Non-mass Enhancement in Breast MRI: Characterization with BI-RADS Descriptors and ADC Values

Objectives: The purpose of this study was to assess the accuracy of contrast-enhanced magnetic resonance imaging and diffusion-weighted imaging in distinguishing benign from malignant non-mass-like breast lesions. Methods: 103 lesions showing non-mass-like enhancement in 100 consecutive patients were analyzed. Distribution, internal enhancement patterns, and contrast kinetic curve patterns were classified according to the BI-RADS lexicon. Apparent diffusion coefficient (ADC) values were obtained from manually placed regions of interest (ROIs) on diffusion-weighted images. The optimal ADC value threshold for the distinction between benign and malignant lesions was determined by ROC analysis. Univariate and multivariate analyses were performed to identify independent predictors of malignancy, and the probability of malignancy was calculated for various combinations of findings. Histological diagnosis obtained by means of core needle biopsy was used as gold standard. Results: According to the univariate and multivariate analysis, odds ratios for malignancy were significantly elevated for clumped or clustered ring internal enhancement and low ADC values (p < 0.001), whereas distribution patterns and contrast kinetic patterns were not significantly correlated with benignity or malignancy. In non-mass lesions with homogeneous or heterogeneous internal enhancement and ADC values greater than 1.26×10mm/s, no malignancy was detected, while all other combinations of findings had a probability of malignancy ranging from 22.2 to 76.6%. Conclusions: A combination of BI-RADS descriptors of internal enhancement and ADC values is useful for the differential diagnosis of lesions showing non-mass enhancement. Lesions with homogeneous or heterogeneous enhancement and high ADC can be followed up, while all other lesions should be biopsied.


Introduction
Breast MRI is a highly sensitive method for the early detection and characterization of breast cancer [1][2][3]. However, distinguishing benign from malignant breast lesions based on MR imaging characteristics remains difficult [4,5]. One attempt at improving diagnostic accuracy has been standardization of reading and reporting of breast MR examinations by the introduction of the Breast Imaging Reporting and Data System (BI-RADS) by the American College of Radiology [6,7].
The BI-RADS lexicon distinguishes three types of breast abnormalities: mass lesions, non-mass enhancement, and foci. Mass lesions are defined as space-occupying lesions that are visible on pre-and postcontrast scans, and may displace or otherwise affect the surrounding breast tissue. In contrast, non-mass enhancement (NME) is the contrast enhancement of an area that is not a mass and may extend over small or large regions of the breast, and whose internal enhancement characteristics can be described as a pattern discrete from the normal surrounding breast parenchyma.
The underlying pathology is variable and includes invasive carcinoma, ductal carcinoma in situ (DCIS), high risk lesions such as atypical ductal hyperplasia and lobular carcinoma in situ, and benign mastopathic changes [8]. Malignancy rates of non-mass enhancement have been reported to be in the range of 39-51.7% [9,10]. In a review of non-palpable, mammographically occult invasive cancers detected by MRI, 57% were cancers with non-mass enhancement [11].
Characteristics of non-mass enhancement as described in the BI-RADS lexicon, include distribution (foal, linear, segmental, regional, multiple regions, or diffuse), internal enhancement pattern (homogeneous, heterogeneous, clumped, or clustered ring enhancement), and contrast kinetic curves [6]. However, the value of BI-RADS descriptors for characterization of non-mass enhancement is significantly lower than for mass lesions [10,[12][13][14]. In a metaanalysis of 858 non-mass lesions, pooled estimates of sensitivity and specificity were 50 and 80%, respectively [9]. This diagnostic uncertainty may lead to a significant number of unnecessary biopsies.
Diffusion weighted (DWI) imaging has been shown to be useful for the characterization of breast abnormalities. Although there is only limited experience with DWI in non-mass enhancement, preliminary reports indicate that the addition of apparent diffusion coefficient (ADC) mapping to contrast-enhanced MRI could improve diagnostic accuracy [15][16][17][18]. The purpose of this study was to investigate the performance of BI-RADS descriptors combined with ADC values for the characterization of non-mass enhancement on contrast-enhanced MR imaging of the breast.

