Objective malignancy grading of squamous cell ... - Wiley Online Library

2 downloads 0 Views 572KB Size Report
Sep 1, 1995 - The authors thank Hans Jargen G. Gundersen for guidance in the field of ..... Lee T-K, Horner RD, Silverman IF, Chen Y-H, Jenny C, Scaran-.
797

Objective Malignancy Grading of Squamous Cell Carcinoma of the Lung Stereologic Estimates of Mean Nuclear Size Are of Prognostic Value, Independent of Clinical Stage of Disease Morten Ladekarl, M.D.,* Torben Bek-Hansen, M.D.,t Regitze Henrik-Nielsen, M.D.,t Christian Mourifzen, M.D.,$ Ulrik Henriques, M.D.,t and Flemming Brandt Sgrensen, M.D.*,t

Background. The prognostic value of quantitative histopathologic parameters was evaluated in 55 consecutively treated patients with operable lung carcinoma of squamous (N = 39) and mixed, adenosquamous (N = 16) cell type. Patients alive were followed for at least 12 years. Methods. Using a projection microscope and a simple test system in fields of vision systematically selected from the whole tumor area of one routine section, five quantitative histopathologic variables were estimated: the mean nuclear volume, the mean nuclear profile area, the density of nuclear profiles, the volume fraction of nuclei to tissue, and the number of mitotic profiles per lo3 nuclear profiles. For each patient, information was recorded regarding sex, age at diagnosis, and clinical stage of disease. Results. Single-factor analyses showed that a favorable prognosis was associated with early clinical stages (Stages I and 11)and young age (P5 0.03), and that females tended to do better than males (P = 0.09). Estimates of the mean nuclear volume were of prognostic significance ( P = 0.02), small nuclei being associated with the worst progFrom the *Stereological Research Laboratory, University Institute of Pathology and Second University Clinic of Internal Medicine, Institute of Clinical Experimental Research, University of Aarhus, TUniversity Institute of Pathology, Aarhus Kommunehospital, and Department of $Chest Surgery, Aarhus University Hospital, Skejby Sygehus, Aarhus, Denmark. Supported by grants from the Danish Cancer Society, Foersom's Foundation, the Danish Foundation for the Advancement of Medical Science, Lily B. Lund's Foundation, Ib Henriksen's Foundation, the Novo Nordisk Foundation, Else & Mogem Wedell-Wedellsborg's Foundation, M. C. & J. K. Moltum's Foundation, Aage Bang's Foundation, and Frits, Georg & Marie Cecilie Glud's Foundation. The authors thank Hans Jargen G. Gundersen for guidance in the field of stereology and Maj-Britt Lundorff for technical assistance. Address for reprints: Morten Ladekarl, M.D., Stereological Research Laboratory, Bartholin Building, University of Aarhus, DK8000 Aarhus C, Denmark. Received January 30, 1995; revision received April 21, 1995; accepted April 21, 1995.

nosis. In a multivariate Cox analysis, clinical stage, age, and mean nuclear volume were found to be parameters of significant, independent prognostic value. Conclusions. The present feasibility study indicates that estimates of the mean nuclear volume are of prognostic value, independent of the clinical stage of disease. This quantitative histopathologic variable is highly reproducible and easily obtained using an unbiased stereologic method. Thus, the mean nuclear volume may be a parameter of future importance in the clinical management of patients with carcinoma of the lung. Cancer 1995;76:797-802. Key words: lung cancer, morphometry, nuclear volume, prognosis, squamous cell, stereology.

Lung cancer is often a rapidly lethal disease with a reported 5-year survival ranging from 1O-35%.',2 However, a subset of operable patients may have a relatively favorable prognosis if treated adequately. To identify these patients, who might benefit from intensive treatment, and to avoid overtreating those who follow an inevitable lethal course, reliable prognostic factors are needed. The histologic type of lesion is a clinically important prognostic variable. For example, small cell lung cancers are known to be particularly aggre~sive,~ but are, in contrast to most other lung carcinoma types, highly responsive to chemotherapeutic treatment.4 In the current study, we focus on patients with tumors of the squamous cell type. This is a common type of tumor, constituting 40-60% of all cases3z5;it is predominantly found in men,5 nearly always associated with smoking,6 and it is generally considered insensitive to chemotherapy * Even within individual morphologic types, lung carcinomas exhibit a very heterogeneous picture regarding disease course and outcome. This may, in part,

