Clustering of Breast Microcalcifications: Revisited

Clustering of Breast Microcalcifications: Revisited

Clinical Radiology (2000) 55, 114–118 doi:10.1053/crad.1999.0220, available online at http://www.idealibrary.com on Clustering of Breast Microcalcifi...

404KB Sizes 16 Downloads 127 Views

Clinical Radiology (2000) 55, 114–118 doi:10.1053/crad.1999.0220, available online at http://www.idealibrary.com on

Clustering of Breast Microcalcifications: Revisited J . M . PARK * , H. K . C HOI * , S.- J. B AE * , M. -S. LE E † , S. -H. AHN‡ , G. GONG§ *Department of Diagnostic Radiology, †Department of Preventive Medicine, ‡Department of General Surgery, §Department of Pathology, University of Ulsan, Asan Medical Center, Seoul, Korea Received: 29 October 1998

Revised: 2 April 1999

Accepted: 6 April 1999

AIM: To verify the diagnostic value of the traditional definition of ‘clustering’ of microcalcifications (more than five in the area of 1 cm2 or 1 cm3) on mammography in the differential diagnosis of benign and malignant breast disease. METHODS AND MATERIALS: Three radiologists without knowledge of the final pathology retrospectively counted the number of microcalcifications per 0.25 cm2 (0.5 × 0.5 cm) unit area on mammography in 57 pathologically proven non-palpable lesions including 26 cancers and 31 benign diseases. Pleomorphism of the microcalcifications, associated architectural distortion or mass or increased density and distribution of microcalcifications were also evaluated. RESULTS: The mean numbers of microcalcifications per 0.25 cm2 were 16.4 in malignant and 16.7 in benign diseases (no statistically significant difference between the two groups). Pleomorphism of the microcalcifications, associated architectural distortion or mass or increased density were, however, important determining parameters. Clustering was more frequently observed in benign diseases. CONCLUSION: In this study, the mean number of microcalcifications per unit area is much larger than the traditional definition of ‘clustering’ and clustering itself is not effective in the differential diagnosis of benign and malignant breast lesions. Imaging features other than numbers of calcification per unit area are more important in assessing the significance of mammographic clustered microcalcifications. Park, J. M. et al. (2000). Clinical Radiology 55, 114–118. q 2000 The Royal College of Radiologists Key words: breast, calcification, neoplasm, mammography, diagnosis.

There is an increased risk for malignancy with increasing numbers of mammographically visible microcalcifications [1,2]. When calcifications are detected on mammography, their number, morphologic appearance, size, associated findings and distribution should be examined to characterize them [2,3]. Clustering of microcalcifications is said to be an important diagnostic criterion [4]. However, the definition of ‘clustering’ is variable and there still remains doubt as to ‘what is a significant cluster?’ [5]. According to the literature, the definitions are ‘more than 5 microcalcifications in an area of 1 cm2’ or ‘more than 5 microcalcifications in an area of 0.5 × 0.5 cm’ or ‘more than 5 microcalcifications in the volume of 1 cc’ [2,4,6]. Alternatively, Bird reported that the number of microcalcifications was of little help in analysis [7]. There is considerable overlap in the X-ray appearance of microcalcifications associated with benign and malignant lesions and both frequently show clustered microcalcifications on mammography. There is the impression that classifying calcifications as ‘clustered’ according to the traditional definitions is not discriminatory. We have analysed other features of clustered microcalcifications as well as their numbers to assess the diagnostic efficacy of the traditional definitions of ‘clustering’ and suggest a more practical new definition. Author for correspondence and guarantor of study: Jeong Mi Park, Department of Diagnostic Radiology, University of Ulsan Asan Medical Center, 388-1 Poong Nap-Dong, Song Pa-Ku, Seoul, 138-736, Korea. 0009-9260/00/020114+05 $35.00/0

