Meta-analysis of dynamic contrast enhancement and diffusion-weighted MRI for differentiation of benign from malignant non-mass enhancement breast lesions

Purpose The objective of this study was to conduct a meta-analysis comparing the diagnostic efficacy of models based on diffusion-weighted imaging (DWI)-MRI, dynamic contrast enhancement (DCE)-MRI, and combination models (DCE and DWI) in distinguishing benign from malignant non-mass enhancement (NME) breast lesions. Materials and methods PubMed, Embase, and Cochrane Library were searched, from inception to January 30, 2023, for studies that used DCE or DWI-MRI for the prediction of NME breast cancer patients. A bivariate random-effects model was used to calculate the meta-analytic sensitivity, specificity, and area under the curve (AUC) of the DCE, DWI, and combination models. Subgroup analysis and meta-regression analysis were performed to find the source of heterogeneity. Results Of the 838 articles screened, 18 were eligible for analysis (13 on DCE, five on DWI, and four studies reporting the diagnostic accuracy of both DCE and DWI). The funnel plot showed no publication bias (p > 0.5). The pooled sensitivity and specificity and the AUC of the DCE, DWI, and combination models were 0.58, 0.72, and 0.70, respectively; 0.84, 0.69, and 0.84, respectively; and 0.88, 0.79, 0.90, respectively. The meta-analysis found no evidence of a threshold effect and significant heterogeneity among trials in terms of DCE sensitivity and specificity, as well as DWI specificity alone (I2 > 75%). The meta-regression revealed that different diagnostic criteria contributed to the DCE study’s heterogeneity (p < 0.05). Different reference criteria significantly influenced the heterogeneity of the DWI model (p < 0.05). Subgroup analysis revealed that clustered ring enhancement (CRE) had the highest pooled specificity (0.92) among other DCE features. The apparent diffusion coefficient (ADC) with a mean threshold <1.3 × 10−3 mm2/s had a slightly higher sensitivity of 0.86 compared to 0.82 with an ADC of ≥1.3 × 10−3 mm2/s. Conclusion The combination model (DCE and DWI) outperformed DCE or DWI alone in identifying benign and malignant NME lesions. The DCE-CRE feature was the most specific test for ruling in NME cancers.


Introduction
According to the Breast Imaging-Reporting and Data System (BI-RADS) MRI vocabulary, "non-mass enhancement (NME)" refers to distribution and internal enhancement that do not meet the requirements for a mass after injecting a contrast medium (1,2).However, there is a lack of characteristic manifestations of breast NME lesions, and the overlap of various features of non-mass breast cancer (BCa) with benign breast lesions, like fibrocystic and inflammatory alterations, as well as focal adenosis, can be found in malignant lesions such as lobular carcinoma, diffuse invasive BCa, invasive ductal carcinoma, ductal carcinoma in situ (DCIS), and, on rare occasions, specific forms of BCa (3,4).
The descriptors of BI-RADS for NME are limited to morphological enhancement and distribution patterns (5).It can be difficult to distinguish between benign and malignant NME using breast MRI for BI-RADS diagnosis (6).Because MRI-guided biopsies are rarely used, it is common to perform unnecessary or delayed surgery.
Currently, most institutions prefer dynamic contrast enhancement (DCE) MRI to diagnose NME.DCE can reveal morphological features (focal, linear, and segmental), enhancement characteristics (homogeneous, heterogeneous, clumped, and clustered ring enhancement), and kinetic patterns, which are valuable for distinguishing benign NME lesions from malignant NME tumors (6).
The internal enhancing characteristics of NME lesions provide less information about malignancy than do those of masses.NME is the most common presentation for DCIS and non-palpable invasive malignancies (7,8), despite the prevalence of NME lesions on DCE-MRI being significantly lower than that of masses (76% versus 13%).High-risk lesions, benign disorders such as fibrocystic disease, and hormone alterations have also been connected to NME (9,10).Consequently, NME cancers present a challenge for breast DCE-MRI.
Diffusion-weighted imaging (DWI) is one functional imaging technique that is gaining popularity as a means of increasing accuracy; it may also prove to be a useful tool for the diagnosis and treatment of individuals with NME lesions.DWI is becoming a routine technique for breast MRI, despite the fact that the apparent diffusion coefficient (ADC) value is not included as a classification indicator in the BI-RADS 2013.Several studies have demonstrated the effectiveness of DWI in the detection of NME lesions (11,12).DWI with ADC mapping is now advised in conjunction with DCE-MRI within a clinical multiparametric (mp) MRI strategy (13)(14)(15).It significantly enhances specificity in distinguishing between benign and malignant breast tumors, avoiding unnecessary breast biopsies (11,12,16,17).
Shao et al. investigated the diagnostic utility of the DCE in NME lesions approximately 10 years ago (18).Many studies have now demonstrated the efficacy of DWI in the diagnosis and treatment of NME breast cancers.They discovered that DWI provided similar but superior diagnostic information to DCE and that DWI could be combined with DCE to improve accuracy (11,19).It is suspected that mpMRI and DWI are not as effective for distinguishing benign from malignant NME lesions (14).A detailed systematic review would be useful for analyzing the vast amount of information now available.
Therefore, we conducted a meta-analysis to examine the diagnostic efficacy of models based on the DWI, DCE, and combination models (DCE and DWI) in detecting NME cancer.

