Background
For decades, angiogenesis has been recognized as a driving factor for rapid growth and metastasis of malignant solid tumors [
1]; as such neovascularization plays a critical role in breast cancer growth and dissemination. Importantly, the structural abnormalities of malignant tumor microvessels are evident not only by high density but also by complexity of the newly formed vessels presenting with irregularity and tortuosity [
2,
3]. Direct assessment of vessel morphological changes as biomarkers for cancer detection by imaging modalities is an emerging research interest.
Previous research has demonstrated the ability to obtain microvascular features of breast tumors at super-resolution scales [
4] although no morphometric analysis was performed. A few studies have proposed ultrasound imaging of tumor microvessels for differentiation of breast masses without using contrast agents [
5]; however, these efforts were limited to a pixel count method and visual inspection of images for the assessment of vessel shapes and distribution. Recently it has been demonstrated that slow blood flows can be successfully decoupled from strong tissue clutter signal when highly registered spatial–temporal data were processed in the singular-spectrum domain [
6]. This approach includes a set of novel vessel enhancement processing algorithms to extract small vessels and suppress unrelated structures in the Power Doppler (PD) image. This approach provides tissue-blood echo separation and prepares the image for quantitative analysis of the microvasculature network and its morphology
. Noting that this method can visualize small sub-millimeter vessels, as small as 300 µm, it has been termed high-definition microvasculature imaging (HDMI) [
6,
7]. HDMI is primarily based on using ultrafast imaging, providing a significantly higher number of coherent imaging frames by utilizing plane-wave scanning instead of the conventional focused line-by-line scanning methods [
6]. Quantitative analysis of microvessel morphological parameters as new biomarkers is described in [
8]. In this work, morphological parameters of 2D HDMI images of breast lesions were extracted and used for lesion characterization [
9]. In a more recent comparative study [
10], it has been shown that adding the morphological parameter to those of shear wave elastography parameters improves the overall lesion characterization performance.
The integrity of the microvasculature morphology is vital for quantitative analysis [
11]. As angiogenesis leads to formation of chaotic and tortuous vessels in malignant lesions [
12], vascular information obtained from single imaging plane approach is incomplete [
13]. Furthermore, 2D imaging methods overlook some important 3D structural features of microvessels and their connectivity, leading to underestimation or overestimation of different morphological parameters in a 2D plane [
14,
15]. Therefore, 3D microvasculature imaging would be helpful for overcoming these limitations as more complete microvasculature information can be obtained.
3D microvasculature imaging for differentiation of breast masses including volumetric photoacoustic imaging, [
16,
17] and 3D ultrasound localization microscopy [
4,
11‐
13] are active research area; however, the latter requires IV placement, contrast medium injection and is costly. Efforts towards using contrast-free 3-D ultrasound imaging of microvessels, taking advantage of a high frame rate ultrasound technique, and providing volumetric information to differentiate breast masses using a vascular index have been reported [
18]; however, morphometric analysis was not provided in that study.
In this work, we introduced a contrast-free quantitative three-dimensional high-definition microvasculature imaging (q3D-HDMI) method to provide microvasculature morphological information within a tumor volume. The q-HDMI methods (3D and 2D) objectively classify breast tumors in benign or malignant, which makes this method less operator dependent and eliminates the observer/reader variability providing a reliable clinical ultrasound imaging. Both the q2D-HDMI and q3D-HDMI techniques are not FDA cleared/approved and they remain investigational. We tested the hypothesis that the proposed contrast-free 3D microvasculature imaging provides complete and more accurate information regarding vessel morphological features and outperforms q2D-HDMI approach in distinguishing malignant from benign breast masses.
Discussion
In this study, we presented in vivo imaging of breast tumor microvessels volumetrically along with comprehensive quantitative analysis of tumor microvessels extracted from the novel q3D-HDMI for the first time. No contrast agent was applied. We demonstrated that q3D-HDMI outperforms the 2D approach and increases the accuracy in malignancy prediction of breast lesions.
The results of our study included one false positive. The pathology confirmed the lesion to be a fibroadenoma with lactational changes in background breast parenchyma. It is known that fibroadenoma is a hormone sensitive tumor and often becomes enlarged with increased vascularity during pregnancy and lactation. Furthermore, fibroadenoma tends to get atypical radiological features during pregnancy and lactation, necessitating core needle biopsy to confirm the diagnosis [
25]. The study also included 3 false negatives. The pathology of two of false negatives was DCIS with low nuclear grade cribriform type. Studies show that low-grade DCIS tend to have less vascularity and may develop into low-grade invasive carcinoma [
26]. Indeed, Grade 1 DCIS generates much less angiogenesis than high nuclear grade of DCIS [
27]. In other words, the false negativity of these low-grade DCIS in our study is more physiologic than a technical issue. The third false negative was IDC Nottingham grade II (of III), subtype category of Luminal A. It is known that luminal A subtypes are slow growing cancers and have low microvessel density with less structural complexity and they maintain the best prognosis among all subtypes [
28]. In other words, this false negative IDC was more physiologic than a technical issue.
Radial scars are often difficult to differentiate from IDC on conventional imaging and pathology review is often non-trivial [
29]. All three radial scars included in this study were correctly predicted as benign. Thus, with q3D-HDMI, the microvasculature features of radial scars could assist for more accurate differentiation. Furthermore, the proposed prediction model correctly predicted all the masses with fibrocystic breast changes without evidence of hyperplasia as benign. Our finding is supported by report of previous study indicating that fibrocystic breast disease without hyperplasia shows lower grade microvessel density compared to those associated with hyperplasia [
30].