METHODS

Among 125 patients who underwent mammography-guided hook-wire localization biopsy for lesions between May 1994 and January 1998 in our institution, 57 lesions in 56 patients with microcalcifications on pre-operative mammography were included in this study. Film-screen mammography was performed in 45 patients and digital mammography in 11 patients. Parameters for digital mammography were variable, but specific enhancement for microcalcifications was not used. Only films taken in our institution were used for analysis with films taken elsewhere being excluded. Two mammography units were used – Senographe 500T (GE Medical Systems, Milwaukee, Wis) and Giotto Hi-Tech Mammography (IMS, Bologna, Italy). Patient age range was 30–59 years; mean age 46 years. Specimen radiography was performed to confirm removed microcalcifications. Two views per breast were viewed together to assess any associated architectural distortion, mass or increased density in 54 patients, including one patient who had bilateral subpectoral implantations. Two patients had had previous unilateral mastectomy. Surgery was performed after hook-wire localization biopsy within 2 months of mammography in 54 patients. One patient underwent surgery 3 months and one 8 months after mammography was performed. The histological diagnoses were invasive ductal carcinoma in 10 lesions (17.5%), ductal carcinoma in situ q 2000 The Royal College of Radiologists

115

CLUSTERING OF MICROCALCIFICATIONS

Fig. 1 – An example of more than 30 microcalcifications. Craniocaudal mammography shows more than 30 very fine microcalcifications in the unit area in inner portion of left breast. Pathologic diagnosis was fibrocystic change.

(DCIS) in 16 lesions (28.1%), fibrocystic change in 28 lesions (49.0%), and one each with fibroadenoma, LCIS, atypical ductal hyperplasia (1.8%). Three radiologists retrospectively counted the number of microcalcifications per 0.25 cm2 (0.5 × 0.5 cm) area on the relevant mammography without knowing the final diagnoses (a dedicated breast radiologist, a radiologist who had read mammography for more than 2 years, and a resident who recently rotated through the mammography section). We chose an area of 0.25 cm2 (0.5 × 0.5 cm) rather than 1.00 cm2 (1.0 × 1.0 cm) because the latter was too large and counting the number of microcalcifications was not practical. A piece of transparent film, with two holes each with an area of 0.25 cm2 (0.5 × 0.5 cm) for film-screen mammography (0.43 × 0.43 cm for digital mammography because digital mammography uses an 86% scale), was overlain on the mammography films to count the number of microcalcifications per unit area. The most clustered area was selected each time subjectively so the actual area assessed may have been different for each assessment. We justify this procedure because we wished to simulate as far as possible real practice. Wire localization also selected the most dense cluster for surgical removal. The number of microcalcifications in a cluster was divided into seven categories, 1–4, 5–9, 10–14, 15–19, 20–24, 25–29, and more than 30. Because there was no ‘gold standard’ for the true number of microcalcifications, analyses were done twice with an interval of 1 week by each of the three radiologists and intra- and interobserver agreements (observer 1 vs 2, observer 2 vs 3, observer 1 vs 3) were calculated using kappa and weighted kappa analyses [8]. Values above 0.75 were regarded

Fig. 2 – An example of pleomorphic microcalcifications. Craniocaudal mammography shows 20–24 clustered microcalcifications in the unit area with pleomorphism in central portion of right breast. Pathologic diagnosis was infiltrating ductal carcinoma.

as good correlation, values between 0.4 and 0.75 as fair to good and values below 0.4 as poor correlation [9] (Fig. 1). Pleomorphism of the microcalcifications, associated architectural distortion, mass or increased density were also recorded as either ‘a’ (present) or ‘b’ (absent). Distribution of the microcalcifications was also categorized into clustered, segmental, regional or diffuse (Figs 2, 3). Because the diagnostic significance of each feature is different from mere agreement (one or other radiologist could misinterpret a finding and still the intra- or interobserver agreement could be good), the statistical significance of each feature was calculated in relation to the final pathologic diagnosis. For pleomorphism of the microcalcifications, associated architectural distortion, mass or increased density, only ‘a’ (present) was regarded as having malignant potential. The diagnostic significance of the clustering of the microcalcifications was calculated as clustered vs segmental, regional and diffuse, i.e., only clustering was regarded as having malignant potential and others (segmental, regional and diffuse) were regarded as negative for malignant potential. Only data with acceptable intraobserver agreement (Kappa value above 0.4) were used for this purpose.