798

CANCER September I , 2995, Volume 76,No. 5

be due to differences in the lesion's malignant potential, which is thought to be associated with the degree of histologic differentiation. However, in squamous cell carcinoma, conflicting results regarding the prognostic value of histologic malignancy grading have been published.'r9 The discrepancies may reflect observer variation associated with the morphologic, qualitative evaluation. Objective grading by direct measurements in tissue sections may improve the validity of parameters obtained at the microscopic examination. Quantitative histopathologic variables are often highly reproducible,'"-I2 and in contrast to conventional grading schemes, the resulting estimates are displayed on a continuous scale, reflecting a continuum of tumor biology. From ordinary histologic sections, one- or two-dimensional quantitative histopathologic parameters may be obtained using morphometric methods. Additionally, stereologic techniques can be used for estimation of realistic, three-dimensional variables. In the current study, we investigated the prognostic value of stereologic and morphometric variables describing nuclear and mitotic features and of conventional clinical parameters. The relative, independent significance of prognostic variables was tested using a multivariate statistical approach. Materials and Methods

Pa tien ts Fifty-five patients with operable lung carcinoma of squamous or mixed adenosquamous cell type, diagnosed from 1979 through 1981 at the University Institute of Pathology, Aarhus Kommunehospital, were consecutively included in the study. The patients were surgcally treated, either by lobectomy or by pneumectomy. Based on postoperative data, disease staging was performed according to the TNM clas~ification.'~ The survival status of patients was extracted from the Central Personal Registry, and causes of deaths were established from death certificates or from autopsy reports. Censoring occurred on the date of death caused by any other disease than lung cancer. The median follow-up was 1.6 years. Patients alive were followed for at least 12 years. The age at diagnosis ranged from 48 to 76 years, with a median of 61 years.

Histologic Specimens At the time of diagnosis, routinely processed sections of tumors were typed according to the World Health Organization clas~ification.'~ The diagnosis of mixed adenosquamous tumors was based on special stains for

Figure 1. Microscopic field of vision from a squamous cell-type carcinoma of the lung projected onto a test system. For the estimation of the mean nuclear volume, all profiles of cancer nuclei hit by points and in focus are sampled. The large and small frames are integrated for counting, respectively, mitotic profiles and nuclear profiles, using an unbiased counting rule" (€1& E, original magnification X 1850).

mucus, requiring both squamous and adenosquamous differentiation in the tumor bulk. For quantitative histopathology, sections were cut at 4 pm from the original paraffin embedded tissue blocks, stained with hematoxylin and eosin, and examined by one observer. The methods used have been described in detail earlier.15,'6In short, a standard microscope equipped with a projection attachment and a lOOX oil lens (numerical aperture, 1.4) was used for measurements, which were performed at a final magnification of 1850X. Fifteen microscopic fields of vision were sampled systematically from the whole tumor area of the specimen, excluding areas of heavy inflammation, necrosis, or nuclear pyknosis. Each field of vision was projected onto a test system with points and two counting frames (Fig. 1). The mean nuclear volume, vv(nuc), was estimated using the stereologic method of "point-sampled intercepts" Is: for each point in the test system that hit a nuclear profile in focus, the distance from nuclear border to nuclear border was measured through the sampling point in one arbitrary direction. These so-called "intercept lengths" (lo) were individually cubed, and the average value multiplied with a/3 to obtain an unbiased, volume-weighted estimate of the mean nuclear size. It was assumed that the nuclei, globally, did not possess a preferred direction of orientation in space. A mean number of 126 intercepts (range, 71196) were measured per tumor. Using the points and counting frames of the test system and formulas shown in Table 1, estimates were obtained of the mean area of nuclear profiles, &(nuc), the number of nuclear profiles per square millimeter tumor, ND, the volume fraction

Objective Grading of Lung CancerlLadekarl et al. Table 1. Formulas Used for Calculation of Quantitative HistopathologicVariables Mean nuclear volume ? r -

799

for obtaining Gv(nuc).This analysis showed that the relative contribution to the total observed interpatient variance from measurement ”noise” was only 1870, whereas 82% could be attributed to biologic variation.

