False-positive incidental lesions detected on contrast-enhanced breast MRI: clinical and imaging features

To identify demographic and imaging features of MRI-detected enhancing lesions without clinical, ultrasound, and mammographic correlation associated with false-positive outcomes, impacting patient care. A retrospective multi-institutional study of imaging studies and patient’s chart review of consecutive women with MRI-detected enhancing lesions without clinical, mammogram, or ultrasound correlation between January and December 2018, who underwent MRI-guided biopsy. According to the BI-RADS lexicon, lesions’ frequency and imaging features were recorded. The demographic and imaging characteristics variables were correlated with histopathology as the gold standard and an uneventful follow-up of at least one year. Univariate logistic regression analysis was used to explore the correlation between the baseline variables such as age, genetic mutation, family history of breast cancer, personal history of breast cancer, MRI indication, background parenchymal enhancement, and MRI characteristic of the lesion with the false-positive results in main data and subgroup analysis. Two hundred nineteen women (median age 49 years; range 26–85 years) with 219 MRI-detected enhancing lesions that underwent MRI-guided vacuum-assisted biopsy during the study period fulfilled the study criteria and formed the study cohort. Out of 219, 180 lesions (82.2%) yielded benign pathology results, including 137 benign outcomes (76%) and 43 high-risk lesions (24%). Most demographic and imaging characteristics variables did not help to differentiate malignant from benign lesions. The variables that showed statistically significant association with true-positive results in univariate analyses were age (OR 1.05; 95% CI 1.02–1.08; p = 0.0015), irregular mass-lesion shape when compared with oval/round mass lesion (OR 11.2; 95% CI 1.6–78.4; p = 0.015), and clumped and clustered ring of enhancement when compared with homogeneous (OR 3.22, 95% CI 1.40–7.40; p = 0.0058). For participants with mass breast lesion, the hyperintense signal on the T2-weighted sequence (compared to the normal fibroglandular signal) was significantly related to the false-positive result (OR 0.13; 95% CI 0.02–0.76; p = 0.024). Young patients, oval/round mass-lesion shape, and homogeneous pattern of non-mass enhancement showed the strongest association with false-positive results of enhancing lesions depicted by MRI. For participants with mass breast lesion, T2-bright mass lesion showed significant association with false-positive result. It may impact the patient’s management with a suggestion of follow-up rather than interventional procedure when these demographic and imaging parameters are present, consequently decreasing the patient’s anxiety and health care costs.

Nevertheless, breast MRI has frequently been reported to result in a large number of false-positive diagnoses. Among 366 asymptomatic women of average breast cancer risk, Kuhl reported 55.2% had a false-positive biopsy [11]. In principle, all suspicious findings that prompt biopsy but yield non-malignant histology are considered false positives, which specifically in the breast, is caused by a quite heterogeneous group of pathology changes.
In fact, contrary to what is observed on mammography, MRI-detected false-positive lesions are usually associated with proliferative results or high-risk lesions, such as atypical ductal hyperplasia and lobular neoplasia [11], for which most practice guidelines recommend intensified surveillance, preventive surgery, or even chemoprevention [12]. Thus, although their diagnosis may provide valuable information to guide further patient management, such false-positive diagnoses add to the overall cost of screening because they require additional workup by imaging, shortinterval imaging follow-up, or biopsy, may cause physical harm because of additional morbidity associated with biopsy procedures, and may cause emotional harm because they may generate anxiety in the patient [13][14][15][16]. Accordingly, the reported high number of false-positive diagnoses has been a reason for limiting the acceptance of breast MRI as a screening tool [17][18][19].
The review of the literature reveals that there are several articles on sensitivity, specificity, and false-positive rate of breast MRI, but few studies focused on evaluating lesion characteristics associated with false-positive results [20,21]. For example, Batzer et al. [20] showed that non-masslike lesions were more likely associated with false-positive results, while Myers et al. [21] did not show this association. Furthermore, the larger diameter of the lesion despite the lesion type shows association with malignancy in the Myers study [21] but was only confirmed for mass lesions in the Batzer study [20]. As such, these discrepant results highlight the need for additional studies in this setting. Over time, machines and software have also developed, potentially affecting results.
Therefore, this study aims to identify demographic characteristics and imaging findings, including morphological appearance and kinetic pattern of enhancement of the MRIdetected enhancing lesions without clinical, ultrasound, and mammographic correlation related to false-positive results. These findings will potentially help reduce the overall cost associated with MRI additional workup and patient harm.

