Shear wave elastography for breast masses is highly reproducible

This HIPAA-compliant study was approved by the Ethics Committees/Institutional Review Boards of each of the 16 participating sites in the United States and Europe (France, Germany, Italy and the UK). The study was funded by SuperSonic Imagine (Aix en Provence, France), the manufacturer of the prototype ultrasound system equipped with SWE. All SWE images were provided free of charge to participants. The integrity of the database management was audited by an independent guarantor (Alison Roberts, Qualitus, Strasbourg, France).

Women 21 years of age or older with a breast mass detected by palpation, mammography, ultrasound or MRI, who were referred for breast ultrasound, were invited to participate. We excluded pregnant and lactating women, those with breast implants, women receiving chemotherapy or radiotherapy for any cancer, skin masses and any that had been biopsied, and patients with a history of ipsilateral breast surgery. All participants provided written informed consent. Reference standard of biopsy, fine needle aspiration or at least 1 year of follow-up was available for all masses assessed as BI-RADS® 3 or higher [9]. BI-RADS 2 masses were presumed to be benign. Patient management was based on the standard B-mode ultrasound available at each site (using their “house system”).

All participants underwent diagnostic breast ultrasound on the house breast ultrasound system used in routine practice (with a minimum of 12-MHz peak frequency linear array transducer). All investigators were specialists in breast diagnostic imaging with experience in acquiring and interpreting a minimum of 300 breast ultrasound examinations in the prior 2 years. The participants then underwent an additional B-mode ultrasound examination using the experimental ultrasound system (RUBI, a prototype of the Aixplorer® ultrasound system [Supersonic Imagine, Aix-en-Provence, France]), and B-mode features and BI-RADS assessments were recorded. After identifying the mass of interest using the prototype system, SWE was performed. Three separate images of the mass were acquired in SWE mode. Upon completion, the SWE images were recalled and the user was asked to assess the qualitative similarity of all three images and the quantitative SWE features of the mass were measured. In addition, an expert observer (WAB) performed a blinded review of one set of static images of all masses in B-mode and SWE mode.

The experimental RUBI system was equipped with a 15–4 MHz linear array transducer with default software pre-sets for breast imaging. Initial house and B-mode RUBI images were reviewed. Mass size (longest diameter, perimeter and area), BI-RADS features of shape (oval, round, irregular), orientation (parallel or not) and margin (circumscribed or not) were recorded for each mass, together with a BI-RADS assessment (2, benign; 3, probably benign; 4a, low index of suspicion; 4b, intermediate index of suspicion; 4c, moderate index of suspicion; or 5, highly suggestive of malignancy) [9].

After recording the B-mode features, SWE imaging was performed with the same transducer on the RUBI system [10]. The recommended technique for SWE was to image the mass with no pressure induced by the transducer before switching to SWE mode. After a few seconds of immobilisation to allow the SWE image to stabilise, the SWE image was frozen and saved. At each site, investigators were first trained on phantoms and then on at least 10 non-study patients before enrolling their first participant.

Three consecutive and distinct SWE images of the mass were acquired by the same operator in the same imaging plane, the probe having been removed and re-applied between each acquisition. In each image, the tissue stiffness of each pixel in the SWE image was displayed as a semi-transparent colour overlay with a range from dark blue, indicating the lowest stiffness (at just over 0 m/s), up to red, indicating the highest stiffness (set at 7.7 m/s for this study) (Fig. 1). Areas of black on the SWE images represent tissue in which no shear wave was detected. Using the experimental system’s review capability, the perimeter of the mass (including surrounding tissue stiffness) was traced and the operator determined the largest diameter on each SWE image. Where no zone of altered stiffness was apparent and the mass was homogeneously soft (blue) on SWE, the dimensions recorded were those of the mass seen on the underlying B-mode image.

Each operator visually assessed two features on the colour overlay SWE images: the shape of the mass including the surrounding areas of stiffness (oval, round, irregular), and stiffness homogeneity (not homogeneous, reasonably homogeneous, very homogeneous). The qualitative similarity among the three SWE acquisitions was recorded as: all three images very similar, all three images reasonably similar, some images similar (2 out of 3) or all three images very dissimilar.

In addition to the size measurements, quantitative elasticity was measured on each of the three SWE images using the system’s quantification tool, known as the “Q-Box™”, which defined a 2 mm × 2 mm region of interest (ROI) that was positioned over the stiffest part of the mass or surrounding tissue on the SWE image (Fig. 2). For each of the three acquisitions, an additional Q-Box was placed in the adjacent fatty tissue; the intention was that this be located in breast fatty tissue, though we recognised that this was not always feasible and was not always performed. The system calculated the minimum (Emin), maximum (Emax) and mean (Emean) elasticity values in kPa or m/s within the mass and in the fatty tissue, as well as the ratio between the mean elasticity in the mass and the fatty tissue (Eratio).