Vv (nuc) = - . 1; 3

Results

Mean nuclear profile area

Nuclear volume fraction Vv (nuc/tis)

=

Pp (nuc/tis)

N ‘ a(p) =_ _ n,.A

Nuclear profile density

ND=-

(nu4 nf,Al

QI

Mitotic profile frequency

MF =

Q(mit).A,

QI(nut). A

A: area of large counting frame; A,: area of small counting frame; a(p): area nuclear intercept length; N: associated with each point in the test probe; lo: total number of points hitting nuclei; n, number of fields of vision; Pp (nuc/tis): fraction of points hitting nuclei per tissue; Q(mit): total number of mitotic profiles counted; Ql(nuc): total number of nuclear profiles counted.

of nuclei to tissue, Vv(nuc/tis), and the number of mitotic profiles per 1000 nuclear profiles, MF. To obtain these variables, a mean number of six mitotic profiles were counted per tumor (range, 0-17) using the large counting frame, whereas an average of 80 nuclear profiles (range, 42-187) were counted using the small counting frame.

Statistics Differences between group means were tested with a Student’s f test, whereas a least-square linear regression analysis was used for investigating the correlation between continuous parameters. The distributions of discontinuous and categoric parameters were tested by Kendall’s 7.19Using the BMDP Statistical Software (Los Angeles, CA), univariate survival analyses were performed by log rank tests and illustrated by KaplanMeier plots. A multivariate survival analysis was performed using Cox’s models with standard stepwise, backward elimination, removing variables at P > 0.06. The model assumption of proportional hazard rates was checked by log-minus-log plots for quantitative parameters after categorizing data into equally sized groups. Quantitative data were, in all other cases, analyzed using the median values as cutoff point. In the tests, which were all double-sided, P < 0.05 was considered the level of significance. A nested analysis of variance” was used to estimate the efficiency of the sampling scheme used

Baseline data of quantitative histopathologic variables are shown in Table 2 . Several significant associations were found between these parameters, strongest between Tv(nuc)arid ZH(nuc)(r = 0.72), between ND and Vv (nuc/tis) (r = 0.56), and between ND and ZH (nuc) (r = -0.55). Age and ND were significantly associated (r = 0.33), whereas no significant differences in means of quantitative variables were seen regarding clinical stage or histologic tumor cell type. Sex and histologic type were not related to clinical stage, whereas men were on average 8 years older at diagnosis than women ( P = 0.01). Finally, significantly more patients were pneumectomized in advanced clinical disease stages (7 = 0.43). At the end of follow-up, 5 patients were alive, 37 had died of lung cancer, and 13 had died of other diseases. The results of single-factor survival analyses for the variables investigated are shown in Table 3. Early stage of disease, large Tv(nuc),and young age were significant, favorable prognostic signs, whereas a marginally significant survival advantage for women compared with that of men was found (Fig. 2). In a multivariate Cox’s model, including variables of significant and marginally significant value by singlefactor analysis, only sex was rejected (Table 4). Thus, clinical stage of disease, &(nuc), and age were significant, independent prognostic indicators. From the relative strength of these variables, a prognostic index (PI) was constructed:

PI = clinical stage X 0.65 - Vv(nuc)X 0.0015

+ age X 0.039.

The constants correspond to the estimated P-values of the Cox analysis. The prognostic index ranged from 1.27 to 4.29. Using the median value of 2.58 as a cutoff point, a clear separation in prognostically poor and favorable cases was achieved (Fig. 3). The 5-year survival rate for patients with a prognostic index larger and smaller than the median was 8% and 67%, respectively, and all long term survivors had a prognostic index below median. Discussion

The results of the current study confirm that stage of disease is a parameter of great prognostic and clinical value. Within individual stages, patients may show very

800

CANCER September 1,2995, Volume 76, No. 5 Table 2. Baseline Data of Quantitative Histopathologic Variables Data Mean Median Range

vv (nuc) bm3) 716 71 7 293-1460 (0.35)

-

aH(nuc) (pm2) 48.0 47.6 27.6-92.1 (0.22)

ND (mrn-*)

Vv (nuc/tis)

MF

3790 3550 1990-8850 (0.29)

0.176 9.21 0.172 8.62 0.099-0.272 0.00-24.6 cv (0.20) (0.59) ZH (nuc): mean nuclear profile area; M F mitotic profile frequency; ND: nuclear profile density; Vv (nuc): mean nuclear volume; V, (nuc/tis): nuclear volume fraction; CV: coefficient of variation (standard deviation/mean).