Methods
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses-Diagnostic Test Accuracy (PRISMA-DTA) statement (20) was followed in this meta-analysis.The Cochrane Collaboration Diagnostic Test Accuracy Working Group (21) was used to conduct this meta-analysis.

Method for searching the literature
A comprehensive search for published research up to January 30, 2023, was conducted utilizing the PubMed, Embase, and Cochrane Library databases.The search terms were ("Nonmass" OR "Nonmass-like" OR "Non-Mass" OR "non-mass enhancement") AND ("MR imaging" OR "MRI" OR "magnetic resonance imaging" OR "MR") and "breast".The search was limited to original studies written in English or Chinese and published on paper.Furthermore, the references of the included articles were manually searched.

Study selection
Two researchers (L.**.and J.**. with 8 and 6 years of breast MRI experience, respectively) individually assessed the whole texts of any papers that might be qualified after screening the titles and abstracts of the papers that were retrieved.All disagreements regarding potential eligible papers were resolved by a third researcher (L.C.**.with 10 years of experience in breast MRI).
The inclusion criteria were as follows: a) patients: patients with NME lesions; b) index test: breast MRI was used as a diagnostic test for NME lesions; c) comparison: pathological and/or clinical follow-up results; d) outcomes: diagnostic accuracy for differentiating NME benign lesions from cancers; and e) study design: retrospective or prospective trials.We excluded studies without absolute numbers of true positive (TP), false positive (FP), true negative (TN), and false negative (FN) cases; and letters to the editor, case reports, conference abstracts, review articles, meta-analyses, and animal experiments.

Data extraction
The data extracted from each study were the study characteristics (first author; publication year; country; study design; reference standard; sample size; cancer prevalence; percentage of DCIS in all cancers; subtype of malignant lesions; data source; diagnostic criteria; sample inclusion time; TP, FP, FN, and TN numbers), patient characteristics (age, gender, menopausal status, and number of total non-mass-like lesions), and imaging characteristics (MRI pulse sequences, b value, magnetic field strength, position of the patients, and slice thickness).If mean results were reported in the case of multiple reviewers, the results were used for the analyses; if not, the results of the more experienced reader were used (22).The two authors (L.**.and J.**.) extracted the data independently.Inconsistencies between the two researchers were re-evaluated by a third researcher (L.C.**.).

Study quality assessment
Two authors (L.**.and J.**.) independently evaluated the quality of the included studies using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) criteria (23).The final results were based on a consensus discussion.