RESULTS

Intraobserver agreement for each parameter was mostly above 0.4 (Table 1). Only observer 3 (resident) showed insignificant values for pleomorphism and architectural distortion. Interobserver agreement was fair to good for number, associated density, and distribution and poor for pleomorphism and architectural distortion (Table 2). Because both intra- and interobserver agreement were fair, the data from observer 1

116

CLINICAL RADIOLOGY

Table 1 – Kappa and weighted kappa values of intraobserver agreement*

Number Pleomorphism Architectural distortion Associated mass Associated density Distribution

Observer 1

Observer 2

Observer 3

0.636 0.443 0.702 0.472 0.673 0.830

0.739 0.905 1.000 0.472 0.882 0.928

0.636 NA† 0.183 0.603 0.653 0.646

* Values above 0.75 were regarded as good correlation. Values below 0.4 were regarded as not significant. † NA kappa value invalid due to unequal number of categories.

Table 2 – Kappa and weighted kappa values of interobserver agreement*

Number Pleomorphism Architectural distortion Associated mass Associated density Distribution

Observer 1 vs 2

Observer 2 vs 3

Observer 1 vs 3

0.582 0.328 0.193 0.497 0.549 0.832

0.582 NA† 0.346 0.264 0.652 0.600

0.539 NA† 0.160 0.331 0.542 0.677

* Values above 0.75 were regarded as good correlation. Values below 0.4 were regarded as not significant. Only the first evaluation results were used. † NA kappa value invalid due to unequal number of categories. Fig. 3 – An example of associated density. Craniocaudal mammography shows 10–14 microcalcifications in the unit area with associated density in inner portion of right breast. Pathologic diagnosis was invasive ductal carcinoma.

(observer 3) also showed fair diagnostic accuracy for associated mass and density (Tables 6–7). Only the significance of clustering was 40.4–49.1% (Table 8). Clustering and diffuse pattern were more frequently shown in benign lesions (Table 9). The specificities of architectural distortion, associated mass, and associated increased density were in the range of 71.9–93.5.

(a dedicated breast radiologist) were representatively used to obtain the mean numbers of microcalcifications in the unit area for both benign and malignant lesions. The total numbers of cases in each category of observer 1’s data were multiplied by each representative value for the seven categories (3 for 1–4, 7.5 for 5–9, 12.5 for 10–14, 17.5 for 15–19, 22.5 for 20–24, 27.5 for 25–29, 32.5 for more than 30, respectively). Data were summed and divided by the total number of cases for both benign and malignant lesions (Table 3). The results were 16.4 for malignant and 16.7 for benign lesions. There was no statistically significant difference between the two groups. The diagnostic significances of pleomorphism, architectural distortion, associated mass and associated increased density were 50.9–66.1% (Tables 4–7). The least experienced radiologist

DISCUSSION

Microcalcifications demonstrated on screening mammography is the reason for biopsy in 40% to 50% of non-palpable breast cancers, both for in situ and invasive diseases [2,3]. Calcifications associated with a mass cause much difficulty for diagnosing cancers on mammography. Microcalcifications without associated findings cause the most diagnostic difficulty. As many microcalcifications have non-specific features and need further evaluation [3,6], the number and distribution of the

Table 3 – The total numbers in each category of observer 1’s data and mean value from the representative values

Benign (31)† Malignant (26)

1–4

5–9

10–14

15–19

20–24

25–29

> 30

Mean*

0 1

6 5

9 6

9 6

2 4

1 3

4 1

16.7 16.4

* Mean numbers of two groups are statistically not significant by Wilcoxon’s rank sum test. Representative values are 3, 7.5, 12.5, 17.5, 22.5, 27.5 and 32.5 for each category. † Numbers in parentheses represent the total number of cases.

117

CLUSTERING OF MICROCALCIFICATIONS

Table 4 – Statistical values* for each observer for pleomorphism

Sensitivity Specificity PPV NPV Accuracy

Order

Observer 1

Observer 2

1 2 1 2 1 2 1 2 1 2

61.5 38.5 58.1 77.4 55.2 58.8 64.3 60.0 59.6 59.6

80.8 84.6 32.3 29.0 50.0 50.0 66.7 69.2 54.4 54.4

Table 7 – Statistical values* for each observer for associated density

Observer 3† Sensitivity Specificity PPV NPV Accuracy

* Sensitivity ¼ TP/TP þ FN Specificity ¼ TN/FP þ TN Positive predictive value (PPV) ¼ TP/TP þ FP Negative predictive value (NPV) ¼ TN/TN þ FN Accuracy ¼ TP þ TN/TN þ FN þ TP þ FP † Observer 3’s data was excluded because intraobserver agreement was poor.

