The BIRADS 3 and 4 categories are the gray zone of lesion characterization. The former can cause confusion to the interpreter as well as patient anxiety [15]. The current BIRADS lexicon offers significant statistical difference regarding the classification of BIRADS 4 lesion that ranges from greater than 2% to less than 95% malignant [16]. In practical terms, this means that there is a high percent chance that a biopsy can yield a benign result and in retrospect be deemed unnecessary [17].
Our study encompassed lesions that were classified into the BIRADS 3 and BIRADS 4 lesions and aimed to highlight the role of further MRI with emphasis on DW-MRI.
It was noted that most benign lesions we encountered tended to occur in the younger age group while the incidence of malignancy increased with age which is in accordance with what Alawi et al. stated [18]. We both also concluded that they comprise a wide range of pathology.
A respective sonomammographic sensitivity and specificity of 73% and 80% were achieved in our study which is much higher than the overall reported sensitivity of 30–60 and 40–80 stated by Mehnati et al. [5]. However, it is comparative with Eisa et al. who reported a sensitivity of 68 and specificity of 74% [19].
Mammography remains the mainstay for diagnosis of breast cancer; however, its sensitivity decreases with increasing breast density owing to the superimposition of glandular tissue [20]. This was the case in our study as the false negatives deduced in our study were seen in dense breasts categorized as ACR C and D. Complementary ultrasound can increase the mammographic sensitivity and thus the number of detected cancers [21].
Based on its ability to assess both morphological and enhancement patterns, MRI has been known to act as a problem solver [22]. We found that the morphological characteristic of a lesion is an important discriminator (sensitivity 92.7% and specificity 89.8%), and this was comparable to the results of Seifeldein et al. who achieved a sensitivity of 97.5% and a specificity of 85% [23]. According to and in agreement with Ebrahim et al., the spiculated margin seen in malignant lesions is a cornerstone when trying to evaluate a pathology [24]. In our study, we were faced with an incidentally discovered lesion in an older patient during her first screening sonomammogram that appeared rather well defined with the exception of mild lobulation to one of its borders. On MRI, the combined analysis of its margins and the lack of any enhancement allowed us to overcome the need of further biopsy. Further follow-up confirmed the stable benign nature of the lesion.
The low specificity was highlighted in our study when benign pathologies such as adenosis and fat necrosis demonstrated confusing non-mass enhancement. On the other hand, one of the malignant entities (DCIS) failed to enhance in our study.
Another fundament in the assessment of a lesion detected on MRI that is an important part of the multiparametric MRI approach is the dynamic contrast enhancement. In the current study, we were able to achieve a sensitivity of 90.2% and a specificity of 81.6%. Aribal et al. [25] reported a higher sensitivity of 97%, and Ebrahim et al. [24] were able to achieve a sensitivity of 100%. Our low specificity was in agreement with the above studies which showed a specificity of 88.9% and 76%, respectively. In our study, we allotted no enhancement or type one curve to benign lesions and jointly allotted both type two and three curves to malignant lesions.
Failure to correctly classify a lesion as benign was seen in two cases, papilloma and granulomatous mastitis with both lesions demonstrating a washout curve rather than more benign rising one. Several papers have similarly described papillomas as falsely malignant acquiring this benign characteristic with a range of 55–72.7% [26].
We also had three false-negative results that yielded benign rising curves rather than malignant washout curves when in fact their pathology was IDC (Fig. 3) and DCIS.
By analyzing the morphological and kinetic features of lesions, together we attained a higher sensitivity and lower specificity of 95.1 and 77.6%. This increase in sensitivity when both parameters were combined was also reached by Ebrahim et al. who reported sensitivity of 100% [24].
More false-positive cases (11 cases) were noted on the final MRI assessment (combined) as we considered the most suspicious finding. Two cases (proved to be sclerosing adenosis and atypical ductal hyperplasia) with suspicious non-mass enhancement showed type 1 rising curve. On the other hand, two cases showed benign morphology yet washout curves, and for the remaining seven cases, plateau curves were attained with either benign or suspicious morphology. In those cases, a final BIRADS categorization of 4 was given necessitating the need for further biopsy.
