Diffusion-weighted imaging of the breast—a consensus and mission statement from the EUSOBI International Breast Diffusion-Weighted Imaging working group

To address the above-described issues and support standardized implementation of DWI, promote its clinical use, and facilitate its adoption as an integral part of breast MRI interpretation within BI-RADS, an International Breast DWI Working Group was established by the European Society of Breast Radiology (EUSOBI). Members of the working group were invited based upon proven expertise in breast MRI and/or breast DWI, representing 25 sites from 16 countries. Within the working group, a scientific committee was appointed. The full composition of the working group is given in Appendix 1.

The primary goals of the International Breast DWI working group are:

Members of the working group were first invited to provide information on their protocols with details of acquisition and processing parameters. This revealed substantial heterogeneity of applied DWI protocols, even among experts. The members of the Scientific Committee were then surveyed with a list of points on basic requirements, acquisition parameters, hardware, quality control, analysis and reporting to define the minimum standards for breast DWI, along with technical recommendations to meet these standards.

In this statement, we report the requirements for breast DWI that obtained general consensual agreement. The items that reached more than 80% consensus are shown in the Supplementary Material. This general guidance should be used to provide consistency across clinical trials and in clinical practice.

The group acknowledges that there are many items that may influence both the acquisition and reporting of breast DWI that are not discussed in detail in this statement. Those items did not reach consensus at this stage, requiring further discussion and/or more extensive research. Some of these issues include field strength effects, diffusion gradient waveform dependence, image registration, eddy current distortion and compensation, and gradient nonlinearity correction. Each of these topics has received attention in the breast DWI literature (see below), providing a roadmap for their incorporation in a consensus process. In that spirit, further work is intended from subgroups of our working group to methodically address such items, especially regarding clinical implementation, standardization and quality control, and advanced diffusion MRI methods. The results of these efforts are to be presented in a second round of topical guidelines.

In general, even though breast DWI has high specificity for lesion characterization, the sensitivity of DCE-MRI still exceeds that of breast DWI [41‐45], albeit the sensitivity of DWI alone may be equal to or even higher than that of commonly used screening techniques such as mammography and ultrasound. Thus, currently, its use for cancer detection as part of an unenhanced MRI examination requires further investigation and should be considered only in cases where DCE-MRI is not accessible or not appropriate. In particular, DWI may be a valuable alternative option in patients with contraindications to gadolinium-based contrast agents (such as patients with severe kidney dysfunction at risk of nephrogenic systemic fibrosis and patients with a previous acute reaction to a gadolinium-based contrast agent [44]).

Breast DWI is not helpful in the assessment of breast implant integrity, which relies on other dedicated non-contrast techniques such as T2-weighted imaging and silicone specific sequences and thus is not needed for this purpose. It is, however, as useful as in any other indication for lesion classification in patients with implants.

Breast DWI should be undertaken in a closed bore magnet with field a strength of 1.5 T or more. The gradient hardware should be capable of reaching a maximum gradient strength of at least 30 mT/m, and the use of a dedicated breast coil with at least four channels is strongly recommended. When possible, DWI should be performed before DCE-MRI, as the presence of contrast agents may physically reduce measured ADC values. It should be mentioned, however, that no significant effect has been shown on overall diagnostic performance when DWI is obtained after contrast administration as long as fat suppression using the STIR technique is avoided [5, 9, 45].

The recommended minimal requirements for breast DWI are shown in Table 1. The working group acknowledges that for different MRI systems, different parameters might need to be adapted in order to obtain the best possible results. Therefore, Table 1 should be regarded as a guide rather than as a checklist for optimizing a DWI sequence. For instance, the axial orientation was motivated by anatomy and ease of interpretation more than technical optimization, which depends on the coil configuration. Specifying more parameters is beyond the scope of this paper but will be addressed at a later stage.

