Sie sind auf Seite 1von 6

Breast Imaging

Hong et al.
BI-RADS for Sonography

BI-RADS for Sonography: Positive


and Negative Predictive Values of
Sonographic Features
Andrea S. Hong1 OBJECTIVE. The purpose of this study was to assess the positive predictive value (PPV)
Eric L. Rosen and negative predictive value (NPV) of features described in the new sonographic BI-RADS
Mary S. Soo lexicon for evaluating solid masses with known histologic diagnoses.
Jay A. Baker MATERIALS AND METHODS. Sonograms of 403 solid lesions were analyzed by one
Hong AS, Rosen EL, Soo MS, Baker JA of three dedicated breast radiologists. Each lesion was described using features from the sono-
graphic BI-RADS lexicon. Lesion description and biopsy results were correlated. PPV and
NPV were calculated.
RESULTS. Histologic results showed that 141 (35%) of 403 masses were malignant. Sono-
graphic BI-RADS descriptors showing high predictive value for malignancy include spiculated
margin (86%, 19/22), irregular shape (62%, 102/164), and nonparallel orientation (69%, 75/109).
Sonographic BI-RADS descriptors highly predictive of benign lesions include circumscribed
margin (90%, 160/178), parallel orientation (78%, 228/294), and oval shape (84%, 200/237). For
the sonographic BI-RADS features of mass margin, shape, orientation, lesion boundary, echo pat-
tern, and posterior acoustic features, descriptors chosen were significantly (p < 0.001) different
for malignant and benign masses.
CONCLUSION. Descriptors from the new sonographic BI-RADS lexicon can be useful
in differentiating benign from malignant solid masses.

I-RADS was developed in 1993 sessed the utility of sonographic features in

B by the American College of Radi-


ology (ACR) to standardize the
language of mammography re-
distinguishing benign from malignant le-
sions [8–10]. However, to our knowledge,
no studies to date have assessed the PPV and
porting, to clarify mammographic inter- negative predictive value (NPV) of sono-
pretations, and to facilitate communication graphic features as described in the new
between clinicians [1–4]. Until recently, sonographic BI-RADS lexicon. The purpose
BI-RADS had been applied to mammography of this study was to evaluate the sonographic
only and did not pertain to other breast imag- features of solid breast lesions with known
ing techniques. Breast sonography is now a histologic diagnosis and determine the pre-
well-established adjunct to mammography. dictive value of features from the new sono-
In light of the widespread use of sonography, graphic BI-RADS lexicon for malignant
the ACR recently developed a BI-RADS lex- versus benign diagnosis.
icon for breast sonography to standardize the
characterization of sonographic lesions [4, 5]. Materials and Methods
This lexicon includes descriptors of features Patient Population
Received May 19, 2004; accepted after revision such as mass shape, orientation, margin, and Institutional review board approval was ob-
August 11, 2004.
posterior acoustic transmission, and other tained for this retrospective study; however, in-
Supported by National Institutes of Health grant number sonographic features. formed consent was not required.
R01-CA090561-01A1.
The positive predictive value (PPV) of Cases for analysis in this study were selected
1All authors:
Department of Radiology, Duke University mammographic features described in the from those recommended for biopsy. Between Feb-
School of Medicine, DUMC 3808, Durham, NC 27710.
Address correspondence to J. A. Baker.
original mammography BI-RADS lexicon ruary 23, 2000, and February 28, 2002, 654 female
has been investigated [6, 7]. These studies patients with 738 lesions underwent sonographi-
AJR 2005;184:1260–1265
found that the mammography lexicon was cally guided biopsy at our facility. A lesion was in-
0361–803X/05/1844–1260 useful in differentiating benign and malig- cluded in this study if it corresponded to a solid
© American Roentgen Ray Society nant breast lesions. Several studies have as- mass on sonography and if both mammographic

