCENT, the eccentricity of the nucleus, assuming that it is an ellipse, defined as ECCENT=(MAXDIAM/MINDIAM); 10) (XCENTER,YCENTER), the x-y coordinates for the center of gravity of pixels inside the nucleus, i.e., XCENTER equals the arithmetic mean of x-coordinates and YCENTER equals the arithmetic mean of y-coordinates; 11) PERIMETER, the sum of distances between consecutive edge-pixels around the nucleus;
12) CONTOUR, equal to the PERIMETER/sqrt(AREA), which assumes the value of 3.54 (=2*sqrt(pi)) for a perfect circle, and >3.54 for all other objects [4]; and 13) FRACTAL, the `fractal dimension' of the edge [38].
Fractal dimension for the edge of a two-dimensional object is a measure of the eventual irregularity of that edge. For a classical two-dimensional geometric object (circle, square, polygon), the estimated perimeter of the object eventually approaches a constant as the measuring device is increasingly refined, and the edge assumes a fractal dimension of one. One estimates the perimeter by placing a raster, or grid, over the object, then adding up the distances between consecutive edge points where the object intersects the grid. The grid is refined by closer spacing of the grid lines, i.e., a smaller interval-distance, d. For a (non-classical) fractal object, the sum-of-distances grows indefinitely larger as the grid is increasingly refined, because the fractal edge is indefinitely irregular. The fractal dimension, s, describes the rate at which perimeter grows as a function of decreasing grid-interval-distance, d, as follows:
s = 1+lim(d->0)(log2((PERIMETER at grid 2*d)/(PERIMETER at grid d)). That is, as the grid-interval, d, approaches zero, then the fractal dimension, s, equals the limit of: the log-base-2 of: the PERIMETER estimated at grid-interval 2*d divided by the PERIMETER estimated at grid-interval d. For a classical geometric object, the PERIMETER estimated at grid-interval 2*d almost equals the PERIMETER estimated at grid-interval d, so the quotient approaches 1 and the log-base-2 of the quotient approaches zero. For an object whose perimeter is 50% larger every time the grid-interval is halved, the fractal dimension, s, equals 1+log2(1.5), or 1.xxx. For an object whose perimeter doubles every time the grid-interval is halved, the fractal dimension, s, equals 1+log2(2), or 2. Such an object is so irregular that its edge forms a `space filling curve'. For a computer image with pixel size greater than zero, it is impossible to calculate s for the limit as d approaches zero, since the irregularity of an object in a computer image cannot be estimated beyond the interval-distance of a single pixel. IMAPATH [ed: now ISAP] calculates s by the above formula, where d = 1 pixel.
Database functions.
When IMAPATH [ed: now ISAP] calculates measurements for a nucleus, these measurements are added to a comma delimited database file, that can be imported into popular commercial spreadsheets, such as MicroSoft Excel (MicroSoft Corp, Redmond, WA), Lotus (\\\\\\\\\\), or Quattro-Pro (Borland International, Inc., Scotts Valley, CA). Row data entries are separated by commas (defining the columns), with a carriage return delimiter between rows. Users can view their databases and rename files within the IMAPATH [ed: now ISAP] environment, but further investigations of the data should be performed either with a spreadsheet or with a statistics program.®PG¯
DISCUSSION
Flow cytometry is a popular method for obtaining quantitative information regarding cancerous and precancerous lesions. However, flow cytometry requires large numbers of cells for meaningful analysis. Image analysis is ideal for measurements on small numbers of cells appearing in stained biopsy preparations. Cytologic specimens are particularly valuable because whole cells with intact nuclei are available for measurement. Such is not the case for histologic specimens, whose cells fall in various focal planes and whose nuclei are truncated by sectioning [10,45,47]. A variety of commercial and public domain image analysis systems have been used to study both histologic and cytologic specimens [1,9,11,48]. Of particular interest is Image, a public domain image capture and analysis system developed exclusively for the Macintosh computer (Apple Computer, Cupertino, CA) and available by modem from InterNet (anonymous FTP zippy.nimh.nih.gov[128.231.98.32]).