Though, 3D power Doppler ultrasound for breast cancer diagnosis [
31] and contrast agent mediated 3D ultrasound localization microscopy on tumor bearing rats [
13] demonstrated the potential of such volumetric techniques; our study demonstrated the effectiveness of the q3D-HDMI through a comprehensive quantitative in vivo study for characterization of breast masses in human.
It is known that angiogenesis, the formation of new microvessels toward and within a malignant breast tumor, starts when the tumor reaches the size of 2–4 mm in diameter [
32]. One of the important findings in this study is that q3D-HDMI was able to capture the microvasculature structures in a breast lesion as small as 3 mm. Furthermore, the q3D-HDMI displayed more vessels with better connectivity in this small tumor than what was possible with q2D-HDMI. As a result, the prediction model based on q3D-HDMI quantitative biomarkers correctly diagnosed this small lesion as malignant. It is known that male breast cancer has even more intense angiogenic reaction than female breast cancer [
33], this is in support of our study, which included two men with breast masses and q3D-HDMI quantitative biomarkers correctly diagnosed them as invasive ductal carcinoma.
To the best of our knowledge, this study is the first to compare 3D and 2D microvessel images quantitatively and demonstrates that the new 3D microvasculature imaging technique, q3D-HDMI, is superior to the current q2D-HDMI method for several reasons. With 2D microvasculature imaging, only one plane of the lesion can be evaluated [
13]. In our study, for the 2D analysis, the microvessel image corresponding to the B-mode image with the largest lesion area was selected. However, the microvasculature information in the malignant lesion is highly spatially heterogeneous, and therefore the microvessel information could largely vary from slice to slice, making it hard to select a representative 2D slice for analysis. Therefore, the microvasculature morphological features obtained on a single plane in 2D imaging may not be a true estimation and could be underestimated or overestimated [
14]. For example, our study showed that the mvFD, NB and tortuosity information obtained from q3D-HDMI was significantly higher than the same parameters obtained from q2D-HDMI and significantly different in benign and malignant. In addition, q2D-HDMI can visualize the vessel only in a two-dimensional space, hence vessels extending outside the imaging plane and their morphological parameters would not be captured. On the other hand, q3D-HDMI captures the entire vessel network and their morphological parameters, providing a more accurate estimation of such parameters. This finding is in agreement with previous study [
14]. Our study also showed that the vessel density measured with q2D-HDMI was significantly higher than that measured with 3D, indicating that q2D-HDMI overestimated microvessel density. However, this is expected as the vessel density in 2D is calculated per unit area, whereas in 3D it is calculated per unit volume.
In this study, the recently introduced parameters for breast cancer detection, Murray’s deviation [
8,
34], representing the relationship between mother vessel and daughter vessels and bifurcation angle [
8,
35] based on information between the daughter vessels, were significantly different in benign and malignant in q3D-HDMI approach, indicating its ability to accurately visualize branches information in a 3-dimensional space. However, for q2D-HDMI, most of the lesions resulted in NaN values for these two parameters, suggesting that no daughter vessels were observed.
Microvessel morphology and its distribution feature vary between benign and malignant breast tumors and are likely to be an important discrimination marker [
2,
36]. Studies have shown that morphological parameters of tumor microvessels obtained from contrast-free quantitative 2D-HDMI increases the sensitivity and specificity in discriminating malignant from benign breast masses [
7,
9,
10]. The study presented here shows that morphological biomarkers of microvasculature network obtained by 3D-HDMI outperforms those of q2D-HDMI and increases the sensitivity and specificity in differentiating malignant from benign breast masses. It is known that the morphology of the vasculature in the immediate vicinity of breast tumors plays a significant role in differentiation between cancerous and benign masses. Studies in [
19,
37] used a combined optoacoustic and ultrasound images of breast tumors to show that aggressively growing malignant tumors tend to recruit vasculature from the immediate proximity so that small arteries penetrate cancerous tumors radially. In contrast, benign tumors push surrounding vasculature out, so those vessels appear on and parallel to the surface of fibroadenomas.
The performance of combined OA and grayscale US for breast lesion differentiation has been shown in [
19,
37]. Study in [
19] also showed that OA/US increased the specificity of breast mass assessment compared with the device internal grayscale ultrasound alone. A direct comparison of the sensitivity/specificity for HDMI and OA/US requires a separate comparative study. However, since HDMI and OA/US present different characteristics of breast lesions, it is expected that they could complement each other in breast lesion differentiation. Therefore, it will be interesting to combine q3D-HDMI with OA/US in the future to further improve the specificity of breast lesion differentiation and reduce the unnecessary biopsies.
Our study had limitations. The sample size was small. Also, to have pathology as gold reference standard, only the participants with recommendation for breast biopsy, nearly all BI-RADS 4 and 5, were enrolled in this study. This prevented a direct comparison between q3D-HDMI and gray scale ultrasound. In the future, we plan to apply this method on a larger population irrespective of their BI-RADS category. Also, there is a potential for data degradation due to breathing motions, as patients were allowed to breathe normally during data collection. In the future we plan to utilize and expand the motion correction algorithms [
38‐
40] to reduce potential motion artifacts or to use deep learning technique [
41,
42] with potential for correction of motion artifacts.
Acknowledgements
The authors would like to thank Dr. Rohit Nayak, PhD, Dr. David Rosen, PhD, Ms. Shaheeda A. Adusei, MS., and Mr. Jeremy Webb, MS., for their assistance in data acquisition at different periods during the patient studies. Also, the authors would like to thank Mr. Duane Meixner, R.V.T., R.D.M.S., Ms. Kate Knoll, R.V.T., R.D.M.S for scanning patients, Ms. Cindy Andrist and Ms. Patricia O'Neil for their valuable help in patient recruitment, and Dr. Lucy Bahn, PhD for her editorial help.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.