Single-shot or multi-shot echo-planar imaging (EPI) is regarded as baseline techniques for DWI acquisition. EPI has a wide breadth of applications and a correspondingly deep library of tools to address its shortcomings in image quality. Examples are nonlinearities in diffusion gradients, eddy current or magnetic field–induced image distortions, and motion. Gradient nonlinearities have been identified as a major source of inaccuracy for breast DWI [46], and prescriptions for correction have been offered [47]. Eddy current and inhomogeneous field distortions are well-known issues for diffusion MRI with evolving solutions. Inhomogeneous field distortion correction [48] and motion correction strategies [49] have been applied to breast DWI. Multi-shot EPI, especially within the RESOLVE sequence framework [50] has the potential to reduce susceptibility-induced geometric distortion and motion artifacts, at the expense of the acquisition time. However, corrective elements for breast DWI acquisitions are currently not sufficiently standardized to recommend a specific approach and will be addressed in future work of our working group. Nonetheless, the group made a recommendation on using an acceleration factor of 2, which reduces eddy current effects (and distortion). Vendor-based standard product pulse sequences with inline workflow elements (eddy current correction, gradient nonlinearity correction) may be used, but the optimal correction strategies are yet unknown.

The choice of b values is critical. Due to the non-Gaussian nature of water diffusion in tissues (which results in a curvature of the DWI signal attenuation plot across b values), the ADC value is highly dependent on the choice of b values, with the ADC values getting smaller as larger b values are used (Fig. 1). Higher b values may increase the specificity of DWI [11] and also lead to a decreased signal-to-noise ratio. A high b value of 800 s/mm² was chosen by the group as a good compromise for standardization. This value can be established theoretically [51] and is backed up by empirical evidence [9].

The number of directions to acquire diffusion measurements from which the final ADC is calculated can vary from one direction to more than 6 directions for diffusion tensor imaging (DTI). Using several different diffusion directions allows calculation of an average ADC and reduces the influence of anisotropy in the tissue, thus improving generalizability. The “3 orthogonal directions” method (commonly available clinically and less time consuming than the 6 directions scheme required for DTI) provides an approximate isotropic weighting (a.k.a. pseudo trace-weighting contrast as an approximation to the true trace which can be obtained from the eigenvalue average derived from the full diffusion tensor). This 3 directions approach may also mitigate residual variations of b values across directions. However, variable eddy current interactions with the EPI readout train may result in different image distortions according to the diffusion-encoding direction, which, when averaged over the 3 directions, can confound image quality and reduce sensitivity to small lesions. Alternatively, acquisitions along only 1 direction but using gradient pulses simultaneously on 2 or 3 axes (sometimes called “diagonal”) allow high b values to be reached using shorter TEs, thus increasing SNR and reducing acquisition time compared with a sequential 3 direction acquisition. Diagonal diffusion gradients, however, would introduce variability in the presence of diffusion anisotropy known to exist in fibroglandular breast tissue. Given these mixed advantages, the compromise recommended is to utilize 3 orthogonal directions. This scheme is universally available and provides approximately isotropic contrast. However, we must emphasize that this approach can introduce non-negligible error and bias in the estimation of diffusion kurtosis imaging-derived indices of the breast and is thus not appropriate for this application [52].

The signal-to-noise ratio (SNR) of DWI is also important, especially when a high b value is used as ADC values may be largely underestimated (mimicking malignant tissues) when the SNR is low (Fig. 2). This requires that a sufficiently high SNR is achieved. One qualitative requirement in that vein is that fibroglandular tissue (FGT) (if present) should be well depicted on low b value images. Quantitatively, while the term “sufficient SNR” has yet to be defined in DWI of the breast, acquisition with multiple excitations for averaging, especially at b = 800 s/mm², will be necessary to reach this goal. Unfortunately, the working group acknowledges that there are currently no universal methods to quantify SNR easily in a clinical setting, especially with the use of multichannel coils. In that setting, prescriptions exist for accurate SNR determination using difference methods [53] for average noise assessment or the Kellman approach [54] for spatially varying SNR, but they require additional acquisitions and advanced reconstruction that are not sufficiently ubiquitous to reach a consensus recommendation. Consequently, before implementing a new DWI sequence in routine clinical care, it is advisable to first test the sequence and assess image quality qualitatively. Ideally, quality control should be performed using dedicated calibrated phantoms with known ADC values, especially for multicenter studies (see below). Such phantoms will also allow the correct application of diffusion gradients (i.e., b values across axes) to be checked. In clinical practice, it remains important to check regularly whether the obtained diffusion images have a sufficient SNR and are reasonably free of structural artefacts that might hamper interpretation. While previous studies have addressed quality controls in DWI in general [55‐57], one study investigated in detail factors influencing SNR and quantitative ADC measures in breast DWI [58]. The results of this study corroborate the hypothesis that scanners should be adequately characterized in breast DWI. Quality control is also required after the maintenance of the MRI system and after each upgrade of hard- and/or software.