Precancerous lesions have been identified for a wide variety of tissues. Precancerous lesions of epithelial tissues are usually designated by a three-letter acronym beginning with the first letter of the tissue and ending with the first letters of `Intraepithelial Neoplasia' (e.g., CIN, VIN, PIN, for cervical, vulvar, or prostatic intraepithelial neoplasia, respectively). Nonepithelial tissues (connective tissue, bone, lymph nodes) also have described preneoplastic conditions [39]. The number of classified precancerous lesions is much smaller than the number of cancer entities. It is our view that a single morphologic variant of a preneoplastic lesion may be the precursor for a variety of morphologically distinct cancers. For instance, there is only one recognized morphologic precursor for bronchial cancer, namely, squamous dysplasia. Some researchers have concluded that squamous dysplasia is a precursor lesion for squamous carcinoma only. Others hypothesize a common origin for all lung carcinomas, implying that bronchial squamous dysplasia may give rise to bronchogenic adenocarcinoma, squamous cell carcinoma, small cell carcinoma, and large cell carcinoma.
A variety of morphometric parameters have been proposed which may detect preneoplastic cells in a variety of tissues, including cervix [8,19], liver [44], ovary [18], breast [24], prostate [21,35], oral cavity [21], kidney [5], and esophagus [23]. In general, once an image has been successfully segmented, information from the image file can be used to evaluate many mathematical parameters [27], using many different algorithms relating to contour [4,33,46] and texture [17,19,20,25]. One of the most useful quantitative measurements in image analysis is nuclear DNA content [8,36].
Limitations in the value of mathematical parameters for characterizing preneoplastic lesions are imposed by specimen preparation [29,45], interobserver differences in cell selection [42], limits on image resolution, glare, and other noise that reduces measurement accuracy [6,41], reproducibility between laboratories [30], selection of measurements with no biologic relevance, and common statistical errors [22], including counting methodology [28], small sample size [45], inadequate sizes of training and test sets for multivariate regression or neural network analyses [2], and the lack of proven clinical utility of cytometry for certain types of pathologic specimens [49].
All software used in the present report was written at U.S. government installations, and as such is available in the public domain. Commercial software is also available that can segment images (mark the boundaries of cells or nuclei) and a wide variety of image measurements pertaining to nuclear features of size, contour, summed optical density and texture. However, the actual computational algorithms used in commercial software is often proprietary, making the enterprise of image analysis a "black box." Furthermore, most image analysis software leaves the matter of image calibration up to the individual researcher, along with all the requisite problems of dealing with image noise, image reproducibility and camera performance. As yet, there are no established standards for dealing with these important problems. For this reason, it is important that images and glass slides must be transferrable between remote laboratories so that interlaboratory quality assurance can assure the validity of image measurements.
Image capture and analysis can be carried out at a hardware cost less than that of a single microscope. Image analysis is currently being tested as a potential aid in the diagnosis of preneoplastic lesions of cervix. As as the cytometric properties of preneoplastic lesions in other tissues becomes clarified, image analysis will have an increasingly prominent role in the early detection and prevention of cancer.
REFERENCES
1. Baak J.P.A., Lowe S., Oort J., Ploem J.S. (1991) Equipment for quantitative cell and tissue analysis. In: Manual of Quantitative Pathology in Cancer Diagnosis and Prognosis, pp. 57-76. Editor: J.P.A. Baak, Springer-Verlag, Berlin.
2. Baak J.P.A., Langley F.A., Hermans J. (1991) Classification and Prognosis for New Cases: Some Aspects of Univariate and Multivariate Analysis. In: Manual of Quantitative Pathology in Cancer Diagnosis and Prognosis, pp. 189-209. Editor: J.P.A. Baak, Springer-Verlag, Berlin.
3. Bailar J.C., Smith E.M. Progress against cancer? (1986) New Engl. J. Med., 314,1226-1232.
4. Barcos M. (1993) Mycosis fungoides: diagnosis and pathogenesis. Am. J. Clin. Pathol., 99,452-458. 5. Banner B.F., Ernstoff M.S., Bahnson R.R., Titus-Ernstoff L., Taylor S.R. (1991) Quantitative DNA analysis of small renal cortical neoplasms. Human Pathol., 22, 247-253.
