ARTICLE IN PRESS

The Breast (2006) 15, 29–38

THE BREAST

KEYWORDBreast;MicrocalcifiDCE-MRI;Screening

0960-9776/$ - sdoi:10.1016/j.b

�CorrespondiE-mail addr

www.elsevier.com/locate/breast

ORIGINAL ARTICLE

Differentiation of benign from malignant breastdisease associated with screening detectedmicrocalcifications using dynamic contrastenhanced magnetic resonance imaging

P.J. Kneeshawa,�, M. Lowryb, D. Mantonb, A. Hubbardc, P.J. Drewa,L.W. Turnbullb

aAcademic Surgical Unit, Castle Hill Hospital, Cottingham, Hull, UKbThe Centre for Magnetic Resonance Investigations, Hull Royal Infirmary, Anlaby Road, Hull, UKcThe Hull and East Yorkshire Breast Care Unit, Castle Hill Hospital, Cottingham, Hull, UK

Received 28 September 2004; received in revised form 5 April 2005; accepted 13 May 2005

S

cations;

ee front matter & 2005reast.2005.05.002

ng author. 4 Warwick Dess: peter@kneeshaw1.

Summary Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) isan effective diagnostic modality for symptomatic breast disease. However, its role inevaluating clinically occult disease associated with mammographically detectedmicrocalcification remains unclear.

Women recalled following screening mammography with microcalcification hadDCE-MRI examination of the breast. The data were evaluated subjectively andobjectively using both empirical and 2-compartment pharmacokinetic modellingtechniques to evaluate signal intensity parameters.

Eighty-eight patients aged 50–75 years (median 58) were recruited. Comparingmalignant and benign lesions, the mean values in arbitrary units for theenhancement index at 1min in the most enhancing 9-pixel square 71 standarddeviation were 0.6170.40 vs. 0.2270.26 p ¼o0:001 with sensitivity, specificity,PPV, NPV and accuracy of 80.0%, 82.4%, 57.1%, 93.3% and 81.8%, respectively. Thecorresponding values attained by the radiologist were 75.0%, 89.7%, 68.2%, 92.4%and 86.4%.

DCE-MRI is able to differentiate malignant from benign clinically occult lesionsassociated with microcalcification and may therefore offer an alternative to opensurgical biopsy for women with equivocal findings following initial triple assessmentfor microcalcification in the breast.& 2005 Elsevier Ltd. All rights reserved.

Elsevier Ltd. All rights reserved.

rive, Beverley HU17 9TB, UK. Tel.: +44 7976827909.karoo.co.uk (P.J. Kneeshaw).

ARTICLE IN PRESS

P.J. Kneeshaw et al.30

Introduction

It is now generally accepted that breast screeningprogrammes reduce cancer deaths by more than30% of the patients included in the program.1 As aresult of increased uptake more women are nowdealing with diagnostic and treatment decisionsrelating to a finding of microcalcifications.2 In theHull and East Riding UK Breast Screening Program(HEYBS) we perform approximately 27,000 screen-ing mammograms a year and of these approxi-mately 1000 ladies are recalled for furtherassessment of suspicious areas. Twenty per centof these recalls are to assess areas of microcalci-fication in the breast.

Microcalcifications are caused by either benign ormalignant processes and are classified by theirmorphological appearances including size, shapeand distribution. Most represent benign conditions,but between 20% and 30% of women biopsied havemalignant disease even when no palpable mass isdetected.3 Many patterns of microcalcification fallinto the indeterminate group, when a confidentdiagnosis of benign or malignant disease cannot bemade. Patients with clinically occult disease andmicrocalcifications that are classified as mammogra-phically malignant, suspicious or indeterminate, willundergo stereo-tactic biopsy. Most women toleratethis procedure, however, approximately 5% fail due toeither awkward lesion position relative to the patientsize; inability of the patient to tolerate the procedureor equipment failure. These ladies may progress toopen surgical radio-localised biopsy.

Clustered microcalcifications on X-ray mammo-graphy are a sensitive sign of ductal carcinoma insitu (DCIS), but the specificity ranges from 10% toonly 35%.4 Ultrasound with a high-frequency trans-ducer can reliably depict clusters of microcalcifica-tions larger than 10mm3,5 however, these tend tobe large, superficially placed microcalcificationssituated within a mass.

