RESEARCH ARTICLE
Diagnostic Performance of Breast Magnetic
Resonance Imaging in Non-Calcified
Equivocal Breast Findings: Results from a
Systematic Review and Meta-Analysis
Barbara Bennani-Baiti
1,2☯
*, Nabila Bennani-Baiti
3
, Pascal A. Baltzer
2☯
*
1 Department of Pharmaceutical Chemistry, University of Vienna, Vienna, Austria, 2 Department of
Biomedical Imaging and Image-guided Therapy, Vienna General Hospital (AKH), Medical University of
Vienna, Vienna, Austria, 3 Division of Hematology, Mayo Clinic, Rochester, Minnesota, United States of
America
☯ These authors contributed equally to this work.
*
barbara.bennani-baiti@meduniwien.ac.at (BBB); pascal.baltzer@meduniwien.ac.at (PAB)
Abstract
Objectives
To evaluate the performance of MRI for diagnosis of breast cancer in non-calcified equivo-
cal breast findings.
Materials and Methods
We performed a systematic review and meta-analysis of peer-reviewed studies in PubMed
from 01/01/1986 until 06/15/2015. Eligible were studies applying dynamic contrast-enhanced
breast MRI as an adjunct to conventional imaging (mammography, ultrasound) to clarify
equivocal findings without microcalcifications. Reference standard for MRI findings had to be
established by histopathological sampling or imaging follow-up of at least 12 months. Number
of true or false positives and negatives and other characteristics were extracted, and possible
bias was determined using the QUADAS-2 applet. Statistical analyses included data pooling
and heterogeneity testing.
Results
Fourteen out of 514 studies comprising 2,316 lesions met our inclusion criteria. Pooled diag-
nostic parameters were: sensitivity (99%, 95%-CI: 93–100%), specificity (89%, 95%-CI:
85–92%), PPV (56%, 95%-CI: 42–70%) and NPV (100%, 95%-CI: 99–100%). These esti-
mates dis played significant heterogeneity (P<0.001).
Conclusions
Breast MRI demonstrate s an excellent diagnostic performance in case of non-calcified
equivocal breast findings detected in conventional imaging. However, considering the
PLOS ONE | DOI:10.1371/journal.pone.0160346 August 2, 2016 1/16
a11111
OPEN ACCESS
Citation: Bennani-Baiti B, Bennani-Baiti N, Baltzer
PA (2016) Diagnostic Performance of Breast
Magnetic Resonance Imaging in Non-Calcified
Equivocal Breast Findings: Results from a Systematic
Review and Meta-Analysis. PLoS ONE 11(8):
e0160346. doi:10.1371/journal.pone.0160346
Editor: Kenneth G.A. Gilhuijs, Universitair Medisch
Centrum Utrecht, NETHERLANDS
Received: April 3, 2016
Accepted: July 18, 2016
Published: August 2, 2016
Copyright: © 2016 Bennani-Baiti et al. This is an
open access article distributed under the terms of the
Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any
medium, provided the original author and source are
credited.
Data Availability Statement: All relevant data are
within the paper and its Supporting Information files.
Funding: This article was supported by the Open
Access Publishing Fund of the University of Vienna.
Competing Interests: The authors have declared
that no competing interests exist.
substantial heterogeneity with regard to prevalence of malignancy, problem solving criteria
need to be better defined.
Introduction
Breast is the most frequently affected organ by cancer in women [
1]. Imaging plays a major
role in secondary and tertiary prevention of breast cancer. Depending on whether healthy
women are screened for breast cancer or whether assessment of symptomatic patients or
screening findings is perform ed, mammography, breast ultrasound and percutaneous image-
guided biopsies play a major role in diagnosis and to rule-out cancer [
2–4]. However, these
methods individually or in combination can yield inconclusive results, whereby the presence or
absence of bre ast cancer is not clearly ascertained. Not everyone agrees on what qualifies as an
equivocal finding. In clinical practice, a variety of results are usually classified as such: asymme-
try without associated microcalcifications, architectural distortions and other ambiguous
abnormalities such as multiple lesions, discrepancy between clinical symptoms and imaging
findings, benign biopsy results with insufficient radiological-pathological concordance, lesions
that could not be sufficiently localized during biopsy attempts, as well as scars. In these
instances, an additional imaging-based diagnostic test would be most welcome. Microcalcifica-
tions are considered less problematic, since these lesions can be visualized by mammography
and the workup of these lesions either by biopsy or follow-up imaging does usually not require
additional imaging modalities.