1260 AJR:184, April 2005


BI-RADS for Sonography

and sonographic films taken before the biopsy were ditional true lateral and spot compression magnifi- formed by one of five dedicated breast imaging
available for review. Forty lesions were excluded cation mammograms available in almost all cases. radiologists with 6–18 years of experience, and
from the study because they underwent biopsy be- Young patients under 30 years old with palpable ab- mammographic views were available to the radi-
fore acquisition of the mammographic or sono- normalities received a single unilateral mediolateral- ologist at the time the sonogram was performed.
graphic images, because such biopsies can cause oblique view to limit radiation to the developing Sonography was performed using a variable-fre-
tissue changes that distort the appearance of the breast. Mammographic breast composition re- quency linear transducer set at 12 MHz (Sonoline
mass. In addition, 181 lesions were excluded be- vealed BI-RADS type 1 (fatty) in 16 (4%) of the Elegra, Siemens Medical Solutions). For lesions
cause either mammographic or sonographic films 403 cases, type 2 (scattered fibroglandular densi- in the lateral aspect of the breast, the patient was
were unavailable for review, 82 lesions were ex- ties) in 157 cases (39%), type 3 (heterogeneously imaged in the supine-oblique position, and for
cluded because no solid mass was found at histol- dense) in 162 cases (40%), and type 4 (extremely other lesions, the patient was supine. Images were
ogy (including 68 cysts, six ducts, one abscess, and dense) in 68 cases (17%). One hundred ten cases acquired in both radial and antiradial projections
seven cases of normal breast tissue), 13 lesions (27%) were occult to mammography but visible on with and without caliper measurements. Addi-
were excluded because they were found to be sonography. Two hundred thirty-three lesions were tional gray-scale images were obtained in almost
lymph nodes at histologic diagnosis, and 10 lesions visible on mammography as a discrete mass seen all cases to better show the lesion. Doppler, color
were excluded because final pathologic diagnoses on both craniocaudal and mediolateral-oblique pro- Doppler, and power Doppler images were not part
were obtained at other facilities and were unavailable. jections. An additional 60 lesions were visible on of the routine imaging protocol.
Nine lesions were excluded because only nonpalpable mammography as focal asymmetry (n = 24), archi-
calcifications were observable on mammography, tectural distortion (n = 32), and calcifications (n = Imaging Interpretation and Data Analysis
and such lesions would not typically be sent for 4) presenting as a palpable mass. For this study, each case was evaluated by one
sonography and sonographically guided biopsy in of three dedicated breast radiologists with 6–11
clinical practice. The remaining 403 lesions in 369 Sonographic Imaging Technique years of experience. Each case was analyzed by
patients constituted the study population, including Sonographic examinations were performed to only one observer. Information about the patient’s
202 palpable lesions and 201 nonpalpable ones. evaluate mammographically identified masses or age, physical examination findings, family history
