Introduction
Breast cancer is the most common cancer and the sec-ond most frequent cause of death in women in devel-oped countries[1].In the United States1in9women(in Germany,1in14)will develop breast cancer.Early de-tection is of great importance for the prognosis of the disease.The main diagnostic tools in breast cancer de-tection are physical examination,mammography,ultra-sound and cytology.Although mammography detects carcinomas at an earlier stage than physical examina-tion,5±15%of palpable breast lesions are mammo-graphicallyoccult[2,3,4].
Breast MRI is regarded as a potential tool in breast cancer diagnosis and has gained more clinical impor-tance in the recent years.The main advantage of breast MRI is its high sensitivityof88±100%[5,6,7],com-pared with mammography,which has sensitivity of be-tween69%and75%[8,9].That means that nearlyall breast carcinomas can be detected on MR-mammogra-phy.However,due to the relatively low specificity,the clinical use of breast MRI is still limited[5,6,7,10].
U.Wedeg?rtner U.Bick
K.W?rtler E.Rummeny G.Bongartz Differentiation between benign and malignant findings on MR-mammography: usefulness of morphological criteria
Received:3July2000
Revised:29November2000 Accepted:20F ebruary2001 Published online:21April2001
Springer-Verlag2001
U.Wedeg?rtner())
Abteilung für R?ntgendiagnostik, Radiologische Klinik,
Universit?ts-Krankenhaus Eppendorf, Martinistrasse52,20246Hamburg, Germany
E-mail:u.wedeg@gmx.de
Phone:+49-40-428034029
Fax:+49-40-428033802
U.Wedeg?rtner′G.Bongartz Department of Radiology, Universityof Basel,Basel,Switzerland
U.Bick
Department of Radiology, Universityof Berlin(Charit?),Germany
K.W?rtler′E.Rummeny Department of Radiology,
Technical Universityof Munich(MRI), Germany Abstract The purpose of this study
was to evaluate the usefulness of
morphological criteria in differenti-
ating between benign and malignant
lesions on MR-mammography.Fif-
ty-three women(18±82years)with
62lesions scheduled for excisional
biopsyunderwent dy namic contrast-
enhanced MR-mammographyusing
fast3D Gradient-Echo sequences in
coronal orientation(axial orienta-
tion in seven patients).Histology
revealed44malignant and18benign
lesions.For each lesion,five radiol-
ogists evaluated four morphological
features:lesion shape,irregularityof
contour,homogeneityof contrast
enhancement and presence of ring
enhancement.For each feature a
receiver operating characteristic
(ROC)curve was generated with
calculation of the area under the
curve(AUC).Interobserver vari-
abilitywas evaluated bythe k-coef-
ficient.The most reliable morpho-
logical feature indicating malignan-
cywas an irregular lesion contour
with a sensitivity/specificity/AUC of
83%/76%/0.9followed byinhomo-
geneous contrast enhancement
(85%/42%/0.7)and the presence of
ring enhancement(71%/53%/0.64).
The average interobserver agree-
ment for the different features
ranged between0.35and0.4.Mor-
phological criteria are useful fea-
tures in MR-mammographyfor dif-
ferentiating between benign and
malignant lesions.However,due to
the relativelyhigh interobserver
variability,standardization of termi-
nologyis important.
Keywords Magnetic resonance
imaging′Breast cancer′
Morphology′Receiver operating
characteristic curves
The main diagnostic criterion in the differentiation of benign and malignant lesions on breast MRI is the evaluation of time signal intensitychanges.Carcinomas more often present a faster and stronger postcontrast signal intensityincrease than benign lesions.The draw-back is that there is an overlap between benign and malignant breast lesions [5,6,11,12],because there are rapidlyenhancing benign lesions and slowlyenhancing breast cancers [6,13,14].Hence a final diagnosis is not always possible and further examinations remain nec-essary.Recent studies have tried to develop improved dynamic sequences and interpretations of dynamic con-trast enhancement patterns,but theydid not improve specificity.
The purpose of this studywas to evaluate the useful-ness of morphological criteria in the differentiation of benign and malignant lesions on breast MRI.
Materials and methods
Between F ebruary1993and July1996,53women,scheduled for excisional biopsybecause of palpable or mammographicallyma-lignant suspicious lesions,underwent dynamic contrast-enhanced MR-mammography.The ages of the patients ranged from 18to 82years,with a mean age of 49years.MR-mammography detect-ed 62lesions in 53women.There were two patients with two le-sions,another two with three and one with four lesions.Histology revealed 44malignant and 18benign lesions (Table 1).