MRI is considered by most to be one of the most sensitive imaging modalities for the detec-
tion of breast cancer [
5,6]. Thus, breast MRI has for instance been shown to be a good imaging
modality to exclude advanced nodal disease and to be helpful in the differential diagnosis of
architectural distortions [
7,8]. However, the effectiveness of breast MRI as a problem-solving
tool remains controversial. Regular concerns are that the high sensitivity of MRI may not be
high enough to rule-out malignancy and that MRI may associate with a high number of false
positive findings as reflected by a low positive predictive value. As a result, there currently
is a lack of clear recommendations on the application of MRI to resolving breast-imaging
equivocality.
Consequently, the purpose of this systematic review and meta-analysis was to evaluate the
performance of breast MRI for diagnosis of breast cancer in non-calcified equivocal breast
findings. Or, put otherwise: can breast MRI rule-in or rule-out malignancy in non-calcified
equivocal breast findings?
Materials and Methods
Search strategy
Two authors (BBB, PAB), one of them with 13 years of clinical experience in breast MRI, inde-
pendently performed a systematic query of all full-text articles in the openly accessible PubMed
database from 01/01/1986 up to 06/15/2015 (
www.ncbi.nlm.nih.gov/pubmed/). Search terms
were predefined as “breast MRI BI-RADS 0”, “breast MRI BI-RADS 3”, “breast MRI problem-
solving” and “breast MRI equivocal”. A separate search was performed for each search term
combination as indicated by the quotation marks above. Resulting titles/abstracts were ana-
lyzed for eligibility and full texts were retrieved, accordingly. Since no specific MeSH terms for
this systematic literature study were identified, additional results were obtained by backward
MRI of Non-Calcified Equivocal Breast Lesions
PLOS ONE | DOI:10.1371/journal.pone.0160346 August 2, 2016 2/16
snowballing [9]. Results at every step were compared and discrepancies solved in a consensus
review. If no consensus was reached, a third reader (NBB) served as an arbitrator.
Eligibility criteria for study selection
Eligible were peer-reviewed studies applying the index test, dynamic contrast-enhanced breast
MRI, as an adjunct to conventional imaging (mammography, ultrasound) to clarify unequivo-
cal findings without microcalcifications in at least 20 human subjects. The reference standard
for index test findings was defined as presence of histopathological sampling or imagin g fol-
low-up of at least 12 months. A diagnosis of cancer by the reference standard was considered a
positive finding, absence of cancer as a negative finding. No language restrictions were applied.
Data extraction and quality assessment
Two authors (BBB, PAB) independently extracted the following data: publication year, study
design (retrospective/prospective), patient number and demographics, whether patient recruit-
ment was consecutive, and indications for MRI examinations. Moreove r, technical MRI
parameters (field strength, coil, contrast medium dosage, and whether fat saturation was
applied) were also collected. Index test (breast MRI) and reference standard (histopathology,
follow-up) data were retrieved to fit a cross-tabulation with true or false positives (TP, FP) and
negatives (TN, FN). Imaging results were called positive in case of BI-RADS 4 or 5 and negative
in case of BI-RADS 1, 2 or 3. In case only a subpopulation of a study fulfilled the eligibility cri-
teria, these specific data were extracted. Quality of studies and likelihood of bias were indepen-
dently (BBB, PAB) evaluated, using Quality Assessment of Diagnostic Accuracy Assessment
(QUADAS-2), an applet that assesses risk and bias in patient spectrum, reference standard, dis-
ease progression, verification, clinical review, incorporation, test execution, study withdrawals,
and indeterminate results [
10]. Any disparities in the findings were resolved by consensus. If
no consensus was reached, a third reader (NBB) served as an arbitrator.
Study outcome
Diagnostic parameters of breast MRI were defined as the study outcome. These were: sensitiv-
ity (TP/(TP+FN)), specificity (TN/(TN+FP)), positive predictive value, PPV (TP/(TP+FP))
and negative predictive value, NPV (TN/(TN+FN)). The influence of a series of covariates was
determined by subgroup analysis.
Statistical analysis
Analyses were performed using STATA 13.0 (StataCorp, College Station, TX, USA) and Open-
MetaAnalyst 12.11.14 (
http://www.cebm.brown.edu/open_meta/download). The possible pres-
ence of publication bias was further assessed by use of Begg's funnel scatterplot and calculated
by Egger's test. In this setup, the plot allowed us to probe both for bias and systematic heteroge-
neity as a function of study size, whereby a symmetrical inverted funnel shape denotes a largely
unbiased dataset [
11].