6. Bartels P.H. (1991) Videophotometry: sources of error. In: Manual of Quantitative Pathology in Cancer Diagnosis and Prognosis, pp. 182-188. Editor: J.P.A. Baak, Springer-Verlag, Berlin.
7. Berry AV, Baskind AF, Hamilton DG. (1981) Cytologic screening for esophageal cancer. Acta Cytol 25,135-141.
8. Bibbo M., Dytch H.E., Alenghat E., Bartles P.H., Wied G.L. (1989) DNA ploidy profiles as prognostic indicators in CIN lesions. Am. J. Clin. Pathol., 92, 261-265.
9. Bibbo M., Dytch H.E., Bartels P.H., Wied G.L. (1986) Clinical applications for an inexpensive, microcomputer-based DNA cytometry system. Acta Cytol. 30,372-378.
10. Bosari S., Lee A.K.C., Wiley B.D., Heatley G.J., Silverman M.L. (1993) Flow cytometric and image analysis of colorectal adenocarcinoma: a comparative study with clinical correlations. Am. J. Clin. Pathol., 99,187-194.11. Check W.A. (1990) Pap smear's future speeding up. CAP Today, 4,1-23.
12. Chu J., White E. (1987) Decreasing incidence of invasive cervical cancer in young women. Am. J. Ob. Gyn. 157,1105-1107..
13. Dayhoff R.E., Maloney D.L. (1992) Exchange of Veterans Affairs medical data using national and local networks. Ann New York Acad Sci 50-66
14. Eddy D.M. (1990) Screening for cervical cancer. Ann. Intern. Med., 113,214-226.
15. Eddy D.M. (1987) The frequency of cervical cancer screening: comparison of a mathematical model with empirical data. Cancer, 60, 1117-1122.
16. Frable W.J. (1992) Letter. CAP Today, 6:51 (inclusive).
17. Galloway M.M. (1975) Texture analysis using gray run length. Computer Graphics Image Processing 4, 172-179.18. Gil J., Deligdisch L. (1989) Interactive morphometric procedures and statistical analysis in the diagnosis of ovarian dysplasia and carcinoma. Pathol. Res. Pract., 185, 680-685.
19. Hanselaar A.G., Vooijs G.P., Van't Hof-Grootenboer A.E., Pahlplatz M.M. (1990) Cytophotometric analysis of cervical intraepithelial neoplasia grade III, with and without synchronous invasive squamous cell carcinoma. Cytometry, 11, 901-906.
20. Haralick R.M., Shanmugan T., Dinstein I. Textural features for image classification. IEEE Trans. Syst. Man. Cybern. S.M.C., 3, 610-621.21. Helpap B. (1988) Observations on the number, size and localization of nucleoli in hyperplastic and neoplastic prostatic disease. Histopathology, 13, 203-211.
22. James N.T. (1989) Common statistical errors in morphometry. Pathol. Res. and Pract., 185, 764-768.
23. James P.D., Atkinson M. (1989) Value of DNA image cytometry in the prediction of malignant change in Barrett's oesophagus. Gut 30, 899-905.
24. King E.B., Chew K.L., Duarte L., Hom J.D., Mayall B.H., Miller T.R., Petrakis N.L. (1988) Image cytometric classification of premalignant breast disease in fine needle aspirates. Cancer, 62, 114-124.
25. Komitowski D., Janson C. (1990) Quantitative features of chromatin structurein the prognosis of breast cancer. Cancer 65, 2725-2730.
26. Koss L.G. (1989) The Papanicolaou cervical test for cancer detection. JAMA, 261, 737-743.
27. Kunze K.D., Haroske G., Dimmer V., Meyer W., Theissig F. (1989) Grading and prognosis of invasive ductal mammary carcinoma by nuclear image analysis in tissue sections. Pathol. Res. Pract. 185, 689-93.
28. Laroye G.J., Minkin S. (1991) The impact of mitotic index on predicting outcome in breast carcinoma: a comparison of different counting methods in patients with different lymph node status. Mod. Pathol., 4, 456-460.