Using gadolinium-based intravenous contrastagents, fast imaging techniques and dedicatedreceiver coils, dynamic contrast-enhanced mag-netic resonance imaging (DCE-MRI) is becoming aninvaluable tool in the diagnosis and treatment ofbreast disease.6 Sensitivities are now in excess of95% and specificities greater than 90% for invasivebreast cancer.7–9

Early studies suggested that DCE-MRI should notbe used to assess microcalcifications.10,11 DCE-MRIis unable to image small (o0.5mm) calcificationstypically associated with malignant disease. How-ever, new vessel formation, or angiogenesis, can bedetected in vivo by DCE-MRI, which examinesfactors such as blood flow and vessel permeability.

In a similar manner to invasive disease, DCIS caninduce angiogenesis. DCE-MRI has been reported toshow increased contrast uptake in either a focal orbranching pattern in DCIS. In a study by Gilles etal., dynamic MR imaging showed contrast enhance-ment in 34 out of 36 patients with DCIS.12 So themicrocalcification does not have to be visualised todemonstrate the cancer.

The aim of the study was to evaluate the efficacyof DCE-MRI in the investigation of microcalcificationin the breast.

Methods

Approval from the local ethics committee and Trustquality assurance was granted. All patients recalledto the screening assessment clinic at the HEYBS unitbecause of indeterminate, suspicious or malignantfindings at screening were offered participation inthe trial. The purpose of the study was explainedand those wishing to take part were fully consentedwith a witness present. The findings of the screen-ing X-ray mammograms (XRMs) were reviewed by adedicated radiologist specialising in breast ima-ging. These were graded as normal, benign,indeterminate, suspicious of malignancy or prob-ably malignant and recorded as R1 to R5, respec-tively. Patients with microcalcification had furtherXRM performed (Alpha RT Instrumentation or Sie-mens Nova 2000). At the first breast screening visitpatients have cranio-caudal (CC) and medio-lateraloblique (MLO) views, but at subsequent visits onlythe MLO view is obtained. At recall a CC view isperformed if not done already, and additionallylateral and magnified lateral views are performedas required. The patient was also assessed clinicallyand by ultrasound scanning. The ultrasound exam-inations were performed using either a HitachiEUB310S system with a 7.5MHz transducer (HitachiMedical Systems, Tarrytown, NY, USA) or an EsaoteIdea AU4 system with either a 7.5/10MHz or a 10/13MHz transducer (Esaote S.p.A., Genova, Italy).The results were recorded in the same manner asthe mammograms. Patients with indeterminate,suspicious or malignant findings on XRM or USS hada stereotactic 14G core biopsy (Opdimas) takenfrom the area of microcalcifcations.

The breast MRI examination was scheduled atleast 10 days after the stereocore biopsy, butbefore any definitive surgery.

MR protocols

DCE-MRI was carried out using a 1.5 T SignaEchospeed MRI scanner (International General

ARTICLE IN PRESS

Differentiation of benign from malignant breast disease 31

Electric, Milwaukee, WI, USA) and a bilateralphased array breast coil (Machnet) for signalreception. Patients were postitioned prone on theexamination table and entered the bore of themagnet feet first. A 20 gauge Venflons was placedin a vein on the dorsum of the hand or in the ante-cubital fossa. A 1250mm long Avons wide boreextension set, attached to a 50ml syringe via a 3-way tap, was primed with 40ml of normal salineand attached to the Venflons. This was used toachieve rapid delivery of the MR contrast agent.

Patients underwent an axial localiser sequence,followed by a series of 28 T1 weighted 3D images(4.0 to 5.0mm slice thickness, no gap, flip 301,256� 192 matrix, field of view (FOV) 28 cm) in thecoronal plane of the abnormal breast. The protondensity sequence was set up at 9 slice locations(thickness/spacing 5.0–9.0mm/2.0mm (dependenton lesion size), 256� 128 matrix, FOV 30 cm) overthe entire breast in the coronal plane to encompassany areas of pathology. The dynamic data wasacquired at these locations using a T1 weighted fastRF-spoiled gradient-echo (FSPGR) sequence (thick-ness/spacing 5.0–9.0mm/2.0mm (dependent onlesion size), 256� 128 matrix, FOV 30 cm, flip301). A bolus injection of gadolinium diethylene-triamine penta-acetic acid (Gd-DTPA MagnevistSchering Health Care, Burgess Hill, UK), at a doseof 0.1mmol/kg body weight, was given by hand-injection over 10 s after the second set of dynamicimages had been acquired. This was followed by a20ml flush of normal saline. In total, data from 35time points, at each slice location, were acquiredat a temporal resolution of 13 s. Finally, 28 T1weighted 3D fat-suppressed, post-contrast imageswere then acquired in the sagittal plane using anFSPGR sequence (4.0–5.0mm slice thickness, nogap, flip 301, 256� 192 matrix, FOV 28 cm). Thetotal scan time was approximately 20min.