Apart from low SNR, a commonly encountered problem is insufficient fat suppression, which leads to ghosting and potentially to underestimation of ADC values. Spectral attenuated inversion recovery (SPAIR), which employs both T1 contrast and spectral selectivity, is recommended given a moderate preference in literature in comparison with short tau inversion recovery (STIR) [59, 60] and based on the consensus among the panel. However, if unsuccessful, STIR may be an acceptable alternative as this technique is independent from B₀ field inhomogeneity [61].

While not listed in Table 1, diffusion time is a parameter that may significantly impact ADC values. Although it is generally hidden to users (gradient pulse time course is usually not reported), with conventional sequences, diffusion time is approximately equal to TE/2. The degree of diffusion restriction (or hindrance) decreases at short diffusion times (increasing the ADC) [62] and it is important to be aware of this parameter as some vendors are now proposing stronger gradient hardware that allows shorter diffusion times. However, on most clinical MRI scanners, achievable diffusion times cannot be shortened below ~ 25 ms, which leads to comparable results between machines.

In a multiparametric breast MRI protocol, lesion detection should be primarily based on evaluation of the contrast-enhanced sequences, which can be aided by inspection of the b = 800 s/mm² images. On the b = 800 s/mm² images, cancers are typically hyperintense given adequate baseline T2 signal. As the basic DWI sequence is T2-weighted, lesions with higher water content (e.g., cysts, myxoid fibroadenoma, highly proliferative cancer) will show a high signal on low b value images and may retain a (relatively) high signal on high b value images. Consequently, a high signal on b = 800 s/mm² images may be due to a very high T2 signal (commonly referred to as T2 shine-through, although physically misleading) or true diffusion restriction with little signal decrease of a moderately high T2 signal. Cross-correlation of b = 800 s/mm² image findings with quantitative ADC maps usually allows discrimination between these instances. Therefore, DWI analysis requires the evaluation of raw DWI data and ADC maps together.

On the other hand, tissues and lesions with a very low water content (e.g., fibrotic parenchyma, scars, some invasive lobular cancers) may demonstrate a very low signal at b = 0 s/mm² (and thus, also on b = 800 s/mm²). Thus, these lesions will be difficult to visualize. In these lesions, the measurement of diffusion-dependent signal loss may not be possible, thus preventing an accurate assessment of diffusion level (Fig. 3). ADC in this condition (that can be referred to as signal blackout) will be low but will not reflect a true diffusion restriction (Fig. 3).

Assessment of location, size, and morphology of lesions is possible on DW images, even though this is limited by a spatial resolution that is inferior to that found in anatomical and contrast-enhanced sequences, and when available, these should be interpreted together. Morphologic assessment on breast DWI may be reported when it is discrepant with other sequences. Lesions can be categorized as foci, masses, or non-mass lesions. For masses, shape (round, oval, irregular) and internal signal pattern (homogeneous, heterogeneous, rim) can be reported while in non-mass lesions, distribution (focal, regional, linear, segmental) and internal signal pattern (homogeneous, heterogeneous) can be reported.