29. Linder J. (1991) The coming era of cytologic automation. Am. J. Clin. Pathol., 96, 293-294.
30. Montironi R., Whimster W.F. (1990) Diagnositic quantitative pathology. Pathol. Res. Pract., 1286, 817-820.31. Moore G.W., Berman J.J., Moore G.W., Brown L.A., Richie S.R., Kuzmak P.M., Dayhoff R.E. (1993) Pathology image analysis in a networked image capture system. Computer Applications for Early Detection and Staging of Cancer. National Library of Medicine, Bethesda, MD
32. Moore G.W., Moore G.W., Berman J.J., Brown L.A. (1993) Impath Version 1.0, Baltimore VA Medical Center, Baltimore, MD
33. Murphy G.F., Partin A.W., Maygarden S.J., Mohler J.L. (1990) Nuclear shape analysis for assessment of prognosis in renal cell carcinoma. J. Urol., 143, 1103-1107.
34. Ng A.B.P., Reagan J.W. (1988) Pathology and Cytopathology of microinvasive squamous cell carcinoma of the uterine cervix. In: Compendium on Diagnostic Cytology, 6th edition, pp.114-122. Editors: Wied G.L., Keebler C.M., Koss L.G., Reagan J.W. Tutorials of Cytology, Chicago.
35. Petain M., Michel P., van Velthoven R., Pasteels J., Brawer M.K., Davis J.R., Nagle R.B., Kiss R. (1991) Morphonuclear relationship between prostatic intraepithelial neoplasia and cancers as assessed by digital cell image analysis. Am. J. Clin. Pathol. 96, 628-634.36. Prenna G., Leiva S., Mazzini G. (1974) Quantitation of DNA by cytofluorometry of the conventional Feulgen reaction. Histochem. J., 6, 467-489.
37. Russ J.C. (1992) The Image Processing Handbook, pp 5-6. CRC Press, Boca Raton.
38. Sanders H., Crocker J. (1993) A simple technique for the measurement of fractal dimensions in histopathological specimens. J. of Pathol., 169, 383-385.39. Seidman J.S., Berman J.J. (1993) Premalignant non-epithelial lesions: a biological classification. In press, Modern Pathology, September.
40. Shu S.J. (1983) Cytopathology of the esophagus: an overview of esophageal cytopathology in China. Acta Cytol 27, 7-16.
41. Smeulders A.W.M., Kate T.K. (1987) Accuracy of optical density measurements of cells I. Low resolution. Appl. Opt. 26, 3249-3257.
42. Smeulders A.W.M., Dorst L. (1985) Measurement issues in morphometry. Analyt. Quant. Cytol. Histol. 7, 242-249.
43. Takahashi K. Outline of gastric mas curvey by x-ray. In, Early Gastric Cancer, Murakami T, ed., Univ. of Tokyo Press, 1971, pp 21-26.
44. Thomas R.M., Berman J.J., Yetter, R.A., Moore G.W., Hutchins G.M. (1992) Liver cell dysplasia: a DNA aneuploid lesion with distinct morphologic features. Human Pathol., 23, 496-503.
45. Van Diest P.J., Smeulders A.W.M., Thunnissen F.B.J.M., Baak J.P.A.(1989) Cytophotometry: a methodological study on preparation techniques, selection methods and sample size. Analyt. Quant. Cytol. Histol., 11, 225-231.
46. van Diest, P.J., Baak J.P.A. (1991) Morphometry. In: Bibbo M, ed., Comprehensive Cytopathology, pp. 946-964. Editor: Bibbo M. W.B. Saunders, Philadelphia.
47. Van Driel-Kulker A., Mesker W., Van der Barg M., Ploem J. Preparation of cells from paraffin-embedded tissues for cytometry and cytomorphologic evaluation. Anal. Quant. Cytol. Histol., 9, 225-231.48. Wells W.A., Rainer R.O., Memoli V.A. (1993) Equipment, standardization and applications of image processing. Am. J. Clin. Pathol. 9, 48-56.49. Wheeless L.L., Badalament R.A., deVere White R.W., Fradet H., Tribukait B. (1993) Consensus review of the clinical utility of DNA cytometry in bladder cancer. Cytometry, 14, 478-481.
Last modified: January 6, 2008