Patients
This study was approved by the ethics committee of the Medical University of Innsbruck. Patient informed consent was waived because of the retrospective design. A retrospective search of the MRI database of the Department of Radiology at the University Hospital Innsbruck was performed to identify breast MRI examinations showing nonmass enhancement between June 2016 and June 2020. A total of 123 consecutive cases of non-mass enhancement with histopathological confirmation were identified. DWI was not performed in 14 lesions. Six lesions were excluded because of technical issues with the DWI sequence, including inadequate fat suppression (n= 5) and motion artifacts (n= 1). The remaining 103 non-mass enhancements in 100 women (median age 51 years; range, 21-84 years) were included in the study.
Indications for MR imaging examination included evaluation of indeterminate findings on mammography or ultrasound in 35 patients, high-risk screening in 22 patients, preoperative tumor staging in 17 patients, oncologic follow-up after breast conserving therapy in 15 patients, and clinical abnormalities in 14 patients. Clinical abnormalities included palpable lesions in eight patients and nipple discharge in six patients. The study lesions detected in the patients who underwent MR imaging for staging of otherwise detected breast cancer were additional lesions distinct from the known biopsy-proven cancer.
Histopathologic diagnosis was obtained with MR-guided vacuum-assisted core biopsy in 82 cases, with ultrasound-guided core biopsy in 19 cases and with stereotactic vacuum-assisted core biopsy in two cases. The final diagnosis was obtained by surgical excision in all malignant and high-risk lesions.

MR Image Interpretation
Evaluation of MR images was done on a dedicated workstation (syngo.via, Siemens Healthcare). Post-processing included multi-planar reconstruction of contrast-enhanced images, maximum intensity projections, calculation of parametric images (maximum enhancement integral), and calculation of time-signal intensity curves.
All images were retrospectively viewed by two experienced readers (one having a 10-year experience and another having a 30-year experience in breast imaging) who made their assessment in consensus. Both readers were blinded to all additional imaging results and to clinical, follow-up, and histopathological data. Axial images and reconstructed sagittal and coronal images were used to determine the spatial distribution of enhancement. Enhancement distribution and internal enhancement patterns were classified according to the 5th edition of the BI-RADS MRI lexicon [6].
Construction of time-signal intensity curves and measurement of ADC values was done by one of the two readers. Kinetic curves were obtained from manually placed circular regions of interest (ROI) that covered the areas with the greatest degree of early enhancement. Kinetic curves were classified according to the BI-RADS MRI lexicon. Curves with a persistent delayed phase ware classified as type 1, curves with a plateau delayed phase were classified as type 2, and curves with a washout were classified as type 3. Mean ADC values were measured on the ADC maps by using circular ROIs with a size of 10-20 mm 2 . The ROIs were placed in the areas of lowest ADC-values on visual assessment. At least three measurements were obtained for each lesion and the lowest mean value was selected. Partial volume effects due to surrounding tissue and necrosis were avoided as far as possible by sparing the lesion borders and obvious areas of necrosis, as shown on T2-weighted and contrast-enhanced images.

Statistical Analysis
For patient characteristics, numbers by benign/high-risk/malignant lesions were computed. Positive predictive values were calculated as PPV3 (i.e., the percentage of cancers in cases that underwent biopsy), as recommended by the 5th edition of the ACR BIRAS lexicon [6]. In order to determine a feasible ADC value threshold, a receiver operating characteristic (ROC) analysis was applied and the threshold was defined as the ADC value where sensitivity and specificity crossed. Comparisons of the means of two independent samples were made using the t-test for normally distributed variables and the Mann-Whitney U test for non-normally distributed variables.
For univariate analysis, odds ratios (OR) for malignant lesions compared to benign and high-risk lesions were computed for ADC, distribution of enhancement, internal enhancement, contrast kinetic curve type, tumor size, and presence of a correlate on mammography or ultrasound. We calculated 95% confidence intervals and p values by applying the likelihood ratio test. For multivariate analysis, all variables tested in the univariate analysis with a p value ≤ 0.1 were put in the model. Finally, the prediction probability for malignancy for the combinations of ADC values and internal enhancement was determined.
All statistical analysis was done using Stata Version 13.1 (StataCorp LP, 4905 Lakeway Drive College Station, TX 77845, USA). Statistical significance was established at alpha = 0.05 (except for building the multivariate model).