Radiologist reporting

One expert MR radiologist reported the DCE-MRI.The report was based upon the lesion morphologyand the signal intensity time data acquired fromthe dynamic sequence. The dynamic data wereanalysed using empirical data analysis software(FuncTool, General Electric, Milwaukee, WI, USA)on a Sun-Unix Advantage Windows workstationinterrogating the percentage change in signalintensity over time. The shapes of the signalintensity curves were classified as previouslydescribed by Kuhl.13 The radiologist was aware ofthe breast quadrant where the microcalcifications

were situated. Information regarding the radiolo-gical level of suspicion (R1–R5) or histology resultfrom the stereocore biopsy was not available to theMR radiologist. The result was recorded as benignor malignant for statistical purposes.

Post-processing DCE-MRI analysis

Regions of interest (ROI) which excluded anyperipheral non-enhancing tissue were drawnaround the areas of contrast uptake using imageanalysis software (Analyze Biomedical ImagingResource V1.0) on a Sun Unix workstation. Thedynamic data were also analysed empirically andpharmacokinetically using software developed in-house. Empirical parameters were determinedfrom the signal intensity time curves generatedfrom the dynamic data sets for each lesion (Fig. 1).

Pharmacokinetic modelling is a mathematicalprocess that uses all the signal intensity data togive numeric values for the lesion permeability andthe contrast exchange rate between the plasmaand the extra-vascular and extra-cellular space of alesion. We used established methodology derivedfrom the 2-compartment pharmacokinetic model ofBrix, adapted in house, to give the amplitude ofcontrast uptake and exchange rate.

The empirical parameters used were: enhance-ment index (EI) which is the percentage rise insignal with reference to the initial signal; EI at1min (EI 1min) which is the percentage rise insignal at 1min with reference to the initial signaland the normalised maximum intensity time ratio(nMITR) where,

nMITR ¼Smax � So

So�

1T

and Smax ¼ maximum signal intensity (arbitraryunits) achieved, So ¼ baseline signal intensity(arbitrary units), and T ¼ time for the change insignal intensity (seconds) to occur.

These parameters were used to assess the wholeROI and the most enhancing 9-pixel square wasdetermined by an in-house developed computeralgorithm that interrogates all the pixels in the ROI,and selects the most enhancing 3� 3 pixel region.

Statistical analysis of the empirical and pharma-cokinetic data was performed using independentsamples t-test. To translate the data into clinicalapplications a cut-off value was determined foreach parameter using receiver operator character-istic curve analysis. Selection of the most appro-priate value to use as a diagnostic cut-off was madeby determining which, of all possible cut-off values,

ARTICLE IN PRESS

Figure 1 T1 weighted fat-suppressed post-contrast images of (a) high-grade DCIS tumour and (b) fibroadenoma with thecorresponding signal intensity curves used to differentiate malignant from benign disease. Note the more rapid initialrise in the SI curve in the malignant case compared to the benign lesion.

P.J. Kneeshaw et al.32

produced the point closest to the co-ordinate (0,1)a point that represents maximal sensitivity andspecificity.

The accuracy of the DCE-MRI, X-ray mammogra-phy and ultrasound examinations reported andclassified by radiologists, were expressed in termsof sensitivity, specificity, PPV and NPV.

The final diagnosis was obtained from thehistology report following definitive surgery. If nosurgery was performed final outcome was notedfrom the core biopsy result.