The carcinomas had an average tumour size of 1.6cm with a range of 0.3to 5.0cm.The average tumour size of the benign le-sions was 1.2cm with a minimum of 0.2cm and a maximum of 2.5cm.The lesion size of five mastopathies could not be deter-mined due to their diffuse nature.
MRI of the first seven patients was performed on a 1.5T Mag-netom 63SP (Siemens),from F ebruary1993to August 1994.F or-ty-six patients were imaged on a 1.0T Magnetom Impact (Sie-mens)from September 1994to July1996.
All patients were examined in prone position using a double breast coil.
On the Magnetom SP ,dynamic contrast-enhanced MR-mam-mographywas obtained using two-dimensional F LASH (fast-low-angle-shot)gradient echo sequences in an axial orientation with a flip angle of 25 .One sequence pre-and five sequences post-con-trast medium injection at 58-s intervals were carried out.
On the Magnetom Impact,dynamic contrast-enhanced MR-mammographywas obtained using three-dimensional LASH gradient echo sequences in a coronal orientation with a flip angle of 25 .One sequence pre-and three sequences post-contrast me-dium injection at 88s intervals were performed.
GdDTPA was administered as bolus injection using an injector at a dose of 0.2mmol per kg bodyweight after the first sequence,followed bya saline flush of 20ml.The chosen pulse sequences were a compromise between spatial and temporal resolution.Ad-ditional subtraction images were obtained in all cases.Technical details are listed in Table 2.
F or this studyeach lesion was documented on film in four to eight slice positions depending on lesion size;the precontrast as well as the first and third (SP fourth)postcontrast series together with the corresponding subtraction images were shown.The win-dowing was standardized as follows:the window width of the na-tive and postcontrast images corresponded to twice the signal in-tensityof subcutaneous fattytissue.The window width of the sub-traction images was set to two-thirds of the window width of the native and postcontrast images.With regard to the standardized window width settings there was no standardized window level.Window level was adjusted for everyexamination,in a waysuch that optimal image qualitywas achieved.
F ive radiologists separatelyevaluated four morphological fea-tures for each of the 62lesions without anyclinical information.All radiologists were board-certified,and had thorough experience in MRI,reviewing breast MRIs on a regular basis as part of their dailyroutine.However,onlytwo of the readers had extensive ex-perience (>3years)in breast MRI.Each lesion was marked on the
1646
Nature of lesion Type
Number Malignant
Non-invasive
Ductal carcinoma in situ 2Invasive
Invasive ductal carcinoma 32Invasive lobular carcinoma 6Adeno carcinoma 2Papillarycarcinoma 1Medullarycarcinoma 1Total
44Benign
Fibroadenoma 10Papilloma
2Fibrocystic mastopathy
5Other benign lesions with mammographic follow up (2years)1Total
18
Table 1Histologyof the 62le-sions
Table 2Imaging parameters Parameters
Magnetom SP Magnetom Impact Repetition time (TR)220ms 14ms Echo time (TE)6ms
7ms
Field of view 300 300160 320Matrix
256 256128 128Slice thickness
6mm 2mm Acquisition time (TA)
58s
88s
film for the observer.The readers were given a10-min training session which explained the characteristics of the morphological features.For each lesion the following morphological criteria were evaluated:lesion shape(round,oval,lobular,irregular),irregular-ityof contour(smooth to irregular on a five-point scale),homoge-neityof contrast enhancement(homogeneous to inhomogeneous on a five-point scale)and the presence of ring enhancement(posi-tive or negative).
Figures1,2and3demonstrate examples of the different mor-phological features.
The inter-observer variabilitywas evaluated bythe k-coeffi-cient,which displayed the agreement between the different ob-servers in the evaluation of each morphological criterion.A weighted k-test was applied which takes the range of disagree-ments between the observers into account.A low k-coefficient means that a lesion abnormalitywas evaluated differentlybydif-ferent observers.A k-value<0.2indicates poor agreement, k=0.21±0.40indicates fair agreement,k=0.41±0.60indicates moderate agreement,k=0.61±0.80indicates good agreement and k=0.81±1.00indicates verygood agreement[15].