Once our concerns of data bias were addressed, we used data from individual cross-tabula-
tions to construct forest plots for sensitivity, specificity, positive predictive value (PPV), and
negative predictive value (NPV). Data heterogeneity was analyzed using Cochran´s Q and I-
squared statistics. We calculated to that effect Q as the sum of squared differences between the
effects of individual studies and those pooled across all studies. Since Q can be affected by sam-
pling size, we also calculated I
2
, the latter being thought to represen t the percent of variability
that it is relatively independent from sampling errors, and defined as I
2
¼
Qdf
Q
x 100%,
MRI of Non-Calcified Equivocal Breast Lesions
PLOS ONE | DOI:10.1371/journal.pone.0160346 August 2, 2016 3/16
wherein df represents Q's degrees of freedom [12]. Pooled estimates for breast cancer preva-
lence, sensitivity, specificity, PPV, NPV, and likelihood ratios were calculated by applying ran-
dom effects models or bivariate analyses using maximum likelihood estimates, as appropriate.
In this study setting, the likelihood ratio can be used to calculate the post-test odds from the
pre-test odds of breast cancer. In addition, a bivariate summary Receiver Operating Character-
istics (sROC) curve was calculated and meta-regression using random effects models was used
to investigate sources of heterogeneity. To put the results into context with Bayes´ theorem, a
Fagan´s nomogram and probability-modifying plots for positive and negative breast MRI
results were constructed.
Results
Study design and reporting
Out of 514 peer-reviewed studies, we identified fourteen eligible reports comprising 2,316 cases
[
13–26]. In five of these fourteen studies, we extracted only data of the eligible subgroups
[
17,18,21,22,26]. This meta-analysis adheres to the Preferred Reporting Items for Systematic
Reviews and Meta-Analyses, and
Fig 1 depict s a PRISMA flow chart summarizing the selection
process. The completed PRISMA checklist is given in
S1 Table. Three of the included fourteen
studies were prospective [
14,20,26]. The remaining studies focused on retrospectively review-
ing the diagnostic accuracy of breast MRI in a problem-solving population. Indications for
MRI were described as required for further evaluation of unclear or suspicious findings, with
or without quest ionable correlates in conventional imaging. Studies included primarily unclear
BI-RADS 0 or 3 finding analogs, whereas some studies also included lesions labeled as BI-R-
ADS 4. In the latter, however, the lesions could not be localized for biopsy and as such qualify
as equivocal lesions [13,18,23]. Four studies indicated detailed reasons for referral to breast
MRI (e.g. the exact type of clinical or imaging findings) [
15,23,25,26], two of which stratified
breast MRI results by indication [
25,26]. The remaining ten studies did not specify detailed
indications for breast MRI referral [
13,16–22,24].
Nineteen studies were deemed not eligible due to various reasons, including lack of peer
review, lack of extractable raw data, studies with too small of a sample-size (n<20), those
focusing on the exclusion of malignancy in biopsy-proven lesions with uncertain malignant
potential (B3) and recommended open surgery, those only aiming at diagnosing suspicious
mammographic microcalcifications, or those limited to certain risk groups or specific clinical
questions [27–45](Fig 1).
All MRI tech nical data are summarized in
S2 Table. QUADAS-2 analysis of risk of bias and
applicability assessment identified a likely patient selection related bias in four studies: one
recruited patients conditional on palpable findings [
22], one on lesion size [20], and two con-
sidered solitary conventional BI-RADS categories only [
14,26]. An unclear patient selection
bias was assigned in another study due to restriction to two BI-RADS categories [
17], and an
unclear bias was also assigned in two studies since a proportion of the patients were lost to
imaging follow-up [
19,24
]. Regarding applicability of the analyzed studies to the research ques-
tion, bias was deemed low in the majority of studies. An assignment of unclear applicability
bias was assigned to the two studies with patients lost to follow-up [
19,23], and to one study
investigating only BI-RADS 4 findings [
26]. Detailed QUADAS-2 assessment results are given
in
S1 Fig.
Finally, a Begg's funnel plot analysis revealed a symmetrical distribution, indicating a lack of
publication bias (Fig 2). The complimentary Egger's test also did not show a significant risk of
publication bias (p = 0.12).
MRI of Non-Calcified Equivocal Breast Lesions
PLOS ONE | DOI:10.1371/journal.pone.0160346 August 2, 2016 4/16
Prevalence of malignancy and diagnostic performance of breast MRI
Analysis of prevalence showed substantial heterogeneity ranging from 1.8 to 56.7% (Q 162.2, I
2
92%; p<0.0001, Table 1). Using a random effects model, pooled prevalence was 14.3% (95%
CI: 9.8–18.8%). The number of true positives, false positives, true negatives, and false negatives
in each study are listed in
Table 2. A bivariate analysis sROC curve revealed an area under the
curve (AUC) of 96% (95%-CI 94–98%,
S2 Fig). We then run bivariate analyses of breast MRI-
Fig 1. Flowchart depicting the selection process during systematic literature review.
doi:10.1371/journal.pone.0160346.g001
MRI of Non-Calcified Equivocal Breast Lesions
PLOS ONE | DOI:10.1371/journal.pone.0160346 August 2, 2016 5/16