Statistical analysis
Statistical analyses were performed using R 4.1.2.Meta-analyses were carried out using the bivariate random-effects model (24).Pooled sensitivity and specificity were calculated using 2 × 2 contingency tables.Similarly, the areas under the curve (AUCs) of summary receiver operating characteristic (ROC) curves were estimated.Threshold effects were assessed by determining if the ROC curve followed a "shoulder-arm" distribution.The I 2 statistic, which ranges from 0% to 100%, was used to assess the heterogeneity of study results.An I 2 of more than 75% implies significant heterogeneity between groups (25).
Subgroup analyses and meta-regression were employed to investigate the effects of the heterogeneity-causing factors.Subgroup analyses were performed for studies with different DCE patterns of MR distribution: heterogeneous, clumped, and clustered ring enhancement (CRE), washout, plateau and plateau or washout time-signal intensity curve (TIC), and the different cut-offs of ADC values.

Analysis of publication bias
Visual examination of funnel plots was used to evaluate potential publication bias (26).A p-value of less than 0.05 was considered statistically significant.

Clinical utility
The clinical value was tested using a Fagan plot, which yielded the posttest probability (p post) of NME when the pretest probabilities (p pre, suspicion of NME) for the DCE, DWI alone, and combination models were computed (27).

Quality assessment
The results of the QUADAS-2 assessment are compiled in Figures 2A, B. Of the studies, 11.1% (2/18) in the patient selection domain were scored as "unclear" (12, 34).Two studies did not report consecutive patients.Index test result blinding when interpreting the reference standard was unclear in four studies (13, 16,19,28).Index test result blinding when interpreting the reference standard was unclear in three studies (5,33,35).Three investigations did not report whether the threshold was pre-specified (5,17,28).Because there was inconsistent histopathological analysis used as the reference standard for all included cases, two studies were classified as having unclear risk of flow and timing bias domain (5,35).
For DWI combined with DCE (four datasets in four studies), sensitivity and specificity demonstrated relatively small degrees of variation across individual studies (0.79-0.94 of sensitivity and 0.55-1 of specificity).The meta-analytic summary sensitivity and specificity were 0.88 (0.78, 0.94) and 0.79 (0.63, 0.89), respectively.Figure 3  Heterogeneity testing demonstrated that there was considerable heterogeneity among the studies for DCE and specificity of DWI alone (I 2 > 75%).There was no significant heterogeneity (I 2 < 50%) in DCE combined with DWI approaches.The forest plots of all the studies are shown in Figure 4, along with the I 2 score and pooled estimates for each modality.

Subgroup analysis
Based on the DCE and DWI of malignant and benign NME lesions, we performed a subgroup analysis.A subgroup analysis showed the DCE covariates, including segmental, heterogeneous, clumped, CRE, washout, plateau, and washout/plateau features.Distribution features produced the best diagnostic performance for Methodological quality summary of all included studies by using QUADAS-2 (A, B).

Meta-regression analysis
The DCE and DWI-MRI underwent meta-regression analysis to identify the source of heterogeneity.The meta-regression results are shown in Table 4.For the DCE model, the pooled specificity of studies diagnostic criteria with distribution or TICs 0.74 (0.66-0.83) was higher than that of studies with IEP as diagnostic criteria 0.68 (0.57-0.78).Asians had a higher pooled sensitivity of 0.66 (0.57-0.75) as compared to Caucasians, which had 0.45 (0.33-0.58).
For the DWI MRI, the specificity of studies using histopathology or follow-up as a reference standard with 0.76 (0.64-0.88) was higher than that of studies using histopathology as a reference with a specificity of 0.65 (0.53-0.77).Using the threshold of ADC < 1.3 × 10 −3 mm 2 /s, the sensitivity and specificity of 0.85 (0.77-0.92) and 0.71 (0.59-0.83), respectively, were higher than 0.82 (0.77-0.92) and 0.66 (0.51-0.81), respectively, with ADC ≥ 1.3 × 10 −3 mm 2 /s as cut-off.Likewise, Asians had a higher pooled sensitivity of 0.86 (0.79-0.92) as compared to Caucasians, which had 0.80 (0.71-0.90).As the number of studies is limited, we did not evaluate meta-regression analysis for the combination model.

Publication bias
Deeks' funnel plot (Figure 5) of the DCE-MRI and DWI-MR demonstrated that there was no publication bias (p = 0.75 and 0.94, respectively).