Patient selection
Consecutive data of patients with MRI-detected enhancing lesions who have undergone MRI-guided percutaneous biopsy in 2018 from three tertiary hospitals: Sinai Health System, University Health Network, and Women's College Hospital, affiliated with the University of Toronto were retrospectively reviewed. The patient's informed consent was waived. Only those who have undergone surgical excision or were followed clinically or by imaging for at least one year from the date of the last imaging examination were included.
Patients with different MRI indications were included in the study. All the included lesions were incidentally detected by MRI, without clinical, mammography, or ultrasound correlation. The included lesions were classified as BI-RADS 4 or 5 in contrast-enhanced MRI, which make them candidate for MRI-guided biopsy. Patient's demographics and MRI findings, imaging-guided biopsy results, and if applicable, the pathological outcome after excisional surgery were reviewed.

Imaging technique
MRI was performed with standard-of-care technique following the American College of Radiology (ACR) quality standards [22]. In all cases, MRI examinations were performed on a 1.5-T system (Signa Excite, GE Medical Systems) or Espree or Avanto (Siemens Healthcare) and 3.0-T system (Verio; Siemens Healthcare) with a standard, bilateral, dedicated breast coil (Sentinelle Vanguard; Sentinelle Medical, Inc.). The sequences included pre-contrast axial T1-and T2-weighted images with fat suppression and dynamic contrast-enhanced (DCE) T1-weighted imaging sequences. The DCE sequence consisted of a pre-contrast scan and four post-contrast scans. MRI examinations for premenopausal patients were scheduled in the second week of the menstrual cycle to minimize enhancement of benign breast parenchyma (22).

Imaging interpretation
All MRI data were reviewed by two breast fellowship-trained radiologists (AA, 12 years of experience, VF, 20 years of experience) from the same department who were blinded to the surgical or core needle biopsy pathology results and the clinical outcomes. The readers described MRI findings following the Breast Imaging Reporting and Data System (BI-RADS) lexicon [22,23]. Discrepant results of lesion characteristics between radiologists were assessed by consensus. The imaging studies with final assessment category BI-RADS 4 and BI-RADS 5 were considered positive, and all other results were considered negative.
Demographic data were recorded, including age, personal history of breast cancer, risk factors including genetic mutation (BRCA), and family history of breast cancer. In addition, the MRI indication (screening; staging; surveillance; problem-solving; others) and imaging findings were retrieved (background parenchymal enhancement; T1 signal, T2 signal, mass or non-mass enhancement (NME); shape and margin for masses; pattern and distribution for non-mass enhancement; and dynamic pattern of enhancement).

Interventional procedures and pathology
Percutaneous MRI-guided biopsies were performed by attending breast imaging fellowship-trained radiologist at the same hospital within 3 weeks from the enhanced MR, using a 9-gauge MRI-compatible vacuum-assisted device (ATEC, Suros Surgical systems), with 8 to 12 samples obtained.

Histopathology
The histopathologic result was considered the reference standard for lesion evaluation. All biopsied lesions were subjected to rad-path correlation. Invasive cancer and ductal carcinoma in situ (DCIS) were considered malignant histopathological results. Papillary lesions, complex sclerosing lesions, radial scar (RS), atypical ductal hyperplasia (ADH), atypical lobular hyperplasia (ALH), and lobular carcinoma in situ (LCIS) were categorized as high-risk lesions (HRL). Other pathologies were considered benign (fibrocystic changes, Pseudoangiomatous stromal hyperplasia, columnar cell changes or hyperplasia, focal fibroadenomatoid changes, stromal fibrosis). Patients with malignancy had surgical excision as per standard institutional practice. All included women who did not undergo excisional surgery had a minimum of 12-month follow-up. The standard followup was a clinical exam, mammogram alone, or mammogram supplemented by MRI.