Sensitivity Specificity PPV NPV Accuracy

Observer 1

Observer 2†

1 2 1 2 1 2 1 2 1 2

42.3 38.5 80.6 80.6 64.7 62.5 62.5 61.0 63.2 61.4

19.2

Observer 3‡

Sensitivity

PPV NPV

71.4

Accuracy

58.0

Observer 2

Observer 3

1 2 1 2 1 2 1 2 1 2

45.8 38.5 75.9 86.2 61.1 71.4 62.9 61.0 62.3 63.6

48.0 52.0 80.6 74.2 66.7 61.9 65.8 65.7 66.1 64.3

38.5 34.6 83.9 74.2 66.7 52.9 61.9 57.5 63.2 56.1

Table 8 – Statistical values* for each observer for distribution†

Specificity

93.5

Observer 1

* See Table 6.

Table 5 – Statistical values* for each observer for architectural distortion Order

Order

Order

Observer 1

Observer 2

Observer 3

1 2 1 2 1 2 1 2 1 2

46.2 46.2 35.5 38.7 44.0 38.7 36.0 46.2 40.4 42.1

44.0 52.0 40.6 40.6 36.7 40.6 48.1 52.0 42.1 45.6

76.9 76.9 25.8 22.6 46.5 45.5 57.1 53.8 49.1 47.4

59.6

* Sensitivity ¼ TP/TP þ FN Specificity ¼ TN/FP þ TN Positive predictive value (PPV) ¼ TP/TP þ FP Negative predictive value (NPV) ¼ TN/TN þ FN Accuracy ¼ TP þ TN/TN þ FN þ TP þ FP † Observer 2’s data was calculated and presented only once because intraobserver agreement was 1.000. ‡ Observer 3’s data was excluded because intraobserver agreement was poor.

* See Table 6. † Clustering only is regarded as true positive and segmental, regional, diffuse are all regarded as negative to find out the diagnostic ability of clustering alone.

Table 9 – Distribution of microcalcifications assessed by each observer Pathologic diagnosis

Observer

Order Distribution*

Table 6 – Statistical values* for each observer for associated mass

Sensitivity Specificity PPV NPV Accuracy

Order

Observer 1

Observer 2

Observer 3

1 2 1 2 1 2 1 2 1 2

11.5 0 90.3 93.5 50.0 0 54.9 52.7 54.4 50.9

11.5 23.1 93.5 87.1 60.0 60.0 55.8 57.4 56.1 57.9

32.0 28.0 71.9 78.1 47.1 50.0 57.5 58.1 54.4 56.1

Malignant lesions

1 2 3

s

r

d

1 2 1 2 1 2

12 12 11 13 20 20 14.7

9 7 8 7 2 4 6.2

5 7 6 5 4 1 4.7

0 0 0 0 0 1 0.2

1 2 1 2 1 2

20 19 19 19 23 24 20.7

5 6 5 5 2 4 4.5

3 3 5 4 3 1 3.2

3 3 3 4 3 2 3

mean Benign lesions

1 2 3

* Sensitivity ¼ TP/TP þ FN Specificity ¼ TN/FP þ TN Positive predictive value (PPV) ¼ TP/TP þ FP Negative predictive value (NPV) ¼ TN/TN þ FN Accuracy ¼ TP þ TN/TN þ FN þ TP þ FP

c

mean

* Distributions are divided into c (clustered), s (segmental), r (regional), d (diffuse).