The median age for these patients was 50.0 years densities in 161 (40%) of 403 cases, palpable le- of breast cancer, and personal history of breast
(age range, 18–89 years). sions in 66 cases (16%), and lesions that were both malignancy was available to each radiologist to
palpable and mammographically visible in 132 best reproduce a realistic clinical situation. The
Mammography cases (33%). Incidental lesions in 44 cases (11%) radiologist was blinded to the histologic diagnosis
Most patients underwent initial craniocaudal were noted during sonography of other mammo- during the evaluation.
and mediolateral-oblique mammography, with ad- graphic or palpable masses. Sonography was per- The interpreting radiologist first evaluated the
mammograms. The observer determined the breast
Frequency of Benign and Malignant Masses for Mammography parenchyma density and then determined if the le-
TABLE 1 sion in question was visible on mammography. If
BI-RADS Lexicon Descriptors
visible, the lesion was described using BI-RADS
Descriptor na Benign Malignant p mammographic descriptors of mass margin (cir-
Mass shape < 0.0001 cumscribed, obscured, microlobulated, ill-defined/
Oval 113 (48)b 90 (80)c 23 (20)d indistinct, or spiculated), shape (oval, round, lobu-
lar, or irregular), and density (fat-containing, low,
Round 10 (4) 6 ( 60) 4 (40)
equal, or high). The presence or absence of calcifi-
Lobular 52 (22) 40 (77) 12 (23)
cations was noted, along with any associated find-
Irregular 58 (25) 10 (17) 48 (83) ings (e.g., architectural distortion) or special cases
Mass margin < 0.0001 (e.g., axillary lymphadenopathy).
Circumscribed 29 (12) 29 (100) 0 (0) After initial interpretation of mammograms,
Obscured 108 (46) 91 (84) 17 (16) prior mammographic views were presented for
comparison if available. Prior mammograms were
Microlobulated 10 (4) 5 (50) 5 (50)
available for 195 (48%) of the 403 lesions.
Indistinct 50 (21) 20 (40) 30 (60) After reviewing the mammographic images, the
Spiculated 36 (15) 1 (3) 35 (97) observer was presented with static sonographic im-
Mass density 0.0003 ages. The observer was first asked to assess each
Fat-containing 4 (2) 4 (100) 0 (0) mass using the lexicon popularized by Stavros et al.
[8]. Subsequently, the observer was asked to assess
Low 5 (2) 4 (80) 1 (20)
the same images using features of the new sono-
Equal 209 (90) 136 (65) 73 (35)
graphic BI-RADS lexicon [4].
High 15 (6) 2 (13) 13 (87) For each category from the mammographic
Note.—BI-RADS = Breast Imaging Reporting and Data System [3]. BI-RADS lexicon, the sonographic lexicon described
aTotal number of mammographically visible solid masses = 233.
bNumber in parentheses indicates percentage of 233 masses; percentages occasionally do not add to 1.0 due
by Stavros et al. [8], and the sonographic BI-RADS
lexicon, the radiologist was limited to selecting one
to rounding.
cNumber in parentheses indicates percentage of benign lesions among total number of masses with given descriptor. best feature descriptor. If the lesion could be de-
dNumber in parentheses indicates percentage of malignant lesions among total number of masses with given descriptor. scribed by more than one term, the observer was in-