On the other hand, the false-negative cases were just two where the morphology and kinetic analysis revealed benign features (DCIS, IDC triple negative).
The currently employed sequences in breast MRI including the pre- and post-contrast series when read in conjunction with the ACR MRI BIRADS lexicon allows for the high sensitivity but low specificity of breast MRI [27].
The addition of DWI to the protocol can add more information about the lesion enabling its better characterization as they both provide different data about the lesion with the former concerned with the lesion vascularity and the latter with water diffusivity. An added advantage of DWI is that we can overcome the worry of gadolinium-based contrast agent safety [28].
Reading DWI-MRI combines both qualitative approach as well as the quantitative interpretation that is based on the ADC values. The background signal suppression in high b value DWI allowed better lesion detection in one of our cases where a small satellite was better detected on DW-MRI compared to the DCE-MRI (Fig. 4).
When categorizing our results into benign and malignant lesions, we relied on the EUSOBI recommendations which specified a cutoff value of 1.3 × 10−3mm2/s. Accordingly, we were able to achieve a sensitivity of 73%, and specificity of 83.7%. We understand that this specificity was lower when compared to El Bakry et al. [29] and Yadav et al. [30] who reached a respective specificity of 92.1% and 91.6%.
We can attribute this result to the fact that we had a significant number of false-positive benign lesions especially fibroadenomas which appeared restricted owing to their high cellularity (Fig. 5). These false-positive results have been acknowledged by Brandão et al. [31].
We also encountered false-negative results which were evident in DCIS, invasive ductal carcinomas associated with areas of breaking down, and as well as in small lesions. Durur-Subasi states that these scenarios can result in malignant lesions appearing facilitated rather than restricted [32]. Pinker et al. suggested that lesion less than 10–12mm was missed on blind reading [33].
On further retrospective analysis of our results, we found that using ADC value of 1.435 as a cutoff value gave a sensitivity of 100% (no cancers missed) yet a specificity of 75.5%. However, we believed that the ADC value of 1.365×10−3mm was considered the best cutoff point to differentiate benign from malignant lesions with resultant sensitivity (90.2%) and specificity (81.6%). Unfortunately, the different protocols among institutions and the variations in reported ADC values mean that there is no established cutoff value [32]. However, our results were close to the cutoff value suggested by the European Society of Breast Imaging [11].
A significant drawback to DW-MRI which we encountered is its inability to adequately assess the margins of a lesion. This is in accordance with what Chen et al. stated [34] in their study regarding the inferior spatial resolution and partial volume effect attributed to DW images.
A note of interest we came across in our study is a case of a mucinous carcinoma in a high-risk patient. It appeared well defined and acquired homogenous contrast on DCE-MRI. However, it demonstrated a wash out type 3 curve. On DWI-MRI it appeared to be restricted with a corresponding low ADC value. This contraindicates to Amornsiripanitch et al. who wrote that lesions with high water content including mucinous carcinoma display high ADC values [35]. Baltzer et al. [36] and Pinker et al. [33] also report that in up to 67% of the time they were not visible on DWI.
According to our results, the diagnostic performance of DCE-MRI was higher than DW-MRI and that means that it serves as a better option when trying to solve problematic lesions. In contrary, Daimiel et al. concluded that DWI was more specific (86.3% vs 65.7%, p < 0.001) but less sensitive (62.8% vs 90%, p < 0.001) and accurate (71% vs 80.7%, p = 0.003) compared to DCE-MRI for breast cancer detection [37].
One of our limitations was the sample size, which was considered small to calculate a standardized ADC cutoff value. Moreover, although we encountered a wide diversity of pathologies in our study, yet 34.4% of lesions were fibroadenomas which might have affected our final results.
The relatively high sensitivity but low specificity in our study has led us to believe that DW-MRI is a beneficial tool with limitations. Its role can be especially highlighted when there is concern regarding contrast safety or when cost and availability limit the use of enhanced MRI. There still remains work to be done regarding standardizing both a protocol and an ADC value.