Introduction

Diffusion-weighted imaging (DWI) is increasingly incorporated into breast MRI protocols worldwide [1,2,3]. DWI using apparent diffusion coefficient (ADC) map** has reported sensitivities of up to 96% and specificities of up to 100% for breast cancer detection [4, 5]. Currently, the prime focus of DWI is to differentiate between benign and malignant lesions to prevent unnecessary breast biopsies. With the recent concerns regarding the safety of gadolinium-based contrast agents (GBCAs) [6,7,8], DWI has been proposed as a promising alternative to dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) to detect early breast cancer without the costs and safety concerns associated with GBCAs [9,10,11,12,13,14].

Several studies have demonstrated that the sensitivity of unenhanced MRI with DWI was equal to or superior to mammography [4, 15]; however, there is still room for improvement [16]. Diffusion sensitivity, better known as “b-value,” has important implications for tumor conspicuity and can be controlled by modifying the magnitude and duration of the diffusion gradients. Higher b-values seem to improve lesion conspicuity by suppressing the normal breast tissue and decreasing the T2 shine-through effect [17]. Nevertheless, they require long examination times and the image quality may be compromised due to a low signal-to-noise ratio [18]. Synthetic b-values may overcome these limitations. Synthetic b-values are generated through a mathematical computation technique from at least two different lower b-values in a voxelwise manner [19,20,21] without increasing the scan time or reducing the image quality (in fact, synthetic b-values present a higher image quality than the acquired b-values) [22] and therefore have the potential to improve the sensitivity of breast cancer detection.

The aim of our study was to assess lesion visibility and the diagnostic performance of DWI for breast cancer detection by the addition of different synthetic b-values.

Materials and methods

Patients

This single-institution study and retrospective data analysis was approved by the Institutional Review Board and was conducted in compliance with the Health Insurance Portability and Accountability Act.

Between September 2018 and March 2019, 84 consecutive women who underwent a breast MRI examination (including DCE-MRI and DWI) at our institution and fulfilled the inclusion criterion of presenting with an enhancing lesion on DCE-MRI (categories 2–5 of the Breast Imaging Reporting and Data System (BI-RADS)) were included in this study. Indications for an MRI examination in these women included screening (46.2%), extent of the disease and surgical planning (33.8%), inconclusive findings in other imaging modalities (6.2%), MRI follow-up examinations for previous findings (5%), evaluation of recurrent tumor (6.3%), and nipple discharge (2.5%). Patients undergoing chemotherapy; pregnant women; and those undergoing examinations without DWI series, a biopsy-proven histology, or at least lesion stability for 24 months were excluded.

Due to technical failure of the DWI sequence and the presence of a clip/biopsy change generating obvious image distortion, four patients were excluded, resulting in a final study population of 80 women (mean age 48.1 ± 12.5 years; range 26–76 years) with 103 breast lesions. Forty-five of these patients were pre-menopausal (56.25%) and 35 were post-menopausal (43.75%).

MRI examination

All the examinations were performed using a 3-T MRI scanner (Discovery MR750; GE Healthcare) with a dedicated 16-channel phased-array breast coil (Sentinelle Coil, Hologic). All the women underwent a state-of-the-art multiparametric MRI (mpMRI) protocol with T2-weighted imaging, DCE-MRI, and DWI. DW images were always acquired before contrast agent injection using a single-shot echo-planar imaging (EPI) sequence with 0 and 800 s/mm2 b-values (Supplemental Table 1, Supplemental Digital Content 1). Synthetic DWI b-values 1000, 1200, 1500, and 1800 s/mm2 were automatically generated from the acquired b-values using a built-in software. Synthetic b-values were selected based on previous literature [23, 24].

Image analysis

Three dedicated breast radiologists (I.D., R.L., and C.S.) with 4–5 years of experience in interpretation of multiparametric breast MRI evaluated images independently using OsiriX v.9.0 software (OsiriX). Readers were aware of the presence of lesions in all the examinations but were blinded to any clinical information and conventional and prior imaging.

DWI

Readers first assessed DW images (b-800, b-1000, b-1200, b-1500, and b-1800 s/mm2) and ADC maps blinded to the DCE-MRI. For all the lesions, visibility using each b-value (yes/no), location, and laterality were recorded. If more than one lesion was visible, all lesions were recorded. A visual grading image quality score (1 = bad quality, 2 = average, 3 = good quality) was assigned by each reader for all the b-values based on artifacts and fat suppression. In addition, a preferred b-value was selected by each reader based on lesion conspicuity defined as the visual difference in lesion contrast with the surrounding parenchyma.