AJR:184, April 2005 1261


Hong et al.

structed to choose the term that described a lesion 0.0005) between descriptors of benign masses and architectural distortion, and one mass
that was most suspicious for malignancy. and those of malignant masses is shown. Spic- with Cooper’s ligament changes and skin re-
Sonographically guided core needle biopsy was ulated margin (PPV = 86%, 19/22), irregular traction. The other 51 cases had 24 instances
performed for each lesion using a 14-gauge needle shape (PPV = 62%, 102/164), and nonparallel of Cooper’s ligament change, 15 examples of
(Achieve, Allegiance Health Care; C. R. Bard). The orientation (PPV = 69%, 75/109) display high architectural distortion, seven instances of
histologic diagnosis was recorded for each case and predictive value for malignancy. Circum- duct change, and five cases of edema. Lesions
correlated with the imaging feature analysis. Le- scribed margin, oval shape, and parallel orien- in 42 (75%) of the 56 cases were malignant.
sions initially diagnosed by core needle biopsy as tation are predictive of a benign lesion with Similarly, 76% (16/21) of masses with calci-
atypia, malignant, and having discordant imaging NPVs of 90% (160/178), 84% (200/237), and fications were malignant, although only 21
and histology underwent surgical excision. For dis- 78% (228/294), respectively. masses were described as having either mac-
cordant and atypical lesions, final histologic diag- Surrounding tissue effects were defined in rocalcifications or microcalcifications visible
noses were based on excisional biopsy results. only 56 (14%) of 403 masses. These included on sonography.
Statistical software (SAS software system, ver- one mass with three surrounding tissue ef- Special cases were infrequently identified,
sion 8.2, SAS Institute) was used to calculate the fects (Cooper’s ligament changes, skin thick- which is understandable because only solid
Pearson’s chi-square test when assessing statistical ening, and architectural distortion), two masses that had undergone biopsy were in-
significance. masses with Cooper’s ligament changes and cluded in this study. Foreign bodies, clusters of
architectural distortion, one mass with edema microcysts, and masses in or on the skin are
Results
Of the 403 lesions, 141 (35%) were malig-
nant at histologic diagnosis. Of the malignant
masses, the most common diagnosis was in- Frequency of Benign and Malignant Masses for Stavros Lexicon
TABLE 2
Sonographic Descriptors
vasive ductal carcinoma, which was found in
120 (85%) of the 141 cancers. Other diag- Descriptor na Benign Malignant p
noses included invasive lobular carcinoma in Mass margin < 0.0001
14 (10%) of 141 malignancies and ductal car-
Well-circumscribed 193 (48)b 173 (90)c 20 (10)d
cinoma in situ in three cases (2%). The re-
maining four malignancies were shown to be Microlobulations 46 (11) 24 (52) 22 (48)
a high-grade phyllodes tumor, follicular lym- Angular 123 (31) 55 (45) 68 (55)
phoma, B-cell lymphoma, and papillary car- Duct extension 15 (4) 6 (40) 9 (60)
cinoma in situ. The remaining 261 (65%) of Branch pattern 3 (1) 2 (67) 1 (33)
403 total lesions were benign without atypia Spiculation 23 (6) 2 (9) 21 (91)
by histology.
Mass shape < 0.0001
Of 233 lesions visible as a mass on mam-
mography, 87 (37%) were malignant. PPVs Ellipsoid 296 (73) 231 (78) 65 (22)
for mammographic BI-RADS features are Taller than wide 107 (27) 31 (29) 76 (71)
shown in Table 1. Descriptors chosen for fea- Thin pseudocapsule < 0.0001
tures of mass shape, margin, and density are Absent 337 (84) 199 (59) 138 (41)
significantly (p < 0.0003) different for malig-
Present 66 (16) 63 (95) 3 (5)
nant versus benign masses.
Table 2 shows the predictive value of mor- Acoustic transmission < 0.0001
phology descriptors defined by Stavros et al. Enhanced 91 (23) 58 (64) 33 (36)
[8]. The descriptors of spiculated margin Normal 173 (43) 138 (80) 35 (20)
(91%, 21/23) and taller-than-wide shape Shadowing 139 (34) 66 (47) 73 (53)
(71%, 76/107) were highly predictive of ma- Mass echogenicity compared with fat < 0.0001
lignancy. The presence of a thin echogenic
Hyperechoic 8 (2) 8 (100) 0 (0)
pseudocapsule and a circumscribed margin
was highly predictive of a benign diagnosis, Isoechoic 54 (13) 46 (85) 8 (15)
with NPVs for malignancy of 95% (63/66) Mildly hypoechoic 254 (63) 175 (69) 79 (31)
and 90% (173/193), respectively. Descriptors Markedly hypoechoic 87 (22) 33 (38) 54 (62)
for the Stavros features shown are signifi- Calcifications within mass 0.0001
cantly (p ≤ 0.0001) different between malig-
Absent 381 (95) 256 (67) 125 (33)
nant and benign masses.
The PPV of malignancy for the sono- Present 22 (5) 6 (27) 16 (73)
graphic BI-RADS lexicon features were also Note.—Descriptors as described by Stavros et al [8].
aTotal number of solid masses = 403.
calculated (Table 3). Predictive values for bNumber in parentheses indicates percentage of 403 masses; percentages occasionally do not add to 1.0 due
sonographic mass shape and mass margin in to rounding.
palpable and nonpalpable masses only are also cNumber in parentheses indicates percentage of benign lesions among total number of masses with given descriptor.

shown (Table 4). A significant difference (p ≤ dNumber in parentheses indicates percentage of malignant lesions among total number of masses with given descriptor.