The pixel-wise parametric ADC maps enable quantitative assessment of diffusion restriction or hindrance, which should always be measured. The ADC value is obtained by drawing a region of interest (ROI) on the lesion on the ADC map (or the b = 800 s/mm² image when the workstation allows propagation of the ROI to the ADC map). The ROI should fall completely within the lesion, contain at least 3 voxels and avoid both artifacts and necrotic or hemorrhagic parts of the lesion. This implies that lesions of 6 mm or larger in the axial plane are evaluable with DWI, albeit for small lesions, partial volume effects should be taken into account. While there was no consensus from our group on the size of the ROI to be used (whole lesion or focused ROI), literature suggests that selecting the lowest ADC value within the lesion (potentially reflecting the most active part of the lesion) might provide a more accurate discrimination between malignant and benign breast lesions [17, 29, 63, 64]. Due to the above described signal blackout, the ROI should fall within the enhancing part of the lesion (and hyperintense part on DW images) in order to avoid falsely low ADC values. Using a small ROI in the darkest part of the lesion on ADC maps is analogous to that used for analysis of DCE images. We thus suggest the use of a ROI placed on the darkest part of the ADC map, avoiding necrotic, noisy, or non-enhancing lesion voxels as the preferred method for measuring ADC values in order to reduce inter- and intra-reader variability and improve breast DWI consistency and comparability between sites. For such an ROI, the mean ADC value within the ROI should be reported, and it is suggested that the units should be in 10^-3 mm²/s. Volumetric sampling of the whole lesion may be useful when the clinical indication is the evaluation of tumor response. In any case, the type of ROI (whole or focused) used for lesion assessment should be reported.

Based on ADC measurements and lesion appearance, the proposed classification of diffusion level in lesions is very low, low, intermediate, high, and very high (Table 2 and Fig. 4). Figure 4 shows corresponding ADC ranges based on the most recent meta-analysis of studies evaluating DWI for the differentiation of benign and malignant lesions [10] and considering only those studies that were in line with our consensus b value suggestions. The ADC values provided in Table 2 are solely intended to describe lesions in an objective way according to their diffusion level. Lesion classification should not be based upon diffusion level alone but should always be performed in conjunction with all anatomical and functional information available from all other imaging data.

Of the malignant lesions, invasive ductal and invasive lobular cancers, as well as DCIS with microinvasion, are usually associated with low to very low diffusion levels [16, 18, 65, 66]; pure ductal carcinoma in situ generally shows low or intermediate diffusion (16), whereas particularly invasive mucinous cancer may present with intermediate or even high diffusion levels [67]; and triple-negative cancers with extensive necrosis may also yield high or very high diffusion levels in the necrotic part [68]. Typical examples of benign and malignant lesions are presented in Fig. 5.

ADC measurements are prone to bias from adjacent noise regions (e.g., from voxels containing fat). These can be identified by unrealistically high (> 3 × 10-3 mm²/s), low (< 0.5 × 10-3 mm²/s) or even zero/negative ADC values. Unrealistic ADC values may require repositioning of the ROI.

Two goals of the working group are (a) to promote more widespread integration of DWI into clinical use with standardized acquisitions, increasing the global evidence base by inclusion of DWI in future clinical trials and (b) to derive quantitative discrimination levels supported by standardized and reproducible acquisitions to be incorporated into clinical diagnostic or prognostic guidelines. A practical reality is that these goals are symbiotic and achieving them is an unavoidably iterative process. Thus, the analysis and interpretation scheme based on both qualitative and preliminary quantitative features (ADC value ranges) presented here are preliminary. It is important to realize that these suggestions are not fully vetted by the procedures of quality control, reproducibility, and multisite concordance that underpin formalized quantitative imaging biomarkers (QIB) and thus may evolve with time. This process will also be amplified by the incorporation of experimental optimization (eddy current distortion correction, gradient nonlinearity correction) into the consensus process.

Consequently, the ADC values in Table 2 may be adjusted by our group in the future once more rigorous standardization guidelines are in place, based upon intended multicenter investigation. Similarly, the minimal consensus recommendations of this document may be further refined to minimize variability from protocol variation.