Using a combination of superb microvascular imaging and other auxiliary ultrasound techniques to increase the accuracy of gray-scale ultrasound for breast masses

Breast lesions are usually assessed by mammography, ultrasound (US), or MRI, depending on the characteristics of each patient. Breast US can assist in differentiating benign from malignant lesions based on features like shape, echogenicity, margins, or posterior features. These findings sum up in a Breast Imaging-Reporting and Data System (BIRADS) assessment category (according to the American College of Radiology; ACR) assigned by the radiologist which alludes to the amount of suspiciousness [1].

Some auxiliary modalities have been developed as diagnostic adjuncts to increase the accuracy of grayscale US. Strain elastography (STE) estimates the elasticity of the lesion and adds to the specificity of US [2, 3]. Other assisting technologies detect new vessels in the region. Neovascularization is a key event in the development of malignancy; small branches grow from vessels in and around and advance into the emerging neoplasm [4, 5]. Doppler US (DUS) is widely used for investigating the vascularity of breast lesions US [6]. However, it cannot detect vessels less than 0.1 mm, is angle-dependent, and is not capable to differentiate between low-volume blood flow and tissue movements (cluttering) in the breast tissue; eliminating these features while providing the flow image [7, 8]. Power Doppler US (PUS) is more sensitive than DUS for the detection of low-volume, slow blood flow, and is not angle-dependent; it displays more definite criteria than DUS to differentiate malignant from benign lesions [9, 10].

In these modalities, a higher vessel number is a criterion in favor of malignancy. For the distribution of vessels, peripheral vascularity is less suspicious than central vascularity, and their combination is worse [7]. Four types of morphologies have been defined: dot-like, linear, branching, and penetrating, respectively named types 1 to 4; showing an increasing level of suspiciousness [7].

A more recent technique is superb microvascular imaging (SMI), which filters tissue cluttering, and distinguishes the sluggish blood flow in small vessels [11]. The first work about breast SMI was published in 2015 [12], and then several studies investigated its capabilities reporting that SMI shows the number, distribution and morphology of vessels in detail; even more specifically than contrast-enhanced studies [12, 13].

Two very recent systematic reviews have been carried out on this topic [8, 13], suggesting that SMI might be superior to other supplementary modalities in the differentiation of benign and malignant lesions. Nevertheless, their collective sample size was not large enough, and non-biased studies suitable for the meta-analyses were limited to three countries only (China, Korea, and Turkey). Also, studies about the results of the combination of US, SMI and other auxiliary methods are scarce. Therefore, we conducted this study to investigate the role of SMI as a supplementary measure to US, and also find and present a sensitive and specific novel combination model comprising gray-scale US and these auxiliary modalities for evaluation of breast masses.

We carried out a study to examine SMI accuracy and find a combination to better predict the malignant nature of breast masses during US. Our results showed no privilege for SMI alone, but a good predictivity when combining SMI, STE, and US BIRADS.

Breast US has a high SE (100% in our study) but a low SP, which adjunct modalities try to increase [7]. Diao et al. showed a higher accuracy for SMI + US than PUS + US, similar to us [17]. Lee et al. [18] found that the combination of SMI, SR and US upgraded the SP from 46 to 79% while lessening the SE from 98 to 91%. The trend of their findings is similar to ours, but they used VI for SMI; and the sensitivities and specificities were higher in our study. Uysal et al. [19] showed that SR was more accurate than SMI VI. This corresponds with our findings, but we further explored the issue by providing combinations of modalities.