1262 AJR:184, April 2005


BI-RADS for Sonography

unlikely to be recommended for biopsy, and which further shows the value of BI-RADS to cumscribed margin, and had a width-to-an-
complicated cysts are not solid lesions and thus the practice of mammography interpretation. teroposterior ratio greater than 1.4. Features
were excluded from this study. Three lesions Although mammography is recognized as most predictive of malignancy were irregu-
(1%) in 403 cases were described as special the best method of screening for breast can- lar shape, microlobulated or spiculated mar-
cases (two intramammary lymph nodes and cer, breast sonography has become well es- gin, and width-to-anteroposterior ratio of
one axillary lymph node) but were not found to tablished as a valuable imaging technique. less than or equal to 1.4.
represent lymph nodes at histology. Although there has been some controversy Stavros et al. [8] developed a classification
regarding the utility of sonography when scheme for solid breast masses that has a
Discussion evaluating solid breast masses for the likeli- 98.4% sensitivity and 99.5% NPV for malig-
The BI-RADS lexicon was first developed hood of malignancy [12, 13], several studies nancy. Of particular note, Stavros et al. spe-
in 1993 for use in mammography reporting. have suggested that sonographic appearance cifically described and gave pictorial
Since its establishment, several studies have can be useful in differentiating benign from examples of lesions illustrating each sono-
found that use of mammographic BI-RADS malignant solid breast masses [8–10]. Rah- graphic descriptor in the lexicon used to build
terminology can be helpful in predicting the bar et al. [9] found that the features most the classification system, thus enabling the
likelihood of cancer [6, 7, 11]. These results likely to predict a benign diagnosis in solid continued use of these descriptors in clinical
are comparable with those found in our study, masses were round or oval shape, had a cir- practice. Despite the reported high sensitivity
and NPV of the lexicon popularized by
Frequency of Benign and Malignant Masses for Sonographic BI-RADS Stavros et al. [8], Baker et al. [14] found that
TABLE 3 interobserver agreement was at best moderate
Lexicon Descriptors
for six of seven sonographic features. This
Descriptor na Benign Malignant p
lack of consistency suggested that a more glo-
Mass shape < 0.0001 bal standardized terminology was necessary
Oval 237 (59)b 200 (84)c 37 (16)d for general clinical use.
Round 2 (0) 0 (0) 2 (100) A standardized lexicon for sonography was
developed in 2003 by the ACR in light of the
Irregular 164 (41) 62 (38) 102 (62)
increasing use of sonography in clinical prac-
Mass margin < 0.0001 tice. Like its mammographic counterpart, the
Circumscribed 178 (44) 160 (90) 18 (10) sonographic BI-RADS lexicon was intended
Microlobulated 37 (9) 18 (49) 19 (51) to provide a unified language for sonographic
Indistinct 104 (26) 56 (54) 48 (46) reporting and research and to avoid ambiguity
Angular 62 (15) 25 (40) 37 (60)
in the communication and teaching of sono-
graphic interpretation [3–5]. The study pre-
Spiculated 22 (5) 3 (14) 19 (86)
sented here was undertaken to explore the
Mass orientation < 0.0001 PPV and NPV values of these new sono-
Parallel 294 (73) 228 (78) 66 (22) graphic BI-RADS features.
Not parallel 109 (27) 34 (31) 75 (69) The sonographic BI-RADS descriptors of
Posterior acoustic features < 0.0001 spiculated margin, irregular shape, and non-
parallel orientation showed high predictive
Enhancement 85 (21) 57 (67) 28 (33)
value for malignancy in this study, whereas
No posterior acoustic features 170 (42) 134 (79) 36 (21) circumscribed margin, oval shape, and par-
Shadowing 132 (33) 63 (48) 69 (52) allel orientation were highly predictive of a
Combined pattern 16 (4) 8 (50) 8 (50) benign diagnosis. Similar results were found
Lesion boundary 0.0004 whether features were assessed for palpable
or nonpalpable masses. This stratification
Abrupt interface 342 (85) 244 (71) 98 (29)
can be understood in view of the abnormal
Echogenic halo 61 (15) 18 (30) 43 (70)
processes that these descriptors represent
Echo pattern 0.0005 [4]. As in mammography, sonographic evi-
Hyperechoic 6 (1) 6 (100) 0 (0) dence of spiculated margin suggests infiltrat-
Isoechoic 49 (12) 41 (84) 8 (16) ing growth of the lesion into the surrounding
Hypoechoic 324 (80) 195 (60) 129 (40) tissue, whereas an irregular shape can indi-
cate inconsistent growth and advancement
Complex 20 (5) 18 (90) 2 (10)
of the lesion edge. Nonparallel orientation
Anechoic 4 (1) 2 (50) 2 (50) on sonography can suggest spread of the lesion
Note.—BIRADS = breast imaging reporting and data system [3]. through tissue-plane boundaries. All of these
aTotal number of mammographically visible solid masses = 403.
bNumber in parentheses indicates percentage of 403 masses; percentages occasionally do not add to 1.0 due
characteristics are more likely to be associated
to rounding. with malignant lesions. In contrast, circum-
cNumber in parentheses indicates percentage of benign lesions among total number of masses with given descriptor. scribed margin and oval shape representing
dNumber in parentheses indicates percentage of malignant lesions among total number of masses with given descriptor .
smooth uniform growth without involvement

AJR:184, April 2005 1263


Hong et al.