Results
Pathohistological findings are summarized in Table 1. Of the 103 lesions in our study, 19 were invasive carcinomas, 21 were ductal carcinoma in situ, six were high-risk lesions (confirmed on surgical excision, without upgrade), and 57 were benign. The overall prevalence of malignancies was 38.8%. The mean diameter of benign and high-risk lesions was 31 mm (range, 6-102 mm), and the mean diameter of malignant lesions was 29 mm (range, 7-70 mm; p > 0.05).
In three benign and seven malignant lesions, microcalcifications were visible on mammography in the localization corresponding to the non-mass enhancement. In 13 benign and eight malignant lesions, an ultrasound correlate was detected either on routine sonography (n=4) or on targeted second-look sonography (n=17). The ultrasound findings were mass lesions (n=16), clustered microcysts (n=2), duct changes (n=2), and architectural distortion (n=1). MR imaging characteristics in the 103 non-mass-like lesions are summarized in Table 2. The most common distribution pattern was linear (40.8%; 42/103), followed by focal (29.1%; 30/103) and regional (16.5%; 17/103). The most common internal enhancement pattern was clumped (42.7%; 44/103), followed by homogeneous (34.0 %; 35/103) and heterogeneous (17.5%; 18/103). Clustered ring enhancement was present in six cases, including two lobular invasive carcinomas, one atypical ductal hyperplasia, one lobular carcinoma in situ and two benign fibrocystic changes. Of 21 DCIS, 12 showed a linear clumped enhancement, three a segmental clumped enhancement, two a regional clumped enhancement, two a focal heterogeneous enhancement, one a linear homogeneous enhancement, and one a focal homogeneous enhancement. Of 19 invasive carcinomas, seven showed a linear clumped enhancement, five a focal clumped enhancement, two a segmental clumped enhancement, two a focal heterogeneous enhancement, and one each a regional clumped enhancement, a segmental enhancement with clustered ring enhancement and a regional enhancement with clustered ring enhancement.
Of the distribution patterns, the highest positive predictive value was found in linear distribution (PPV3, 47.6%), followed by segmental distribution (PPV3, 46.2%). The internal enhancement patterns with the highest positive predictive values were clumped enhancement (PPV3, 72.7%) and clustered ring enhancement (PPV3, 33.3%). The positive predictive value for detection of malignancy was 31.9% (23/72) for a contrast kinetic curve showing a persistent delayed phase, 48% (12/25) for a curve showing a plateau phase, and 83.3% (5/6) for a curve showing a washout delayed phase. The six cases with a washout curve were three invasive carcinomas, two high-grade DCIS and one papilloma.

Figure 3. Results of the ROC-based positive test at different ADC values. The optimum cut-off level is determined to be 1.265×10 -3 mm 2 /s
For univariate and multivariate analysis of MR imaging characteristics, linear and segmental enhancement, and regional and diffuse enhancement were combined in one group. In addition, clustered ring enhancement was combined with clumped enhancement. For the assessment of the influence of lesion size, lesions were divided into those under 25 mm and over 25 mm. The presence of a correlate on mammography or ultrasound was also included in the analysis. According to the univariate analysis, odds ratios were significantly elevated (p < 0.05) for clumped or clustered ring enhancement, type 3 contrast kinetic curves, the presence of microcalcifications on mammography, and ADC values less than 1.26×10 -3 mm 2 /s ( Table 3). Multivariate analysis confirmed clumped internal enhancement and low ADC values as independent indicators of malignancy (Table 4).  The prediction probabilities for malignancy for the combination of ADC values and internal enhancement patterns are summarized in Table 5. Non-mass-like lesions with homogeneous or heterogeneous enhancement and high ADC values were all benign. All other combinations had higher probabilities of malignancy, ranging from 22.2% for the combination of homogeneous enhancement and low ADC to 76.6% for the combination of clumped or clustered ring enhancement and low ADC.