118

CLINICAL RADIOLOGY

microcalcifications have been cited as important features in the assessment of their malignant potential. Modern mammographic units are able to show more than five microcalcifications in an area of 1 cm2. Too many lesions are classified as ‘clustered’ using this traditional definition of risk with the result that this parameter is of little diagnostic value in differentiating benign from malignant processes. Egan et al. [10] found no cancers when fewer than five calcifications were present in an area of 0.5 × 0.5 cm and 84% of cancers were associated with more than 10 calcifications. Powell et al. [6] also found no cancers when fewer than five calcifications were present in a cluster. They also concluded that calcification was a strong predictor of malignancy. Hall et al. [11] used five as moderately suspicious of malignancy, but unfortunately they did not state either the unit area or the mammography unit used. De Lafontan [12] counted the number of microcalcifications per cm2. They found that a compact cluster (> 20/cm2) was more frequently observed in malignant lesions and loose clusters (< 10/cm2) were more frequently observed in benign diseases. However, they did not indicate the mean numbers of microcalcifications per cm2. Muir et al. [2] observed that approximately one-third of cases were malignant and two-thirds benign for whatever number of calcifications chosen as the level determining biopsy for non-invasive cancers. They also concluded that associated opacity and palpability must be regarded as suspicious of invasive cancer. Franceschi et al. [13] suggested two predictive criteria for malignancy: linear/ arborescent pattern and number of microcalcifications more than 15/cm2, both of which correlate well with our result. In this study, the diagnostic specificities of architectural distortion, associated mass, and associated increased density were in the range of 72–93%. The diagnostic significances of pleomorphism, architectural distortion, associated mass and associated increased density were generally above 50% (Tables 4–7). Although these are not very high numbers, part of the reason may be that most of the assessed lesions were nonpalpable and small and associated features were absent in most of the malignant lesions. However, the significance of clustering was lower than 50% and this indicates that clustering alone does not necessarily suggest malignancy (Table 8). In summary, we have found that the number of microcalcifications per unit area is much larger than the traditional

descriptions of ‘clustering’ for both benign and malignant lesions. According to our results, a 1 cm2 area may contain more than 60 microcalcifications for most lesions. We suggest that ‘more than 15 microcalcifications in an area of 0.25 cm2 (0.5 × 0.5 cm)’ is a more useful definition of clustering. We have demonstrated that clustering itself is not a very useful diagnostic feature for differentiating malignant from benign breast microcalcifications shown on mammography. Pleomorphism of microcalcifications or associated architectural distortion, mass or increased density should be considered as more important diagnostic predictors of malignancy.

REFERENCES 1 Sickles EA. Breast calcifications:mammographic evaluation. Radiology 1986;160:289–293. 2 Muir BB, Lamb J, Anderson TJ, Kirkpatrick AE. Microcalcification and its relationship to cancer of the breast: experience in a screening clinic. Clin Radiol 1983;34:193–200. 3 Monsees BS. Evaluation of breast microcalcifications. In: Jackson VP, ed. Breast Imaging, RCNA. Philadelphia: W.B. Saunders Co, 1995;33: 1109–1121. 4 Kopans. Breast Imaging, 2nd edn. Philadelphia: Lippincott-Raven, 1998: 321–322. 5 Kopans. Breast Imaging, 2nd edn. Philadelphia: Lippincott-Raven, 1998: 388–389. 6 Powell RW, McSweeney MB, Wilson CE. X-ray calcifications as the only basis for breast biopsy. Ann Surg 1983;197:555–559. 7 Bird RE. Critical pathways in analyzing breast calcifications. Radiographics 1995;15:928–934. 8 Fleiss JL. Statistical Methods for Rates and Proportions. 2nd edn. New York: John Wiley & Sons, 1981:217–225. 9 Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159–174. 10 Egan RL, McSweeney MB, Sewell CW. Intramammary calcifications without an associated mass in benign and malignant disease. Radiology 1980;137:1–7. 11 Hall FM, Storella JM, Silverstone DZ, Wyshak G. Nonpalpable breast lesions: recommendations for biopsy based on suspicion of carcinoma at mammography. Radiology 1988;167:353–358. 12 De Lafontan B, Daures JP, Salicru B, et al. Isolated clustered microcalcifications: diagnostic value of mammography – series of 400 cases with surgical verification. Radiology 1994;190:479–483. 13 Franceschi D, Crowe J, Zollinger R, et al. Biopsy of the breast mammographically detected lesions. Surg Gynecol Obstet 1990;171: 449–455.