of surrounding tissue are associated more Frequency of Benign and Malignant Masses for Sonographic BI-RADS
with a benign lesion. Similarly, parallel ori- TABLE 4
Descriptors, Separated by Physical Exam
entation suggesting containment in one tis-
Descriptor n Benign Malignant p
sue plane is more indicative of a benign
process. Nonpalpablea Mass shape < 0.0001
Sixteen (9%) of 172 masses described as Oval 123 (61)c 108 (88)e 15 (12)f
both oval and circumscribed on sonography Round 2 (1) 0 (0) 2 (100)
were malignant at histology. This relatively
Irregular 76 (38) 32 (42) 44 (58)
high malignancy rate in the setting of appar-
ently benign sonographic features is likely Mass margin < 0.0001
due to concomitant mammographic features Circumscribed 91 (45) 81 (89) 10 (11)
that appeared more worrisome and prompted Microlobulated 12 (6) 8 (67) 4 (33)
biopsy, including interval enlargement of a Indistinct 55 (27) 32 (58) 23 (42)
lesion compared with prior mammographic Angular 35 (17) 18 (51) 17 (49)
findings. In addition, the benign sonographic
Spiculated 8 (4) 1 (13) 7 (88)
features described by Stavros et al. [8] had a
slightly lower predictive value in our study Palpableb Mass shape < 0.0001
than was reported in that article. This is likely Oval 114 (56)d 92 (81) 22 (19)
due to the fact that the observing radiologists Round 0 0 0
in our study were interpreting static images, Irregular 88 (44) 30 (34) 58 (66)
whereas real-time evaluation might have pro-
Mass margin < 0.0001
vided more information, leading to the choice
of a more suspicious descriptor. Whereas Circumscribed 87 (43) 79 (91) 8 (9)
static image interpretation is not typical of our Microlobulated 25 (12) 10 (40) 15 (60)
routine practice, viewing static images repre- Indistinct 49 (24) 24 (49) 25 (51)
sents a common method for interpreting clin- Angular 27 (13) 7 (26) 20 (74)
ical breast sonograms. Spiculated 14 (7) 2 (14) 12 (86)
Identification of surrounding tissue ef-
Note.—BI-RADS = Breast Imaging Reporting and Data System [3].
fects such as edema, architectural distor- a201 total nonpalpable masses.
tion, or changes to the Cooper’s ligaments b202 total palpable masses.
was infrequent. However, identification of cNumber in parentheses indicates percentage of 201 nonpalpable masses; percentages occasionally do not

surrounding tissue effects had a high predic- add to 1.0 due to rounding.
dNumber in parentheses indicates percentage out of 202 palpable masses.
tive value for malignancy, suggesting that eNumber in parentheses indicates percentage of benign lesions among total number of masses with given descriptor.
recognition of such features could be help- fNumber in parentheses indicates percentage of malignant lesions among total number of masses with given descriptor.
ful in the final assessment of a sonogram. In
addition, the observers in this study occa-
sionally encountered some difficulty in de- would be desirable for a more complete as- mammographic appearance of the lesion
scribing a sonographic feature with the sessment of the utility of the sonographic could influence the radiologist’s assessment
terms offered by the sonographic BI-RADS BI-RADS lexicon. of the sonographic image, mammograms
lexicon. For example, in describing the le- This study was limited to solid masses. In and sonograms are routinely interpreted in
sion boundary, the radiologist may have be- particular, for lesions such as simple cysts, tandem in actual clinical practice. This de-
lieved that neither of the two available which are often clearly benign on sonogra- sign was therefore intended to combat the
options—abrupt interface or echogenic phy, the NPV of benign descriptors (e.g., cir- artificial quality imposed by the retrospec-
halo—accurately reflected what he or she cumscribed margin) is likely to be much tive nature of this study and is in keeping
observed. These experiences may reflect higher. with the practice recommendations of tan-
some unfamiliarity with the relatively new This study has three additional weak- dem interpretation as described in the ACR
sonographic BI-RADS lexicon, and further nesses. First, biopsy had already been per- BI-RADS manual [4]. Third, only biopsy-
instruction and experience in the usage of formed for all patients included in this proven lesions were included in this study.
the BI-RADS descriptors may be warranted. study. Therefore, although they were Thus, this study did not address the predic-
Each case in this study was interpreted by blinded to biopsy results, the observers were tive value of BI-RADS features in more be-
one of three radiologists, all of whom were aware that their descriptions and assess- nign-appearing lesions that were interpreted
dedicated breast radiologists working at the ments did not directly affect patient care, as definitely benign or were recommended
same academic institution. No case was as- which could theoretically have affected for follow-up only. During the time covered
sessed by more than one observer. To our their choices. Second, when evaluating the by this study, however, follow-up assess-
knowledge, no studies have yet been pub- sonogram, the radiologist was not blinded to ments were rarely used in clinical practice at
lished showing the degree of intraobserver the mammogram and indeed was asked to this institution: that is, any solid mass that
or interobserver variability with the sono- analyze the mammogram immediately be- was not determined to be a benign finding
graphic BI-RADS criteria. Such studies fore analyzing the sonogram. Although the (i.e., BI-RADS category 2) virtually always