Discussion
The results of our study indicate that internal enhancement patterns combined with ADC values are best suited for distinguishing benign from malignant non-mass enhancement. Among BIRADS descriptors for distribution of enhancement, we found the highest positive predictive values for segmental (46.2%) and linear (47.6%) enhancement. This is comparable to the PPV for segmental enhancement of 57.4% found by Wilhelm et al. [10] and 67% found by Liberman et al. [19]. Tozaki and Fukuda [20] reported a PPV of 100% for segmental distribution, which they attributed to high-resolution imaging with multiplanar reconstructions. Reported positive predictive values for linear enhancement ranged from 31.5 to 39.7% [10,21]. Some authors concluded that BIRADS distribution descriptors, together with other imaging features, could be used for predicting the probability of malignancy [22,23]. However, in our study, distribution patterns were not identified as independent predictors of malignancy. Of 13 cases with segmental enhancement, three were DCIS, three were invasive carcinomas, one was atypical ductal hyperplasia, and six were benign lesions. Of 42 cases with linear enhancement, 13 were DCIS, seven were invasive carcinomas, one was atypical ductal hyperplasia, and 21 were benign lesions. The benign lesions were mainly abnormalities associated with the breast ducts, such as sclerosing adenosis (n= 4), papillomatosis (n= 3), ductectasia (n= 4), or galactophoritis (n= 5). In five cases, pathohistology showed fibrocystic changes or normal breast tissue. These results suggest that segmental or linear enhancement do not indicate malignancy, but only an abnormality associated with the galactophoric ducts.
In our study, clumped internal enhancement was a strong and independent predictor of malignancy. This is consistent with other studies, which found high positive predictive values for clumped enhancement alone or in combination with ductal or segmental enhancement [10,[19][20][21][22]. There are few studies which performed regression analysis in order to adjust for confounding factors in the statistical analysis and to identify independent predictors of malignancy in non-mass enhancement. Yabuuchi et al. [18] found segmental distribution, low ADC values and clumped internal enhancement to be the strongest indicators of malignancy. In contrast, Gutierrez et al. [14] reported that BIRADS descriptors were not significant predictors of malignancy in non-mass enhancement lesions. However, for the analysis, clumped enhancement and heterogeneous, reticular or dendritic enhancement were grouped together, making it difficult to compare their results with other studies.
Unlike in previous studies, clustered ring enhancement was a relatively rare finding in our study. Only in six of 103 lesions, including two lobular invasive carcinomas, two high-risk lesions and two benign lesions, was this pattern clearly detectable. Tozaki et al. [24] described clustered ring enhancement in 63 % of malignant and 4% of benign lesions. Uematsu and Kasami [22] reported this finding as being present in 66 of 124 (53%) cases of non-mass enhancement. However, these studies analyzed clustered ring enhancement not as a separate internal enhancement pattern, as specified in the 5th edition of the ACR BI-RADS atlas, but categorically as present or not present [6]. A recent analysis of non-mass enhancement patterns using the updated BI-RADS atlas reported a frequency of clustered ring enhancement of only 20% [21]. In addition, clustered ring enhancement can be difficult to differentiate from other internal enhancement patterns, such as heterogeneous or clumped enhancement. Data on the inter-observer agreement for the diagnosis of clustered ring enhancement are not available [25].
Many studies have assessed the effectiveness of diffusion-weighted imaging (DWI) in the differentiation of benign and malignant breast lesions. A meta-analysis of 13 studies reported an overall sensitivity and specificity of DWI of 84% and 79%, respectively [26]. However, there are only few studies on the effectiveness of DWI in non-mass lesions [16,18]. Partridge et al. [17] and Kuhl et al. [15] found that the mean ADC was significantly lower for malignant than benign lesions for both masses and non-mass enhancement, although the difference in mean ADC of benign and malignant lesions was smaller for non-mass enhancement. In our study, a low ADC was significantly associated with malignancy.
Our ADC cutoff value of 1.26×10 -3 mm 2 /s was similar to that in the study of Yabuuchi et al. [18]. Other studies reported cutoff values ranging from 0.9 to 1.55×10 -3 mm 2 /s [15,17]. The differences may be explained by different scanning parameters and ROI size and placement. To date, there is no consensus about how to choose the ROIs for ADC measurements, although ROI placement and size significantly influence ADC values [27]. Minimum and mean ADC acquired from 2D-ROIs have shown to be best suited for the differentiation of benign and malignant breast lesions [27]. Small ROIs covering the area with the lowest ADC values, as used in our study, proved to perform better than large ROIs covering the whole lesion [28,29].
Non-mass enhancement often leads to unnecessary biopsies. In a BI-RADS descriptor study of 120 malignant and 31 benign lesions, non-mass enhancement was the major cause of false-positive findings [12]. Therefore, the primary objective of our study was to identify combinations of findings that are associated with such a low probability of malignancy that biopsy can be omitted and short-term follow-up can be performed instead. This goal could be achieved by combining internal enhancement patterns with ADC values. To our knowledge, the only study with a similar approach is that of Yabuuchi et al. [18]. However, they examined only 31 malignant and 14 benign lesions and did not use the classification scheme of the 5th edition of the ACR BIRADS lexicon.
Our study has some limitations. First, it is a retrospective analysis from a single institution with a limited number of cases. Second, we used only those lesions that were referred for biopsy, which could have led to selection bias due to the exclusion of benign lesions without histological diagnosis. On the other hand, the proportion of malignant lesions in our study of 37.8% was comparable to other studies [9,10]. Finally, measurements of ADC values were done by only one reader, and inter-observer agreement was not assessed. However, in previous studies, measurements of mean ADC generally showed good inter-and intra-reader agreement, independent of the approach to placement of ROI [27,30,31].

Conclusion
In conclusion, a combination of BI-RADS descriptors of internal enhancement and ADC values is useful for the differential diagnosis of lesions showing non-mass enhancement. Lesions with homogeneous or heterogeneous enhancement and high ADC can be followed up, while all other lesions should be biopsied.

Funding
The authors received no financial support for the research, authorship, and/or publication of this article.

Ethical Approval
The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of the Medical University of Innsbruck.

Data Availability Statement
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to data protection requirements

Conflict of Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.