The receiver operating characteristic(ROC)curve(Fig.4)de-scribes the accuracyof a diagnostic test,bythe relationship be-tween the true positive fraction(sensitivity)and the false positive fraction(1±specificity)due to different thresholds.Sensitivity var-ies inverselyas the confidence threshold is changed.A ROC curve plots the true positive fraction(TPF)as a function of the false positive fraction(FPF).A larger area under the ROC curve indi-cates greater discrimination capacity,because a larger value of TPF can be achieved for each value of FPF.
In this study,MRI studies of all lesions were evaluated by means of ROC analysis with five independent observers.ROC analysis with calculation of the area under the curve(AUC)was performed using the correlated receiver operating characteristics (CORROC2)computer program developed byMetz[16].Reader-data for each of the morphological criteria were analysed.Data averaging for each feature was performed bycalculating the curve estimate(true positive and false positive)for each of the five readers and averaging the resulting parameter values[16].The ROC curves allowed direct comparison to be made of the diag-nostic accuracies of each criterion estimated bythe AUC curve (AZ).
Results
The inter-rater agreement for the evaluation of the dif-ferent configurations of the morphological featureale-
1647
Fig.1a±e MR-mammographyof a fibroadenoma in the right
breast of an18-year-old patient,documented in a the precontrast,
b,c first and third postcontrast and d,e corresponding subtraction
series.This lesion shows the typical morphological characteristics
of a benign lesion including a lobular lesion shape,a smooth bor-
der definition,homogeneous contrast enhancement and the ab-
sence of ring
enhancement
Fig.2a±d A37-year-old patient with an invasive ductal carcinoma
in the right breast documented in a,b the first and third postcon-
trast and c,d corresponding subtraction series.This lesion shows
the typical features of malignancy including an irregular lesion
shape,irregular border definition,inhomogeneous contrast en-
hancement and the presence of ring
enhancement
Fig.3a±d A37-year-old woman with recurrent invasive lobular
carcinoma close to an implant in the left breast documented in a,b
the first and third postcontrast and c,d the corresponding subtrac-
tion series.This lesion is a good example of irregular border defi-
nition,ring enhancement with peripheral contrast enhancement
and no enhancement at the
centre
Fig.4Receiver operating characteristic(ROC)curve of the crite-
rionairregularityof contouro
sion shapeoshowed an average k-coefficient of0.38with a range from0.59to0.29.
The five readers observed different features for be-nign and malignant lesions.Table3shows the frequen-cies obtained for the different shapes for benign and malignant lesions.The most frequent configuration of the benign lesions(40%)wasalobularofollowed by aroundo(32%),aovalo(15%)andairregularo(13%). With a frequencyof119out of220possibilities(five readers and44carcinomas)an irregular lesion shape was the most prominent feature in the evaluation of the carcinomas,followed byalobularo(26%),aovalo(15%)andaroundo(5%)configurations.
The ROC analysis measured a sensitivity of78%and a specificityof65%with an AUC of0.81.
The criterionairregularityof contourodifferentiated betweenasmoothoandairregularoon a five-point scale. The inter-rater agreement for the evaluation of the bor-der definition showed an average k-coefficient of0.36 with a range from0.52to0.20.
Table4shows the results of the ROC analysis in-cluding sensitivity,specificity and AUC.The criterion airregularityof contourohad a sensitivityof83%and specificityof76%.The AUC was0.9.
The criterionahomogeneityof contrast enhance-mentowas evaluated on the post-contrast and subtrac-tion images,and differentiated between homogeneous and inhomogeneous contrast enhancement on a five-point scale.The inter-rater agreement for the evaluation of homogeneous and inhomogeneous contrast en-hancement had an average k-coefficient of0.35with a range from0.28to0.43.
The ROC analysis measured a sensitivity of85%and a specificityof42%.The AUC was0.70(Table4). Presence of ring enhancement showed the best inter-rater agreement with an average k-coefficient of0.4,a maximum of0.67and minimum of0.10.
Table5shows the frequencyof positive and negative ring enhancement in benign and malignant lesions.In malignant lesions a positive ring enhancement was found in43%,compared with27%in benign lesions.In approximatelytwo-thirds of the lesions there was no ring enhancement observed in either group(benign or malignant),whereas positive ring enhancement was more frequent in carcinomas.The ROC analysis mea-sured a sensitivityof71%and a specificityof53%.The AUC was0.64.