different disease courses, however, and variables of prognostic value, independent from clinical stage, are therefore urgently needed. We found stage and type of operation to be significantly associated, but this is understandable, because more extensive surgery is often performed for advanced tumor stages. In the univariate survival analysis, females tended to do better than males. The difference could, however, be explained by the younger mean age of female patients, and accordingly, in the multivariate analysis, sex was not a variable of independent significance, In this study, we concentrated on squamous and mixed adenosquamous cell carcinoma. Other types of nonsmall cell lung cancer were not considered, because they have been associated with a generally less favorable p r o g n ~ s i s ~and * ~therefore ~ , ~ ~ should be evaluated separately. We included mixed adenosquamous cell type carcinomas due to the pronounced heterogeneity of lung cancer^,^ in which mixtures of cell types are often found if a number of sections from the same tumor are i n ~ e s t i g a t e dNo . ~ ~difference in prognosis between patients with pure and mixed tumors was seen, and the quantitative histopathologic variables obtained in the two tumor types did not differ significantly either. Among the estimated quantitative histopathologic

variables, we found prognostic value of the mean nuclear volume only. Studies of other kinds of cancer have shown a significant correlation of this parameter with p r o g n ~ s i s , ' ~although ~ ' ~ ~ ~ ~in*some ~ ~ cancer types it may be without prognostic effect.26The mean nuclear volume seems suitable for routine use: it has been shown to be highly it can be obtained with sufficient accuracy in about 15 minutes using simple e q ~ i p m e n t , ' ~ ,and ' ~ , ~it~seems robust to variations of the tissue pro~essing.'~ In the current study, the mean nuclear volume was of prognostic value, independent of clinical disease stage, and it may, therefore, be used to stratify patients to a more individualized treatment than is possible by clinical stage alone. Although the current study does not investigate the ability of variables to predict the effect of treatment, aggressive tumors with small mean nuclear size might be more sensitive to chemotherapy, as are small cell-type carcinomas. The average Vv(nuc) of the current study (of 716 pm3)is comparable to the findings of Aru and Nielsen,28 who reported a mean V,(nuc) of 632 pm3 in 10 squamous cell lung carcinomas and a mean of 736 pm3in 10 adenocarcinomas. They found no overlap of C,(nuc) in these tumor types with the Vv(nuc)of 10 small cell car-

Table 3. Results of Single-Factor Prognostic Analyses of Clinicopathologicand QuantitativeHistopathologic Variables ~~

Variable Clinicopathologicparameters Sex Age Type of operation Clinical stage of disease Tumor cell type Quantitative histopathologic variables %(nut) ZH (nu4 ND Vv(nuc/tis) MF

Cutoff point Male/female Median Lobectomy/pneumectomy I/II/III Squamous/adenosquamous Median Median Median Median Median

No.of cases

P value

48/7 30/25 11/44 22/12/21 39/16

0.09 0.03 0.33 0.01 0.98

28/27 29/26 29/26 28/27 28/27

0.02 0.12 0.47 0.69 0.85

ZH(nuc):mean nuclear profile area; M F mitotic profile frequency; ND: nuclear profile density; Vv(nuc): mean nuclear

volume; V&uc/tis): nuclear volume fraction.

801

Objective Grading of Lung Cancer/Ladekaul et al. 10-

a

C ..... stage I

~ - stage _ I

- stageri

N=7 8~-..

$

a-

L-----

II

I_

I

----; L. ............

_-__-

~

’ L..

N=12

............. N=22

O! 0

1

I

lr

5

15

N=21

0 , 0

1

5

15

10

d

b . . . . .s 61 years ->

q\-, L....

g0.S Figure 2 . Plots of disease specific survival according to (a) patient sex ( P = 0.09), (b) age at diagnosis ( P = 0.03), (c) clinical stage of

a ?

.-

-

61 years

-iy(nuc)’717flm’

,

..~

I .

__.._ o V(nuc)r717pm’

. ~

.....

,-.. ........ . .b

-

N=27 .........N=30 .....

L ...

-. L..

L..-

N=25

5 u 01

..............