After an initial training review of 11 proven cases, a second observer with 20 years’ experience in breast ultrasound (WAB) reviewed the static images of all of the masses. This observer was blinded to the diagnosis, other imaging findings, and the clinical history. One or two representative B-mode images, or a B-mode and Doppler image, of each case were reviewed and the BI-RADS features of the mass shape (oval, round, irregular), orientation (parallel or not), and margins (circumscribed or not), were assessed, followed by an overall BI-RADS assessment (2, 3, 4a, 4b, 4c, 5). A representative SWE image and its corresponding B-mode image were then reviewed; when all three SWE images had been scored as “very similar”, the first was shown to the observer, and when they had been scored as “some images similar” then the first of the two similar images was reviewed. For cases in which images were assessed as being dissimilar, one image was arbitrarily chosen (by J-PH) for observer review. The shape of the mass and surrounding stiffness on the SWE image (oval, round, irregular), together with the homogeneity of the elasticity (very homogeneous, reasonably homogeneous, not homogeneous), were recorded. In addition, the second observer assessed the qualitative maximum elasticity in the masses using a six-level visual colour scale (Fig. 3). Interpretations from the second observer were compared to the initial mass assessment performed by investigators on the house and prototype systems.

From September 2008 until September 2010, 958 women with 1000 breast masses were initially enrolled. When a participant had multiple masses, only the malignant mass was retained. There were 42 cases with multiple malignant or multiple benign masses; in these, one was selected at random. Seventeen cases with a missing reference standard (2 papillomas not excised and 15 BI-RADS 3 masses without follow-up), 1 skin lesion, and 1 case with no mass were excluded. Clinical outcomes with SWE for the remaining 939 masses have been reported [10].

The prototype system did not support measurements above 7.7 m/s; this occurred in 113 masses, which were excluded from the reproducibility analysis because there was no possibility of variance as their Emax exceeded 7.7 m/s. An additional 61 cases were excluded because identical rounded values of Emax were observed for one of two reasons: 1) the same SWE acquisition was used more than once for different measurements or 2) the pre-set maximum SWE scale value was reached and was lower than that specified by the protocol (e.g. 7.3 m/s) because of investigator error. Also excluded were 3 cases in which all values of Emax were identical; 2 with missing values; and 2 cases in which the same image acquisition had been used for more than one set of measurements (i.e. data were not independent). In total, 758 masses were retained. Among the 181 excluded masses were 145 malignancies: we repeated the evaluation of agreement with these masses included (i.e. among 939 total masses, including 289 malignancies).

The data were analysed using the statistical software package Stata/IC, release 11.0 for Windows (StataCorp LP, College Station, TX, USA). For continuous variables, intraclass correlation coefficients (ICC) were calculated as a measure of reliability using a one-way analysis of variance (ANOVA). The ICC is the proportion of the total variability (between masses plus measurement error) explained by variation between masses. A log_e(X + 1) transformation was used for the analysis of all continuous variables to make the distribution more symmetrical; and a 1 (one) was added to each value before taking the logarithm to handle observed values of zero, as the logarithm of zero is undefined.

For ordered categorical variables such as qualitative SWE features, weighted kappa (κ) [11] was used as a measure of agreement using quadratic weights [12]. Weighted kappa is preferable to unweighted kappa for an ordered categorical variable, where disagreements can be weighted according to the extent of the discrepancy. Using the quadratic weighting scheme, weighted kappa is identical to the intraclass correlation coefficient [12]. Weighted kappa is interpreted in the same way as kappa and has a maximum value of 1 (perfect agreement). No weighting was used to assess agreement for individual BI-RADS assessment categories. Kappa of 0–0.2 indicates slight agreement; 0.21–0.4, fair agreement; 0.41–0.6, moderate agreement; 0.61–0.8, substantial agreement; and 0.81–0.99 almost perfect agreement [11]. Kappa for BI-RADS assessments were calculated using six categories: 2, 3, 4a, 4b, 4c and 5. Maximum quantitative elasticity across three acquisitions was used to assess agreement with the second observer’s rating of the visual maximum within six pre-specified colour ranges from one acquisition.

Kruskal–Wallis one-way ANOVA was used to compare independent groups for ordered categorical variables. Spearman rank correlation (ρ) was used to assess correlation of ordered categorical variables.