Results

Eighty-eight patients aged 50–75 years (mean 57.4,median 58) were recruited to the study. Clinicalevaluation, DCE-MRI and X-ray mammography wereperformed in all 88 patients, but ultrasound wascarried out in only 84/88. Histological diagnosis wasavailable in 74 patients. The remaining patients(n ¼ 14) were discharged to routine screeningfollowing the additional mammographic assess-ment, as the microcalcifications were classified as

ARTICLE IN PRESS

Differentiation of benign from malignant breast disease 33

R2 or below. A total of 20/88 patients hadmalignant disease and 68/88 had benign disease(Table 1).

The objectively derived parameters namely EI at1min and nMITR, for both the whole region ofinterest and the most enhancing 9-pixel square,were used to compare benign and malignantdisease. The mean value71 standard deviation isshown for each parameter with the statisticalsignificance. Although significant differences innMITR were obtained between benign and malig-nant lesions for both the whole and maximumenhancing 9-pixel square, considerably greaterdifferences were obtained for EI at 1min asindicated in Table 2.

Cut-off values (arbitrary units) of 0.237 for EI at1min, 8.4 for nMITR, 0.305 for EI at 1min in a 9-pixel square and 10.4 for nMITR in a 9-pixel squarewere derived from the ROC curve analysis and thediagnostic accuracy for each parameter calculatedis shown (Table 3). nMITR achieved the best overallaccuracy at 85.2% with values for the otherparameters ranging from 79.5% to 81.8%.

A comparison of the radiologist reported imagingmodalities namely X-ray mammography, DCE-MRIand ultrasound is shown (Table 4). Ultrasound wasperformed in 84 out of the 88 patients and achieveda sensitivity and specificity of 55.0% and 81.2%.DCE-MRI was inferior to X-ray mammography in

Table 1 Final histology from 88 women aged 50–75 yea

0

2

4

6

8

10

12

14

16

18

NOS G3

and

high

grad

e DCIS

NOS G2

and

high

grad

e DCIS

NOS G2

and

low g

rade

DCIS

NOS G1

and

high

grad

e DCIS

IDC G

3 an

d hig

h gr

ade

DCIS

IDC G

2 an

d hig

h gr

ade

DCIS

IDC G

2 an

d low

gra

de D

CIS

High g

rade

DCIS

Inte

rmed

iate

DCIS

Lobu

lar ca

rcino

ma

in

Atypica

l duc

tal

Typica

terms of sensitivity (75.0% vs. 95.0%) but resultedin superior specificity (89.7% vs. 25.0%) anddiagnostic accuracy (86.4% vs. 40.9%). The overallaccuracy of DCE-MRI reported by the radiologistwas comparable with objective analysis usingnMITR (86.4 vs. 85.2%) (Tables 3 and 4).

The pathology and size of the incorrect MRradiologist reported lesions are shown in Table 5.Two invasive tumours each less than 5mm weremissed. These were both associated with high gradeDCIS. In addition, 3 other cases of DCIS were alsodiagnosed as benign disease by DCE-MRI.

The screening assessment XRMs were classified asnormal or benign in 14/88 cases; indeterminate in52/88 and suspicious or malignant in 12 cases. The5 cases incorrectly reported as negative on DCE-MRIby the radiologist all scored R3 or above on XRM butUSS reported 4 of the 5 false negatives on DCE-MRIas normal (Table 6).

Discussion

The rising attendance at breast screening units andthe increasing sensitivity of mammography willincrease the number of patients with microcalcifi-cation requiring further assessment. After stereo-tactic core biopsy failure the current options are to

rs (including R1-R2 XRM group).

situ

hype

rplas

ia

l duc

tal h

yper

plasia

Fibroc

ystic

dise

ase

Fibroa

deno

ma

Benign

Norm

al

R1-R2

XRM

Number of patients

ARTICLE IN PRESS

Table 2 Parameters derived from objective data only showing the difference between malignant and benignlesions.

Histology Number of patients Mean Std. Deviation p-value

nMITR Benign 68 6.132662 7.372763 0.008

Malignant 20 14.328495 12.079885

EI at 1 minute Benign 68 .166062 .192868 <0.000

Malignant 20 .445625 .262762

nMITR 9-pixels Benign 68 8.979888 12.170805 0.019

Malignant 20 34.986595 44.876339

EI at 1 minute 9-pixels Benign 68 .219954 .263214 <0.000

Malignant 20 .607905 .395991

Table 3 Diagnostic accuracy with 95% confidence intervals using objectively derived empirical DCE-MRIparameters; for whole and maximally enhancing 9-pixel square ROI (all values given are percentages).