One 2D region of interest (ROI) per lesion and reader was drawn manually on ADC maps derived from acquired b-values using the OsiriX v.9.0 software (OsiriX). The ROI was placed in a slice containing the tumor maximum diameter and within the area with the lowest ADC values.

Each reader assigned a 1–5 malignancy score to DW images (from 1 = non-suspicious to 5 = highly suspicious) using acquired and preferred synthetic b-values for each visible lesion. The criteria for this score included qualitative parameters based upon the previous literature [15, 25] as well as quantitative ADC values extracted from ADC maps as shown in Table 1. Scores 4 and 5 were considered suspicious for malignancy, whereas scores 1, 2, and 3 were considered non-suspicious.

Table 1 Criteria for DWI malignancy score
Full size table

DCE-MRI

After a wash-out period of at least 21 days, DCE-MRI alone was read. Readers classified lesions according to BI-RADS classification [26]. Lesions categorized as BI-RADS 2/3 were considered non-suspicious, whereas categories BI-RADS 4/5 were considered suspicious for malignancy.

Consequently, the results for both readings were reviewed in consensus for missed lesions on DWI or a lesion mis-match between DCE-MRI and DWI. In the case of mis-matched or missed lesions on DWI by one or two of the readers, they were asked to obtain ADC values for lesion categorization. Lesions missed by all the readers were excluded for categorization. The mean ADC values for all the lesions across readers were then determined (Supplemental Table 2, Supplemental Digital Content 1). Categories for breast composition of fibroglandular tissue (FGT) were recorded for each examination based on its report (A-almost entirely fat, B-scattered FGT, C-heterogeneous FGT, and D-extreme FGT).

Multiparametric MRI

mpMRI with DWI and DCE-MRI was evaluated using an ADC cutoff value of 1.3 × 10−3 mm2/s as recommended by the European Society of Breast Imaging [15]. A final lesion classification was given as follows: If a BI-RADS 4 or 5 was assigned on DCE-MRI, an ADC > 1.3 × 10−3 mm2/s was required to assign a final classification as non-suspicious. If a BI-RADS 2 or 3 was assigned, an ADC ≤ 1.3 × 10−3 mm2/s was required to assign a final classification as suspicious.

Histopathology

The final diagnosis was established by histopathology using image-guided needle biopsy for the majority of the lesions (n = 98). In the event of discordant findings between histopathology and imaging, the final diagnosis was established surgically (n = 2). Benignity was confirmed in three lesions by imaging follow-up of up to 24 months.

Statistical analysis

All calculations were performed using SPSS 25.0 (IBM) and SAS 9.4 (SAS Institute) in a per-lesion analysis. Median and mean ranks were calculated for image quality and preferred b-values. Sensitivity, specificity accuracy, and their 95% confidence intervals (CI) were calculated for the imaging methods and averaged over the three readers [27]. Likewise, diagnostic parameters for breast cancer detection were obtained for each imaging modality for lesions stratified by size (small lesions ≤ 10 mm and lesions > 1 mm). Receiver operating curves (ROC) were obtained using the PROC GLIMMIX statement in SAS 9.4 (SAS Institute) by treating each reader’s assessment as a fixed effect and estimating a robust (sandwich) measure of variance to account for the correlation between multiple readers [16, 31, 32]. This is potentially problematic if one of the future roles of DWI is to be a reliable tool in breast cancer detection and not only in the characterization of lesions found in other imaging modalities. An improvement in the resolution of DWI sequence would be desirable to enhance cancer detection. Regarding synthetic b-values, readers were able to identify the same number of cancers using synthetic b-values of 1000/1200 s/mm2 and the acquired b-value of 800 s/mm2. In contrast, synthetic b-values of 1500 and 1800 s/mm2 missed more lesions, probably due to a reduction in image quality.

Most of the studies have almost exclusively focused on the visibility of breast cancer [20, 22, 33,34,35,36]; therefore, there is limited information on the conspicuity of benign breast lesions at high or synthetic b-values. While Chen et al [37] found no significant differences in conspicuity grades using b-values of 600, 800, and 1000 s/mm2, our results point to a difference in conspicuity. Benign lesions were more conspicuous at lower b-values, while malignant tumors appeared brighter than the surrounding parenchyma at high b-values. The increased conspicuity of breast cancer at high b-values has been demonstrated by other studies with a wider range of b-values than Chen et al [17, 38].