Discussion
Recent studies emphasize the usefulness of analysis of morphological features,because there are malignant lesions,such as certain invasive ductal and lobular car-cinomas and certain ductal carcinoma in situ lesions, that will not enhance rapidlybut in which lesion mor-phologysuggests the presence of malignancy[17,18]. An effective use of morphological criteria is onlypossi-ble provided that there is sufficient agreement among the observers in the evaluation of each feature.Mor-phological criteria with a high inter-observer variability documented bya low k-coefficient are not qualified for clinical use without improvement of terminology.So far, inter-observer variabilityhas onlybeen examined in a few studies.Mussurakis et al.[19]showed onlya mod-erate agreement between two experienced readers in rating morphological characteristics.
In the present studythe statistical analy sis of the in-ter-observer variabilityshowed onlyan average inter-rater agreement for each of the morphological criteria. In contrast with the average overall inter-rater agree-ment,the combination of the two most experienced breast MRI-readers showed a moderate to good inter-observer variability.Furthermore,there is a wide vari-abilitybetween different observer combinations.In the evaluation of the criterionaring enhancementoour in-ter-observer variabilityranged from good to poor(see Table4).With respect to the partiallyachieved good inter-observer variabilitywe conclude that there is a possibilityof obtaining good results after better train-ing.The large variabilitybetween different observer combinations as well as the average inter-rater agree-ment maybe partiallyexplained bydifferent experience
1648
Table3Frequencies of the criterionalesion shapeo
Shape Benign Malignant
Round32%(28/90)5%(12/220)
Oval15%(14/90)15%(32/220)
Lobular40%(36/90)26%(57/220)
Irregular13%(12/90)54%(119/220)
Table4Summaryof results of the receiver operating characteristic(ROC)test and k-coefficient for the morphological criteriaalesion contouro,ahomogeneityof contrast enhancemento,alesion shapeoandaring enhancemento(AUC area under the curve) Morphological criteria k-Coefficient SensitivitySpecificityAUC Lesion contour0.36(0.21±0.52)83%76%0.90 Homogeneityof contrast enhancement0.35(0.28±0.43)85%42%0.70 Shape0.38(0.29±0.59)78%65%0.81 Ring enhancement0.40(0.10±0.67)71%53%0.64
in breast MRI reading of the five observers.Different readers often perceived the same abnormalitydiffer-ently.For example,one reader regarded a mass as well-defined,whereas the others classified parts of the mar-gin to be ill-defined.This occurs due to a lack of stan-dardized terminologyfor the description of the lesions seen with breast MRI.Similar problems are document-ed in mammography.After a standardized terminology (BI-RADS)in mammographyhad been developed,in-ter-observer variabilitydecreased[20].
In previous literature,morphological assessment of breast MRI referred to onlya description.In our study sensitivityand specificitywere evaluated for four dif-ferent selected morphological features.Best results for the differentiation of benign and malignant lesions were obtained for the criterionairregularityof contouro.A specificitybeing as high as76%shows that an irregular border is specific in the characterization of carcinomas and is therefore useful in the differentiation between benign and malignant.Furthermore this criterion is sensitive for the detection of malignant lesions,because more than80%of the carcinomas presented an irregu-lar border.The ROC curve in Fig.4,with an AUC of0.9 demonstrates the high diagnostic accuracyand discrim-ination capacityof this criterion.
The criterionainhomogeneous contrast enhance-mentoseems to be sensitive for detection of carcinomas. Eighty-five percent of the malignant lesions showed an inhomogeneous contrast enhancement.However,due to a low specificityof42%this feature is not useful in the differentiation of benign and malignant lesions,if used as a single feature.A more specific form of inho-mogeneous contrast enhancement is the ring enhance-ment.Peripheral enhancement is explained histopatho-logicallybya necrotic or fibrotic zone in the tumour centre and an angiogenic activityin the viable tumour tissue at the peripheryof the tumour[12,21].We found ring enhancement to be more common in carcinomas (43%),but it was also observed in27%of the benign lesions.Our most specific feature for malignancycon-cerning thealesion shapeowas irregular configuration, whereas in benign lesions a lobular and round lesion shape was more frequentlyobserved.