Objective analysis Sensitivity (95% CI)

Specificity (95% CI)

PPV (95% CI)

NPV (95% CI)

Accuracy (95% CI)

Enhancement index at 1

minute in whole ROI

80.0

(62.5-97.5)

80.9

(71.6-90.2)

55.2

(37.1-73.2)

93.2

(86.8-99.6)

80.7

(72.5-88.9)

NMITR in whole ROI 75.0

(56.0-94.0)

88.2

(80.5-95.9)

65.2

(45.7-84.6)

92.3

(85.8-98.8)

85.2

(77.7- 92.6)

Enhancement index at 1

minute in 9-pixel square

80.0

(62.5-97.5)

82.4

(73.3-91.5)

57.1

(38.7-75.4)

93.3

(87.0-99.6)

81.8

(73.7-89.9)

Normalised MITR in 9-

pixel square

80.0

(62.5-97.5)

79.4

(69.8-89.0)

53.3

(35.4-71.2)

93.1

(86.6-99.6)

79.5

(71.1-87.9)

P.J. Kneeshaw et al.34

ARTICLE IN PRESS

Table 4 Diagnostic accuracy of the radiologist reported imaging methods with 95% confidence intervals (allvalues are percentages).

Radiologist reportedSensitivity (95% CI)

Specificity (95% CI)

PPV (95% CI)

NPV (95% CI)

Accuracy (95% CI)

X-ray Mammography95.0

(85.4-100.0)

25.0

(14.7-35.2)

27.1

(16.7-37.5)

94.4

(83.8-100.0)

40.9

(30.6-51.1)

Ultrasound

(84 patients)

55.0

(33.2-76.8)

81.2

(71.6-90.8)

47.8

(27.4-68.2)

85.2

(76.3-94.1)

75.0

(65.7-84.3)

DCE-MRI 75.0

(56.0-94.0)

89.7

(82.5-96.9)

68.2

(48.7-87.7)

92.4

(86.0-98.8)

86.4

(79.2-93.6)

Differentiation of benign from malignant breast disease 35

repeat the core, return the patient to early recallor proceed to localised surgical biopsy. The benignsurgical biopsy rate of screened patients for ourunit was 0.2% in the year ending 2000. Thereforethere is a need for a technique that has anacceptably high accuracy so that lesions deemedto be benign can be safely left in-situ.

This work has assessed the accuracy of DCE-MRIin evaluating disease associated with microcalcifi-cation in the breast using 2 reporting methods. Theconventional MR reporting in this study wasperformed by dedicated MR radiologists achievingan overall accuracy of 86.4%, but with much higherspecificity and positive predictive values comparedto XRM or USS. However, we have demonstratedthat signal intensity characteristics analysed objec-tively, using empirical parameters, can give anaccuracy rate which is comparable to the radiolo-gist but also superior to either XRM or USS. Thehighest negative predictive value achieved usingobjective analysis was 93.3% compared to 94.4% forX-ray mammography. To attain this result, datafrom a region of interest drawn manually over theenhancing area of the breast is required and nomorphological information is used. The cut-offvalue for separating benign from malignant diseasecan be altered to maximise the negative predictivevalue.

Histopathological assessment of the 5 falsenegative cases by DCE-MRI showed 3 cases of pureDCIS, 1 of intermediate grade with a maximumdiameter of 50mm and 2 tumours both with small

invasive foci (5 and 0.4mm) in addition to DCIS.Four of these cases were reported as R3 and 1 as R4on XRM. The lack of contrast enhancement may bedue to relatively poor or absent angiogenesis.Indeed reports indicate that DCIS will progress toinvasive disease in between 25% and 50% of casesonly.14 It is possible that areas of DCIS that do notenhance may fall in to this group, but long-termfollow up of a larger cohort of these cases would berequired to answer this question, if this wasdeemed ethically justifiable.