1......... N=28 I

cinomas, and, including the current 55 cases, this is still true. Thus, &(nuc) might, in puzzling cases, be useful for diagnostic separation of small cell and nonsmall cell carcinomas. Quite unexpectedly, we found that mean nuclear size is inversely correlated with prognosis, that is, patients with tumors having small nuclei do worse. A ”negative” association has also been found in squamous cell-type carcinomas of the uterine cervix.29The biologic explanation for the inverse relation between v,(nuc) and prognosis is, however, speculative. Although significantly correlated with Sv(nuc), two-dimensional estimates of the mean nuclear profile area were without prognostic significance. This finding is in agreement with a previous Additionally, others found no significance of one-dimensional estiIt theremates of the mean largest nuclear fore seems that only three-dimensional nuclear size estimates carry prognostic information.

Conflicting results have been reported regarding the prognostic value of mitotic counts in lung cancer. An earlier investigation indicated a possible prognostic ~ignificance,~~ which we, as well as others,8,12failed to reproduce, Apart from studying different subsets of patients, some of the discrepancies might be explained by differences in sampling methodology using selective sampling of “worst” areas or a systematic sampling approach, as performed in the current study. Although the latter approach is most efficient in fighting tumor heterogeneity, sampling from several sections may be necessary to obtain stable estimates.32 In conclusion, the current study of squamous and mixed adenosquamous cell type lung cancer shows that stereologic estimates of mean nuclear size are of significant prognostic value, independent of clinical stage. This variable might be clinically useful for stratification of patients within stages to different treatment schemes,

Table 4.Results of the Multivariate Cox Survival Analysis First model

Variable* Sex Age

Vv (nut) Clinical stage

B -0.77 0.031 -0.0014 0.63

B/SE -1.02 1.49

-1.75 3.14

Final model P value

B

0.26 0.14

0.039 -0.0015 0.65

0.08 0.002

B/SE

P value

-

-

1.92

0.058 0.058 0.001

-1.86 3.23

p: regression coefficient estimated by maximizing the likelihood function in the Cox analysis; SE: standard error of 6; v,(nuc): mean nuclear volume. * Sex scored 1 for female and 2 for male: clinical staee scored 1.2. or 3: aee and Cdnuc) entered on a continuous scale.

CANCER September I , 1995, Volume 76, No. 5

802

_____

Pls2.58

11.

__ Pl.2.58

12. 13. 14. 15.

16.

N=27

0

I

0

5

1

lo

1

15

Years Figure 3. Plot of disease specific survival according to the prognostic index, based on information about clinical stage of disease, mean nuclear volume, and age at diagnosis.

based on the expected outcome of disease. However, the results of the current study need to be extended to other types of lung carcinoma and should be confirmed in larger studies.

17.

18.

19.

20.

21.

References 22. 1. Rosado de Christenson ML, Templeton PA, Moran CA. Bronchogenic carcinoma: radiologic-pathologic correlation. Radiographics 1994; 14:426-46. 2. Melamed MR, Flehinger BJ, Zaman MB. Impact of early detection on the clinical course of lung cancer. Surg Clzn North A m 1987; 67~909-24. 3. Kayser K, Bulzebruck H, Probst G, Vogt-Moykopf I. Retrospective and prospective tumor staging evaluating prognostic factors in operated bronchus carcinoma patients. Cancer 1987;59:35561. 4. Raghavan D, Perez R, Creaven P, Takita H, Loewen G, Vaickus, L. Carboplatin for small cell lung cancer: progress toward greater efficacy and reduced toxicity. Semiti Oncol 1994;Zl:l-8. 5. Muller K-M. Histological classification and histogenesis of lung cancer. EurJRespZr Drs 1984;65:4-19. 6. Capewell S, Sankaran R, Lamb D, McIntyre M, Sudlow MF. Lung cancer in lifelong non-smokers. Edinburgh Lung Cancer Group. Thorax 1991;46:565-8. 7. Crump M. Postoperative chemotherapy for resected non-smallcell lung cancer. World I Surg 1993; 17:735-40. 8. Lee T-K, Horner RD, Silverman IF, Chen Y-H, Jenny C, Scarantino CW. Morphometric and morphologic evaluations in stage 111 non-small cell lung cancers: prognostic significance of quantitative assessment of infiltrating lymphoid cells. Cancer 1989;63:309-16. 9. Chung CK, Zaino R, Stryker JA, O’Neill M, DeMuth WE. Carcinoma of the lung: evaluation of histological grade and factors influencing prognosis. Ann Thorac Surg 1982;33:599-604. 10. Sdrensen FB, Sasaki M, Fukuzawa S, Yamabe H, Olsen S, Yoshida 0. Qualitative and quantitative histopathology in transitional cell carcinomas of the urinary bladder: an international

23.