A summaryof architectural features indicating ma-lignancymentioned in the literature[12,17,22,23]in-clude masses with irregular or spiculated borders,pe-ripheral enhancement and ductal enhancement.Fea-tures suggesting benign disease include a mass with smooth or lobulated borders,with no contrast enhance-ment,nonenhancing internal septations and patchypa-renchymal enhancement.
In comparison with the literature,an irregular lesion contour,being the most specific feature of malignancyin our study,is also described as being characteristic of carcinomas in several other studies as well,whereas be-nign lesions are characterized bya smooth and well-de-fined border[6,8,24,25,26].Heywang-K?brunner proclaims focal enhancement with particular irregular contour to be a suspected feature of malignancy[27]. In the literature inhomogeneous,and in particular peripheral enhancement,is regarded as being suggestive of malignancy[12,21,24,28,29,30,31,32].There are some descriptions of the contrast enhancement of be-nign and malignant lesions.Boetes et al.define en-hancements that start in the peripheryand progress to the centre of the lesion as centripetal enhancement[29], whereas central enhancement with an initial signal in-crease in the lesion centre and delayed,weak enhance-ment in the peripheryis regarded as centrifugal en-hancement and a sign of benignity.Rubens et al.[30] described peripheral enhancement with a low central signal intensityin carcinomas,compared with uniform contrast enhancement in sclerosing adenosis and fi-broadenoma.In a more recent studyBuadu et al.[33] examined different patterns of peripheral enhancement in breast MRI.Their conclusion is that earlyperipheral enhancement with progression to the centre appears to be fairlyspecific for carcinomas,whereas earlyen-hancement with minimal or no central progression,al-though more common in malignant tumours,maybe seen in some benign lesions as well.With respect to our results this maypartlyexplain that almost30%of the benign lesions showed a positive ring enhancement. The descriptions about lesions shape are mainlyfo-cused on focal,patchyenhancement[17,24].However, according to the results of our studyan irregular lesion shape seems to be suggestive of malignancy.
Our conclusions are that morphological criteria are useful in the differentiation between benign and malig-nant lesions.With respect to the insufficient results of inter-observer variability,we feel that the terminology for morphological description has to be standardized in order to achieve effective use.Furthermore the inte-gration of both kinetic and morphological information mayultimatelybe needed to achieve optimal discrimi-nation[18].Therefore further investigations and evalu-ations are needed for the development of interpretation models,including a combination of both methods.
1649
Table5Frequencies of the criterionaring enhancemento
Benign Malignant Positive27%(24/90)43%(94/220)
Negative73%(66/90)57%(126/220)
1650 References
1.Wingo PA,Tong T,Bolden S(1995)
Cancer statistics1995.Ca Cancer J Clin 45(1):8±30
2.Samuels JR,HafftyBG,Lee CH,F i-
scher DB(1992)Breast conservation
therapyin patients with mammograph-icallyundetected breast cancers.Radi-
ology185(21):425±427
3.Bird RE,Wallace TW,Yankaskas BC
(1992)Analysis of cancers missed at
screening mammography.Radiology
184(3):613±617
4.Baker LH(1982)Breast cancer detec-
tion demonstration project:five year
summaryreport.Ca Cancer J Clin
32(4):194±225
5.Harms SE,Flamig DP(1993)MR im-
aging of the breast:technical approach
and clinical experience.Radiographics
13(4):905±912
6.Heywang-K?brunner SH(1994)Con-
trast-enhanced magnetic resonance im-aging of the breast.Invest Radiol
29(1):94±104
7.Kaiser WA,Reiser M(1992)False-pos-
itive cases in dynamic MR mammogra-phy.Radiology185
8.Fobben ES,Rubin CZ,Kalisher L,
Dembner AG,Seltzer MH,Santoro EJ (1995)Breast MR imaging with com-
merciallyavailable techniques:radio-
logic-pathologic correlation.Radiology 196(1):143±152
9.Harms SE,F lamig DP,HesleyKL,et al.
(1993)Fat-suppressed three-dimen-
sional MR imaging of the breast.Ra-
diographics13(2):247±267
10.Funke M,Fischer U,Grabbe E(1996)
MR-Mammographie:Aktueller Stand
und Perspektiven.Aktuelle Radiol
6(3):130±135
11.Harms SE,Flamig DP,Evans WP,Har-
ries SA,Brown S(1994)MR imaging of the breast:current status and future
potential.AJR163(5):1039±1047 12.Orel SG,Schnall MD,LiVolsi VA,
Troupin RH(1994)Suspicious breast
lesions:MR imaging with radiologic-
pathologic correlation.Radiology
190(2):485±49313.Fischer U,Brinck U,Vosshenrich R
(1995)Accuracyof MR mammography
in ductal carcinoma in situ(abstract.).