Reasons for false positive results in other studiesinclude normal enhancing breast tissue in the pre-menopausal patient as well as vascular fibroadeno-mas, papillomas and radial scars.15,16 The falsepositive rate in this study is in keeping with otherauthors findings. It has been previously noted thatfibrocystic disease is also a source of false positiveresults.10

MRI identified one 12mm tumour of no specificinvasive type with associated high grade DCIS thatwas missed by the other imaging modalities.Following the MRI examination, the original XRMswere reviewed and an area of microcalcificationsthat coincided with the enhancing area on MRI, notthought to be malignant originally was biopsied.This was found to be malignant. This patient wouldhave been discharged to normal screening as theassessment XRM and USS were classified as R1. Inthis case an area of microcalcifcations could befound at the area of enhancement on MRI and sostereotactic biopsy was possible. Although not a

ARTICLE IN PRESS

Table 5 Comparison of results obtained from radiologist reported DCE-MRI and histology.

Malignant Histology (n=20) Benign Histology (n=68)

Malignant on

DCE-MRI

6 pure DCIS (5 high grade 1

intermediate grade)

9 invasive and DCIS

Total n=15

Fibroadenoma

Lobular carcinoma in situ

Columnar cell change

Typical ductal hyperplasia

Involutional change

Papilloma

Normal

Total n=7

Benign on

DCE-MRI

10mm high grade DCIS

5mm NOS and high grade DCIS

50mm intermediate grade DCIS

0.4mm invasive and 55mm DCIS

11mm high grade DCIS

Total n=5

2 atypical ductal hyperplasia

5 typical ductal hyperplasia

14 fibrocystic disease

4 fibroadenomas

3 LCIS

15 other benign histology

5 normal

14 benign no biopsy (R1-R2)

Total n=61

P.J. Kneeshaw et al.36

problem in this study, enhancing areas occult on X-ray mammography, clinically and by ultrasoundnecessitate the need for MR guided biopsy. Openconfiguration scanners allow continuous access to

the patient but are limited by low field strengthwhich compromises signal-to-noise.17 Closed mag-net systems allow access to the patient outside thebore of the magnet with a localisation device. Early

ARTICLE IN PRESS

Table 6 Comparison of ‘‘level of suspicion’’ scores for XRM and USS for the false positive (7 cases) and falsenegative (5 cases) DCE-MRI examinations.

Patient XRM USS 1 3 1 2 3 1 3 3 1 4 3 4

False negative DCE-MRI

5 4 1 6 4 1 7 3 1 8 3 1 9 2 1 10 4 1 11 2 1

False positive DCE-MRI

12 4 3

Differentiation of benign from malignant breast disease 37

small studies using freehand and localisationdevices have been encouraging. Daniel et al.managed to obtain diagnostic tissue in all 27 lesionsusing a freehand method. Liney et al. advocatedthe use of carbon fibre core biopsy needle toreduce artefact.18 Lehman et al. advovated the useof an 11-gauge MR-compatible device designed foruse in magnetic resonance vacuum-assisted breastbiopsy. They successfully sampled tissue from 15out of 16 lesions.19

Previous MR studies have reported variableaccuracy for classification of microcalcification.Westerhof et al. investigating mammographicallysuspicious microcalcifications reported a sensitivityof 45%, specificity of 72%, positive predictive valueof 71%, negative predictive value 46% and accuracyof 56%.10 In a further study by Gilles et al. theyobserved a sensitivity of 95% and a specificity of51%. Their specificity was potentially impairedbecause the presence or absence of contrastuptake in the breast was the only parameter usedto decide if the area of microcalcification wasassociated with malignancy.11 Hwang et al. as-sessed the efficacy of MRI in patients with knownDCIS for assessment of residual disease, occultinvasion, and multicentricity. They achieved anaccuracy of 88% in predicting residual disease, 82%in predicting invasive disease and 90% in predictingmulticentricity. They concluded that MRI of DCIScan serve as a useful adjunct to mammography byproviding a more accurate assessment of the extentof residual or multicentric disease.20

In this series we have shown that DCE-MRI is ableto differentiate benign from malignant diseaseassociated with microcalcification with consider-

ably greater accuracy than either X-ray mammo-graphy or ultrasound. Moreover DCE-MRI may havea role for patients in whom the findings of initialtraditional triple assessment are equivocal.

Acknowledgements

Yorkshire Cancer Research for their support.The Division of Cancer within the Postgraduate

Medical Institute of the University of Hull (PGMI) inassociation with the Hull York Medical School UK.