Statistical methods
Summary statistics such as means, medians, standard deviations, ranges, frequencies, and proportions were reported to describe participant demographic and clinical characteristics. All biopsied lesions which yielded non-malignant histology were considered false positives. Wilcoxon rank sum test or Chi-square test was used to compare participant characteristics across pathology groups. Univariate logistic regression model was used to evaluate the correlation between participant characteristics with true-positive result for the main dataset and subgroup analysis. Due to limited number of true positives, multivariate model could allow limited predictive characteristics based on the rule of at least 10 events per variable. Thus, we focused on univariate subgroup analysis instead of multivariate analysis. Subgroup analysis was conducted for participants with mass and nonmass lesions, respectively. All statistical analyses were performed using R, version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria). p values less than 0.05 were considered to indicate statistically significant differences.
Out of 219 cases, 65 (30%) underwent surgical excision, including 39 malignant cases and 26 high-risk lesions. The remaining cases were followed up clinically/by imaging.
Out of 43 high-risk lesions, eight (18.6%) patients had intraductal papilloma or papillary lesion; all had followup MRI within 1 year and no significant change was seen. Three patients (7%) had papillary lesions with atypia; 2 of them had lumpectomy and 1 had bilateral mastectomy because of having contralateral breast cancer. From these three patients one had same final pathology and two downgraded. We had 20 patients (46.5%) with ADH or ALH with ADH. Out of these 20 patients, 5 had mastectomy because of having a breast cancer somewhere else in their breasts or being mutation positive, 14 underwent lumpectomy, and 1 chose to have only follow-up. Two downgrade pathologies to fibrocystic changes and 2 upgrade to DCIS and invasive ductal cancer were seen; other had the same final pathology. Among high-risk patients, 10 patients (23.3%) had ALH. 2 of them underwent lumpectomy and 8 had follow-up imaging. One in the follow-up showed an  (7) 9 (8) 2 (5) 2 (7) Regional 16 (9) 12 (11) 0 (0) 4 (13) Multiple regional 1 (1) invasive lobular cancer somewhere else in the same breast; 1 lumpectomy case upgrade to DCIS; and 1 case vanished on the follow-up images. A spindle cell lesion with atypia (2%) was among high-risk patients, who underwent lumpectomy and the final pathology was benign. The only radial scar (2%) that we had chosen to have bilateral mastectomy as she had cancer in contralateral breast. The final pathology of radial scar remained the same.
The false-positive result in our study was 82.2%. Figures 2, 3, and 4 show examples of false-positive cases. Table 2 summarizes the univariate analysis's demographic and imaging features associated with malignant outcome. In addition, subgroup analysis for mass lesions and non-mass lesions were performed (Tables 3 and 4).
The univariate analysis showed that between a variety of clinical and pathological parameters, age in years was the only demographic characteristic that showed a significant influence on the differentiation of false-positive and true-positive groups. Older women had more chance to have a true-positive lesion (OR 1.05; 95% CI 1.02-1.08; p = 0.0015). For masses, compared with oval/round shape, irregular shape significantly increased the occurrence chance of true positive (OR 11.2; 95% CI 1.6-78.4; p = 0.015). In addition, the hyperintense signal on the T2-weighted sequence In the evaluation of non-mass lesions, the only feature that significantly differentiated malignant and benign/ high-risk lesions was enhancement pattern; compared to homogeneous, clumped and clustered ring of enhancement was significantly related to a true-positive result (OR 3.17; 95% CI 1.38-7.28; p = 0.0066). Distribution (p = 0.44), T1 signal (p = 0.77), T2 signal (p = 0.17), and dynamic curve (p = 0.97), did not have significant difference in true-positive and false-positive groups.