Clinical utility
As illustrated in Figure 6, we computed the posttest probabilities on Fagan plots for the DWI, DCE alone, and DCE combined with DWI models.In the event of a positive pretest, using the DCE combined with DWI model would increase the posttest probability to 51 from 20% with a positive likelihood ratio (PLR) of 4, while in the event of a negative pretest, it would decrease to 4% with a negative likelihood ratio (NLR) of 0.15.

Discussion
In the present meta-analysis, we investigated the diagnostic value of the combination model (DCE and DWI), DCE alone, and DWI alone in NME breast lesions.We demonstrated that combining DCE and DWI may improve sensitivity and specificity when compared to either DCE or DWI alone.We first systematically assessed the ability of DWI to differentiate between malignant and benign NME lesions, and we discovered that DWI outperformed DCE in terms of diagnostic accuracy and had a comparatively high pooled sensitivity.The findings confirmed the possibility of DWI as a shorter and simpler protocol for breast MRI.
DCE-MRI is a vital technique for identifying NME lesions, as well as internal enhancement models and morphologic characteristics, and is the most important parameter (19).We found that DCE-MRI had moderate sensitivity and specificity of 0.58 and 0.72, respectively, which were similar to the 2013 meta-analysis of Shao et al. (18), who reported that the pooled weighted estimates of sensitivity and specificity were 0.5, and 0.8, respectively.One possible explanation for moderate sensitivity may be that there was no association between malignancy and some of the included DCE features, such as clumped pattern enhancement and homogeneity, distribution type, and clumped pattern combinations (31).Likewise, various studies have found no connection between homogeneous patterns and malignancy (37,38).Therefore, we used subgroup analysis to find the valuable DCE features.
The subgroup results revealed that the washout/plateau TICs demonstrated the highest diagnostic sensitivity, while the CRE characteristic had the highest diagnostic specificity.CRE is the main feature of intraductal carcinoma of NME, accounting for 72.2% of all malignant lesions in this study, and five studies were not reported.The pathological basis of CRE in intraductal carcinoma is its abundant blood supply.After enhancement, the matrix around the catheter and the guide tube wall can be improved, and small ring enhancement can be demonstrated when perpendicular to the catheter section (31).When lesions exhibit CRE features, they should be diagnosed as malignant.
Research has shown that TIC is useful in the diagnosis of breast disease (11,13,19,(30)(31)(32).This finding may be attributed to the presence of varying degrees of arteriovenous shunts in breast tissue, as well as capillary hypertrophy and high endothelial permeability in malignant breast lesions (31,32).Our findings indicate that plateau curves or washout TICs are highly sensitive but have low specificity in diagnosing NME BCa.Unfortunately, TICs are semiquantitative analyses, and plateau curves might indicate either malignant or benign lesions, reducing TICs' diagnostic ability.
Furthermore, it is worth noting that the dynamic type is related to the pathological type of breast cancer.Dynamic curves can help distinguish between benign lesions and invasive NME BCa, but they cannot tell the difference between DCIS and benign lesions.Moreover, a washout curve could help differentiate DCIS from invasive NME lesions (31).We were unable to define the pathology type due to the small sample size, and more research is still needed to determine how curve type and pathological type are related.
DWI combined with the quantitative ADC value has been useful in the diagnosis of breast lesions (16,19,30).In this investigation, we discovered that DWI and ADC values can help diagnose malignant NME lesions.It appears to have increased sensitivity to NME BCa.However, we cannot recommend ADC thresholds for NME lesions because the ADC values of included studies ranged from 0.9 to 1.35 × 10 −3 mm 2 /s.Moreover, subgroup results indicated that ADC values with a mean threshold of <1.3 × 10 −3 mm 2 /s had slightly higher sensitivity than those with a threshold of ≥1.3 × 10 −3 mm 2 /s.Similar to our results, previous studies have found an ADC mean threshold of <1.3 × 10 −3 mm 2 /s, which is considered a suspicious diffusion hindrance level by the EUSOBI DWI working group consensus (39).Unfortunately, more research was not conducted to determine the optimal individual threshold of ADC value for differentiating between benign and malignant NME lesions due to the small sample size and variability.