An improved conspicuity of malignant tumors at high b-values could be particularly helpful in dense breasts, where lesions can be mammographically masked by the large amount of FGT. In addition, an improvement of tumor visibility without contrast injection could improve the cost-effectiveness of MRI [39]. However, extremely high b-values, i.e., b-1800 s/mm2, have a low signal which can cause lesions located on the fat tissue to be overlooked, especially if fat is poorly suppressed [33]. In light of our results, b-1200 s/mm2 could be the best option for an optimal lesion visualization with the best conspicuity, which could enhance lesion characterization by a better correlation on ADC maps and more accurate ADC values.

Nevertheless, it is worth mentioning that ADC (maps and values) can only be derived from acquired DW images. Synthetic high b-value images are obtained by extrapolating signals acquired at lower b-values (e.g., 0 and 800 s/mm2), assuming a Gaussian model. However, diffusion in tissues is not Gaussian [2]. The calculation of synthetic high b-values is just a strategy to enhance contrast already present in lower b-value images and is potentially useful to detect and depict lesions but lacks the power of non-Gaussian diffusion to characterize tissues [40].

Although synthetic b-values over 1000 s/mm2 have demonstrated an improvement in tumor visualization and image quality [19, 20, 22, 34, 35, 40,41,42,43], DCE-MRI outperforms DWI for breast cancer visualization and detection with a higher sensitivity across all readers. This is in accordance with the current literature: DCE-MRI outperforms unenhanced MRI with or without supportive sequences for cancer visualization [9, 16]. In particular, tumors such as DCIS or NMLE exhibit a lower signal intensity in DWI and, therefore, are prone to be overlooked with unenhanced MRI, especially at high b-values [44]. These limitations are to be addressed to enable unenhanced MRI in a screening setting, where tumors tend to be smaller and NMLE lesions are clinically undetectable. In addition, these types of lesions account for false negative cases in DWI. In our study, a high number of IDC cases exhibited associated DCIS which could explain a slightly lower sensitivity for DWI compared with other studies [4]. Based on our results, DWI alone would currently have no role in the work-up of indeterminate lesions (e.g., BI-RADS IVa and IVb lesions), especially in small ones where its accuracy was lower mainly at the expense of a decrease in sensitivity. In this subgroup, the sensitivity and accuracy for DCE-MRI were also reduced since there is a difficulty in distinguishing morphological features. In these cases, mpMRI continued showing the best accuracy although no significant differences with DCE-MRI were found. Nevertheless, there was an additional value in the combination of DWI and DCE-MRI: a decrease in the number of false positives. This was particularly relevant in the group of small lesions ≤ 10 mm which included most of the benign lesions in our study sample. This is important to prevent unnecessary follow-up examinations in indeterminate lesions as well as benign breast biopsies, which increase costs and patient anxiety.

These results match previous publications investigating a combined DWI and DCE-MRI approach for breast cancer detection [16, 45,46,47].

Overall, inter-reader agreement was moderate to high for all the parameters assessed. Lesion visibility at b-800 s/mm2 achieved the lowest agreement, which could point to a more consistent performance of synthetic b-values for lesion visibility. Inter-reader agreement was moderate for b-values rendering the best tumor conspicuity (1200–1500 s/mm2). This can be explained by the fact that readers preferred a range of b-values rather than a specific value. Images at b-1800 s/mm2 were rated worst by all readers with respect to both lesion conspicuity and image quality.

We acknowledge several limitations in our study. Firstly, no comparison was done with acquired high b-values to maintain clinical acquisition times. Secondly, the larger size of malignant lesions compared with the benign ones and the small number of pure DCIS, ILC, and NMLE compared with invasive carcinomas presenting with a mass may affect the results and their generalization. Nevertheless, this population reflected the clinical practice in our screening and tertiary assessment center under the established inclusion criteria. Thirdly, synthetic b-values generated from different DWI sequences may yield different visual and image quality results. In our study, a single-shot EPI DWI with a short TI inversion-recovery (STIR) fat suppression sequence was used, and therefore, our results may not be extrapolated to other sequences.

In conclusion, the addition of synthetic high b-values (e.g., 1200s/mm2) improves tumor conspicuity without increasing the time of scan, which is particularly helpful in dense breasts. Nevertheless, the role of DWI for the visualization of NMLE and small lesions and its performance in breast cancer detection are still not definite. mpMRI remains the best modality for lesion detection with the best accuracy which is particularly helpful in MRI screening patients and obviates unnecessary biopsies in benign lesions.