Discussion
The malignancy rate of 18% of MRI-detected lesions in our cohort is slightly lower than the BI-RADS lexicon benchmark of 20-50% [24] and aligns with the lower range of previously published studies, ranging from 18 to 60% [25][26][27][28][29][30][31][32][33], suggesting that there is no excessive number of biopsies in our institution. However, the high rate of false-positive results in our cohort, while similar to the previously published study focused on screening [34], warrants knowledge of which lesion characteristics are related to the false positive to minimize the cost of health care and the patient's anxiety. In this sense, young patients, oval/round mass-lesion shape, bright mass lesion on T2-weighted sequences and homogeneous pattern of non-mass enhancement were the identified variables associated with false-positive results.
With respect to demographic characteristics, our results showed that lesions in older women had a greater chance of being true positive than false positive. However, the study by Myers et al. [21] did not show an association of age with malignancy (p = 0.43). This discrepancy can be related to the different populations included in the study. Although MRI indication did not show a statistically significant difference in the false-positive rate in either study, our study was predominantly in high-risk populations, and Myers focused on breast cancer patients undergoing MRI for disease assessment.
Regarding the lesion features for mass-like lesions, contrary to previously published studies [20,21], which showed that margin irregular [20] and spiculated [21] were Fig. 1 Flowchart of the lesions with MRI-guided biopsy parameters associated with malignancy, our study showed that irregular shape and not margin was the feature associated with malignancy. In fact, the difference in the results may be attributed to both, the small number of irregular shape lesions in our study limiting our statistical power, and the subjective analysis of these characteristics, which can cause possible overlapping between irregular margin and shape in the lesion classification. In addition, similar to Myers' study [21], the hyper signal on the T2-weighted sequence (compared to normal fibroglandular signal) was significantly related to false-positive results.
For non-mass lesions, to the best of our knowledge, this is the first time showing that any clumped and clustered ring of enhancement was significantly related to a true-positive result, p < 0.0066.
Moreover, although a smaller number of high-risk lesions (24%, 43/180) were identified in our cohort compared to previously published study (44.8%, 81/202) [11], the rate of a high-risk lesion identified cannot be neglected considering its prognostic importance, which can reflect in different patient's management, including intensified surveillance, preventive surgery, or even chemoprevention. Fig. 2 A 40-year-old woman, high risk. Axial, fat-saturated, contrastenhanced, and subtracted T1-weighted images show a homogeneously enhancing mass in the upper inner quadrant (arrow) b bright on T2-weighted sequence (arrow) c with mixed included washout kinetics on the post-processing color map (arrow). No mammography and ultrasound correlate was detected, and MRI-guided biopsy was performed. Pathology reported fragments of benign breast tissue showing fibrocystic changes, columnar cell changes/hyperplasia, and focal mild to moderate ductal hyperplasia, duct ectasia, and focal chronic inflammation with histiocytic and reactive changes suggestive of reaction to cyst/duct rupture MRI technology has evolved and proton MR spectroscopy and diffusion weighting can also affect false-positive results. However, these techniques require a longer acquisition time and have a substantial number of technical flaws, such as voxel shift, incomplete acquisition, or incorrect shimming, and therefore, their performance was not evaluated in our study [35,36].
Avoiding biopsy of MRI-detected lesions with features related to the false-positive rate can potentially decrease health care cost and patient anxiety. Therefore, it is the main strength of our study.
However, there are several limitations. The main limitation lies in its retrospective design, which may be subject to selection bias. In this sense, the study design, excluding lesions for which biopsy was not recommended (findings that could be true negative or false negative), may have caused over-or underestimation of certain characteristics of the lesion associated with benign or malignant results.
In addition, although the MRI data were reviewed by two breast-trained radiologists and the discrepant results of lesion characteristics among radiologists were assessed by consensus, lesion classification based on subjective parameters defined by the BI-RADS can affect falsepositive rates limiting the generalizability of results. In addition, false positives may be affected depending on the indication of the MRI study; the patient population in this study is broad, consisting of patients undergoing screening MRI, extent of disease, surveillance, and problem-solving and therefore, our results may not be transferable to other settings.
As future steps, we believe that research focusing on functional molecular imaging is likely to be beneficial. Furthermore, recently, several investigators have developed computer-aided methods and machine learning for the diagnosis and quantitative characterization of breast lesions in clinical images. Although most AI methods are still in the technical development and research phase, some studies [37][38][39][40] have shown that artificial intelligence systems can potentially improve the performance of radiologists in differentiating benign and malignant lesions detected on MRI and can be an important promising tool in this setting.

Conclusion
Based on our result, young patients, oval/round mass-lesion shape, and homogeneous pattern of non-mass enhancement showed the strongest association with false-positive results of incidental enhancing lesions depicted by MRI. For participants with mass breast lesion, T2-bright mass-lesion showed significant association with false-positive result. It may impact the patient's management with a suggestion of follow-up rather than interventional procedure when these demographic and imaging parameters are present, consequently decreasing the patient's anxiety and health care costs.    and design. Material preparation, data collection, and analysis were performed by AA, VF, and XL. The first draft of the manuscript was written by AA and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.
Funding The authors declare that no funds, grants, or other supports were received during the preparation of this manuscript.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations
Competing interests The authors have no relevant financial or nonfinancial interests to disclose.
Ethical approval This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of University Health Network (May 14, 2020/REB#20-5119.0).
Consent to participate and consent to publish As a retrospective study, the patient's informed consent and also consent to publish was waived.