Our findings revealed that the DWI combined with DCE model outperformed the DCE-and DWI-alone models in discriminating between benign and malignant NME.This result highlights the importance of actively integrating DWI with the morphologic and functional data from DCE-MRI.
In our experience, these sequences often complement one another, thereby reducing erroneous readings.Moreover, breast DCE combined with DWI can provide unique morphological, functional, and molecular information about breast tumors, which may significantly improve the BCa diagnostic level.Additionally, DWI is fast and sensitive, but its spatial resolution is poor, and it cannot observe lesions comprehensively (16).Therefore, it is better to combine it with DCE, which has high spatial resolution.
However, the specificity and sensitivity of included studies range from one another, which can be explained by a variety of reasons such as variances in research group size, use of different BI-RADS lexicons resulting in distinct internal enhancement classifications, and changes in evaluation.Heterogeneity is common in meta-analyses.In our study, the different diagnostic criteria significantly influenced heterogeneity in DCE-MRI data.The meta-regression analysis revealed that the DCE model had higher diagnostic sensitivity and specificity in segmental or TIC than in IEP.A previous review reported that IEP clumped pattern enhancement and homogeneity are common in benign lesions (31).As a result, the specificity of studies may be limited.Furthermore, study design can impact research quality and lead to heterogeneity.
In the DWI model, the different reference standard was the source of heterogeneity.There is higher sensitivity in reference to histopathology than histopathology or follow-up, but the specificity was limited.Because the accuracy of the diagnostic test is determined by comparing the results of the index test with those of the reference standard, comprehension of the reference standard may influence the interpretation of the index test findings.However, excluding some benign NME lesions that reference standards based on follow-up, this maybe lead to sample selection bias.In addition, different imaging parameters also create heterogeneity.This is the first meta-analysis with adequate MRI data to evaluate DWI, DCE methods, and the current combination model for assessing NME lesions.Furthermore, Deek's funnel plot demonstrated the absence of published bias, implying that our findings are reliable.However, there are certain limitations.First, there was significant heterogeneity for both sensitivity and specificity.As a result, we conducted subgroup analysis and metaregression analysis to investigate the sources of study heterogeneity, as described above.Second, there may have been interpretation bias because six of the included studies used reference criteria ranging from pathological analysis to surgical or biopsy findings to radiological follow-up.Third, the acquisition of various indicators is dependent on multiple imaging parameters (b value, Tesla), which resulted in some variation between investigations.Furthermore, differences in MR sequence input, ground truth, and other variables may influence the results, although this is considered a minor limitation, as most studies had similar source data.To address these limitations, we recommend further largescale studies to unify the use of b values and DCE parameters in diagnosing NME lesions.In conclusion, our findings showed that the combination model (DCE and DWI) provided extremely good diagnostic performance in distinguishing between malignant and benign NME lesions.DWI has a substantially higher sensitivity for detecting NME lesions than DCE.The DCE-CRE feature was the most specific test for detecting NME cancers.Therefore, further study should combine DWI with DCE to test the accuracy of differentiating cancers from NME benign lesions.

FIGURE 1
FIGURE 1 depicts the hierarchical ROC curves of the region-based DCE-MRI, DWI-MRI, and combination model studies.The sensitivity and specificity ROC curves did not demonstrate a threshold effect.

4
FIGURE 4Forest plots demonstrate the pooled sensitivity and specificity of DCE (A), DWI (B) and combination model (C).
TABLE 1A The baseline characteristics of the included studies.

TABLE 1A Continued
TABLE 1B The baseline characteristics of the included studies.

TABLE 2
Results of pooled estimates and heterogeneity measures for MRI studies of the NME lesions.

TABLE 3
Summary of subgroup analyses: breakdown of malignant and benign NME lesions.

TABLE 4
Univariable meta-regression evaluating the effect of confounding factors on sensitivity and specificity of DCE and ADC.