Radiology197:129
14.Boetes C,Strijk SP,Holland R,Ba-
rentsz JO,Van Der Sluis RF,Ruijs JH
(1997)False-negative MR imaging of
malignant breast tumors.Eur Radiol
7(8):1231±1234
https://www.wendangku.net/doc/3015035532.html,ndis JR,Koch GG(1977)The mea-
surement of observer agreement for
categorical data.Biometrics
33(1):159±174
16.Metz CE(1986)ROC methodologyin
radiologic imaging.Invest Radiol
21(9):720±733
17.Nunes L W,Schnall MD,Orel SG,et al.
(1997)Breast MR imaging:interpreta-
tion model.Radiology202(3):833±841
18.Orel SG.(1999)Differentiating benign
from malignant enhancing lesions iden-
tified at MR imaging of the breast:are
time-signal intensitycurves an accurate
predictor?Radiology211(1):5±7
19.Mussurakis S,BuckleyDL,CoadyAM,
Turnbull L W,Horsman A(1996)Ob-
server variabilityin the interpretation
of contrast enhanced MRI of the breast.
Br J Radiol69(827):1009±1016
20.Baker JA,Kornguth PJ,Floyd CE Jr
(1996)Breast imaging reporting and
data system standardized mammogra-
phylexicon:observer variabilityin le-
sion description.Am J Roentgenol
166(4):773±778
21.Kaiser WA(1994)False-positive results
in dynamic MR mammography.Causes,
frequency,and methods to avoid.Magn
Reson Imaging Clin N Am.
2(4):539±555
22.Stomper PC,Herman S,Klippenstein
DL,Winston JS,Edge SB,Arredondo
MA,Mazurchuk RV,Blumenson LE
(1995)Suspect breast lesions:findings
at dynamic gadolinium-enhanced MR
imaging correlated with mammograph-
ic and pathologic features.Radiology
197(2):387±395
23.Harms SE,F lamig DP,HesleyKL,et al.
(1993)MR imaging of the breast with
rotating deliveryof excitation off reso-
nance:clinical experience with patho-
logic correlation.Radiology
187(2):493±501
24.Kuhl CK(2000)MRI of breast tumors.
Eur Radiol10(1):46±58
25.Adler DD,Wahl RL(1995)New meth-
ods for imaging the breast:techniques,
findings,and potential.Am J Roentge-
nol164(1):19±30
26.Gilles R,Guinebretiere JM,Lucidarme
O,Cluzel P,Janaud G,Finet JF,Tar-
divon A,Masselot J,Vanel D(1994)
Nonpalpable breast tumors:diagnosis
with contrast-enhanced subtraction dy-
namic MR imaging.Radiology
191(3):625±631
27.Heywang-K?brunner SH(1990)Con-
trast-enhanced MRI of the breast.Kar-
ger Editor(ed),Basel,Switzerland
28.Aoki S,Sakamoto T,Koutake H,Ta-
naka T,Murata K,Morita R(1993)
Dynamic MR mammography:value of
imaging during the first minute(ab-
stract).Radiology189(P):105
29.Boetes C,Barentsz JO,Mus RD,et al.
(1994)MR characterization of suspi-
cious breast lesions with gadolinium-
enhanced TurboFLASH subtraction
technique.Radiology193(3):777±781
30.Rubens D,Totterman S,Chacko AK,
et al.(1991)Gadopentate dimeglu-
mine-enhanced chemical shift MR im-
aging of the breast.Am J Roentgenol
157(2):267±270
31.Stelling CB(1995)MR imaging of the
breast for cancer evaluation.Current
status and future directions.Radiol Clin
North Am33(6):1187±1204
32.Friedrich M(1998)MRI of the breast:
state of the art.Eur Radiol8(5):707±725
33.Buadu LD,Murakami J,Murayama S,
Ohno S,et al.(1997)Patterns of pe-
ripheral enhancement in breast masses:
correlation of findings on contrast me-
dium enhanced MRI with histologic
features and tumor angiogenesis.
J Comput Assist Tomogr21(3):421±430