There is no fixed cut-point for the suspicious number of vessels, and there are no quantitative ratings for vascular morphology and distribution; thus no threshold for the vascular findings has been standardized. Park et al. [7] defined the above score to quantify SMI findings, and found higher scores for malignant lesions. Also, they showed that US + SMI was more accurate than US + DUS or US + PUS. However, their score gave equal scores to all findings, whereas our radiologists experience showed that the vascular features do not have equal impacts; and that vessel distribution, number, and morphology have ascending values in favor of malignancy. Therefore, we assigned a coefficient from 1 to 3 to each of these to provide a pertinent weight to that group of findings. We scored morphologic and distributional features as Park et al. [7], and the number of vessels according to our calculated cut-off point of 4 (Fig. 3). Then, we considered findings of DUS, PUS, and SMI as quantitative items and defined the cut-off points. As the accuracy of VS was higher than VR, we considered various combinations of the VRs and SR with US-BIRADS to find the highest accuracy. Considered separately, STE alone had the highest accuracy among the four assisting options (Table 4). When added to US BIRADS, the SP increased to around 79%, but the SE decreased to 90%. The best model was when considering US + STE + SMI. The SE was still 100%, and the SP rose to nearly 75% (Table 4). Thus the combination of the US BIRADS, SR, and SMI lead to a high diagnostic capacity for breast lumps.

Other scoring systems also have quantified vascular findings. Kim et al. [20] assessed 62 lesions (12 malignant) by rating the number of vessels from 0 to 4 (0, 1–2, 3–4, and 5–7 vessels), and the distribution and architecture like Park et al. [7]. Despite the lower sample size, the proportion of malignant to benign masses was close to ours. They found that US + SMI increased the diagnostic accuracy from 66 to 90%, while these were 56% and 70% in our study, respectively. However, we developed these findings by the scorings and the triple combinations, and obtained higher diagnostic yields. Two studies were performed by Zhu et al. [21, 22]. First [21], they showed a SE and SP of 85% for the combination of SMI VI and Virtual Touch Quantification (VTQ) of STE. In the second study [22], they reported an SE and SP of 87% and 85%, respectively, for US + SMI. There, they scored “vascular quantity” as 0–3, morphology as 1–7, and distribution as 1–3. The results of SMI + US in our study were better regarding the SE (100%), but the SP was lower (70.11%). Although the high specificity obtained in their study is very valuable, the lower specificity questions the superiority of the scoring system; and the final combination of US + SMI + STE as proposed in our study presents an obvious advantage. Liang et al. [23] assessed the diagnostic performance of US + SR + SMI. For SMI, they considered the vascular morphology only, and upgraded or downgraded the US-BIRADS according to it. This changed the SP from 44 to 81%, and the SE from 99 to 96%. The main deficit is that it only considered the morphology. In our study, the defined VS was based on definite precise data, was practical and easily calculated, and showed appropriate for the defined goal.

Zhou et al. [24] performed a meta-analysis to evaluate which combination of adjunct modalities was more accurate. The heterogeneity of studies was a serious limitation, and they could not directly envisage a combination of 3 modalities. Also, their approach to BIRADS was clinically impractical because they considered BIRADS 4a, 4b, and 4c as separate entities. Nonetheless, their conclusion was that BIRADS 4b + STE, + SMI was probably more suitable. Although approached differently, this result is in line with ours.

Despite the different settings, the studies that considered several combinations of US and auxiliary modes, including that of Liang et al. [23], Zhou et al. [24], and ours; found the highest diagnostic performance in US + STE + SMI.

Thus, regarding the low SP of US, and the need to auxiliary measures, the combination of US with SMI-VS and STE would yield very accurate results.

Our study had some limitations; the low number of malignant lesions and the lack of VI measurement. Also, the US was done by one radiologist, and the inter-observer difference was not assessed; thus the reproducibility of the study was not explored. In addition, the VR could have been calculated by using the mean or the sum of the dimensions of a mass,instead of the greatest dimension; this could have been a better solution in the case of asymmetric changes.

To conclude, the combination of US, SMI, and STE promotes the diagnostic yield in the differentiation of malignant and non-malignant lesions; increasing the specificity from 46.1% to 74.4%, while maintaining the 100% sensitivity. We propose a study that involves a larger sample size, considers inter-observer variability, and uses VS and the combination model to evaluate the accuracy with greater precision. Two other interesting subjects for future studies would be to compare VS with VI; and the outcomes of using this combination of modalities with the traditional method.