1264 AJR:184, April 2005


BI-RADS for Sonography

underwent biopsy. Therefore, few such References and correlation with large-core breast biopsy. Can
cases would be excluded from the study on 1. Kopans DB. Standardized mammographic report- Assoc Radiol J 1998;49:223–228
ing. Radiol Clin North Am 1992;30:257–261 8. Stavros T, Thickman D, Rapp CL, Dennis MA,
this basis. Regardless, it would be useful for
2. D’Orsi CJ, Kopans DB. Mammographic feature Parker SH, Sisney GA. Solid breast nodules: use
future studies to have a broader patient pop- of sonography to distinguish between benign and
analysis. Semin Roentgenol 1993;28:204–230
ulation for analyses, incorporating lesions 3. American College of Radiology. BI-RADS: malignant lesions. Radiology 1995;196:123–134
with long-term follow-up results in addition mammography. In: Breast imaging reporting and 9. Rahbar G, Sie AC, Hansen GC, et al. Benign ver-
to those with histologic diagnoses. data system: BI-RADS atlas, 4th ed. Reston, VA: sus malignant solid breast masses: US differenti-
ation. Radiology 1999;213:889–894
In conclusion, the ACR BI-RADS lexicon American College of Radiology, 2003
10. Harper AP, Kelly-Fry E, Noe JS, Bies JR, Jackson
provides standardized terminology to facilitate 4. American College of Radiology. BI-RADS: ultra-
VP. Ultrasound in the evaluation of solid breast
accurate and consistent breast sonography and sound, 1st ed. In: Breast imaging reporting and
masses. Radiology 1983;146:731–736
data system: BI-RADS atlas, 4th ed. Reston, VA:
mammography reporting. This study shows American College of Radiology, 2003
11. Baker JA, Kornguth PJ, Lo JY, Floyd CE. Breast
that features from the standardized sono- cancer: prediction with artificial neural network
5. Mendelson EB, Berg WA, Merritt CRB. Toward a based on BI-RADS standardized lexicon. Radiol-
graphic lexicon can be helpful in distinguish- standardized breast ultrasound lexicon, BI-RADS: ul- ogy 1995;196:817–822
ing benign from malignant solid masses. trasound. Semin Roentgenol 2001;36:217–225 12. Jackson VP. Management of solid breast nodules:
Further evaluation of the sonographic BI-RADS 6. Liberman L, Abramson AF, Squires FB, Glass- what is the role of sonography? Radiology
lexicon would be useful, including studies us- man JR, Morris EA, Dershaw DD. The Breast Im- 1995;196:14–15
aging Reporting and Data System: positive 13. Hall FM. Sonography of the breast: controversies
ing prospective real-time analyses of sono-
predictive value of mammographic features and and opinions. AJR 1997;169:1635–1636
grams and interobserver variability studies of final assessment categories. AJR 1998;171:35–40 14. Baker JA, Kornguth PJ, Soo MS, Walsh R, Men-
the descriptors. Such studies would help fur- 7. Bérubé M, Curpen B, Ugolini P, Lalonde L, Ouimet- goni P. Sonography of solid breast lesions: ob-
ther to assess the utility of the sonographic BI- Oliva D. Level of suspicion of a mammographic le- server variability of lesion description and
RADS lexicon in clinical breast radiology. sion: use of features defined by BI-RADS lexicon assessment. AJR 1999;172:1621–1625

AJR:184, April 2005 1265