24.

25.

26.

2 7. 28. 29.

30.

31. 32.

investigation of intra- and interobserver reproducibility. Lab Invest 1994; 70:242-54. Baak JPA, Ladekarl M, Sdrensen FB. Reproducibility of mean nuclear volume and correlation with mean nuclear area in breast cancer: an investigation of various sampling schemes. Hum Pathol 1994;25:80-5. Weiss W. The mitotic index in bronchogenic carcinoma. A m Rev Respir Dis 1971; 104:536-43. Hermanek P, Sobin LH, editors. TNM classification of malignant tumours. 4th ed. Berlin: Springer Verlag, 1987. World Health Organization. Histological typing of lung tumours. 2nd ed. Geneva: World Health Organization, 1981. Sdrensen FB. Quantitative analysis of nuclear size for objective malignancy grading: a review with emphasis on new, unbiased stereologic methods. Lab Invest 1992;66:4-23. Ladekarl M, Sdrensen FB. Prognostic, quantitative histopathologic variables in lobular carcinoma of the breast. Cancer 1993;72:2602-11. Gundersen HJG. Notes on the estimation of the numerical densityof arbitraryprofiles: theedgeeffect. IMicrosc 1977; 111:21923. Gundersen HJG, Jensen EB. Stereological estimation of the volume-weighted mean volume of arbitrary particles observed on random sections. JMicrosc 1985; 138:127-42. Sokal RR, Rohlf FJ. Biometry. The principles and practice of statistics in biological research. 2nd ed. New York: W.H. Freeman and Company, 1981. Gundersen HJG, Bsterby R. Optimizing sampling efficiency of stereological studies in biology: or ’do more less well!’. J Microsc 1981; 121:65-73. Lipford 111EH, Sears DL, Eggleston JC, Moore GW, Lillemoe KD, Baker RR. Prognostic factors in surgically resected limited-stage, nonsmall cell carcinoma of the lung. A m Suvg Pafhol 1984; 8: 357-65. Mountain CF, Lukeman JM, Hammar SP, Chamberlain DW, Coulson WF, Page DL, et al. Lung cancer classification: the relationship of disease extent and cell type to survival in a clinical trials population. J Surg Oncol 1987;35:147-56. Fraire AE, Roggli VL, Vollmer RT, Greenberg SD, McGavran MH, Spjut HJ, Yesnet R. Lung cancer heterogeneity: prognostic implications. Cancer 1987; 60:370-5. Nielsen K, Colstrup H, Nilsson T, Gundersen HJG. Stereological estimates of nuclear volume correlated with histopathological grading and prognosis of bladder tumour. Virchows Arch B Cell Patho1 1986; 52:41-54. Ladekarl M. Quantitative histopathology in ductal carcinoma of the breast: prognostic value of mean nuclear size and mitotic counts. Cancer 1995; 75:2114-22. Mogensen 0, Smensen FB, Bichel P, Jakobsen A. Nuclear volume and prognosis in ovarian cancer. lilt J Gynecol Cancer 1992;2:141-6. Ladekarl M. The influence of tissue processing on quantitative histopathology in breast cancer. I Microsc 1994; 174:93-100. Aru A, Nielsen K. Stereological estimates of nuclear volume in primary lung cancer. Pathol Res Pract 1989; 185:735-9. Serensen FB, Bichel P, Jakobsen A. DNA level and stereologic estimates of nuclear volume in squamous cell carcinomas of the uterine cervix: a comparative study with analysis of prognostic impact. Cancer 1992;69:187--99. Cagle PT,Langston C, Fraire AE, Roggli VL, Greenberg SD. Absence of correlation between nuclear morphomehy and survival in stage I non-small cell lung carcinoma. Cancer 1992;69:24547. James RD, Davey AT. Mitotic index in carcinoma of the lung. Lancet 1979;2:852. Gerstl B, Switzer P, Yesnet R. A morphometric study of pulmonary cancer. Cancer Res 1974; 34:248-54.