References

1. Salvador R. Screening in breast cancer—an update. Eur JCancer 2002;38(3):4–5.

2. The Women’s Cancer Resource Centre. Breast calcifications:beware if a six-month mammogram follow-up is recom-mended, Microcalcification Awareness/Action Committee,2001.

3. Kopans DB. Breast imaging. Philadelphia: Lippincott-Raven;1998. p. 388–9.

4. Basset LW. Mammographic analysis of microcalcifications.Radiol Clin N Am 1992;30:93–105.

5. Moon WK, Im JG, Koh YH, Noh DY, Park A. US ofmammographically detected clustered microcalcifications.Radiology 2000;217(3):849–54.

6. Turnbull LW. Magnetic resonance imaging: its role in thebreast. In: Dixon JM, editor. Breast cancer: diagnosis andmanagement. Amsterdam: Elsevier Science; 2000. p. 1–14.

7. Kaiser WA, Zeitler E. MR imaging of the breast: fast-imagingsequences with and without Gd-DTPA. Preliminary observa-tions. Radiology 1989;170:681–6.

8. Drew PJ, Kerin MJ, Turnbull LW, Imrie M, Carleton PJ, FoxJN, Monson JRT. Routine screening for local recurrencefollowing breast-conserving therapy for cancer with dynamic

ARTICLE IN PRESS

P.J. Kneeshaw et al.38

contrast-enhanced magnetic resonance imaging of thebreast. Ann Surg Oncol 1998;5(3):265–70.

9. Heywang SH, Wolf A, Pruss E, Hilbertz T, Eiermann W,Permanetter W. MR Imaging of the breast with Gd-DTPA: useand limitations. Radiology 1989;171:95–103.

10. Westerhof JP, Fischer U, Moritz JD, Oestmann JW. MRimaging of mammographically detected clustered micro-calcifications: is there any value? Radiology 1998;207:675–81.

11. Gilles R, Meunier N, Lucudarme O. Clustered breastmicrocalcifications: evaluation by dynamic contrast-en-hanced subtraction MRI. J Comput Assist Tomogr 1996;20(9):14.

12. Gilles R, Zafrani B, Guinebretiere J, Meunier M, LucidarmeO, Tardivon A, Rochard F, Vanel D, Neuenschwander S,Arriagada R. Ductal carcinoma in-situ MR imaging andhistopathologic correlation. Radiology 1995;196(2):415–9.

13. Kuhl CK, Mielcarek P, Klaschik S. Dynamic breast MR imaging:are the signal intensity time course data useful fordifferential diagnosis of enhancing lesions? Radiology1999(211):101–10.

14. Benson JR. The biology of breast cancer and its relevance toscreening and treatment. In: Morgan MWE, Warren R, Querci

della Rovere D, editors. Early breast cancer. HarwoodAcademic Publishers; 1998. p. 39–40.

15. Weinreb JC, Newstead G. MR imaging of the breast.Radiology 1995;196:593–610.

16. Kuhl CK, Bieling HB, Gieseke J, Kreft BP, Sommer T, LutterbyG, Schild HH. Healthy premenopausal breast parenchyma indynamic contrast-enhanced MR imaging of the breast:normal contrast medium enhancement and cyclical-phasedependency. Radiology 1997;203:137–44.

17. Gould SWT, Lamb G, Lomax D, Gedroyc W, Darzi A.Interventional MR-guided excisional biopsy of breast lesions.J Magn Reson Imag 1998;8:26–30.

18. Liney G, Tozer DJ, Brunsveld van Hulten H, Beerens EGJ,Gibbs P, Turnbull LW. Bilateral open breast coil andcompatible intervention device. J Magn Reson Imag2000;12:984–90.

19. Lehman CD, Aikawa T. MR-guided vacuum-assisted breastbiopsy: accuracy of targeting and success in sampling in aphantom model. Radiology 2004;232(3):911–4.

20. Hwang ES, Kinkel K, Esserman LJ, Lu Y, Weidner N, HyltonNM. Magnetic resonance imaging in patients diagnosed withductal carcinoma-in-situ: value in the diagnosis of residualdisease, occult invasion, and multicentricity. Ann Surg Oncol2003;10(4):381–8.