Introduction
Invasive breast carcinoma is an angiogenesis-dependent malignancy, and studies have indicated that an increased tumor microvessel density is associated with poor prognosis [
1,
2]. Importantly, blood vessels in malignant tumors are extremely heterogeneous and very different from vessels found in normal tissues or benign tumors. Poor oxygen levels in early-emerging tumors stimulate the release of vascular endothelial growth factors (VEGF), which initiates new vascularization and tumor growth [
2,
3]. In turn, the demand for additional oxygen in growing tumors leads to formation of leaky, fragile, tortuous vessels [
4]. In contrast, in most benign cases, tumor growth is controlled by mechanisms similar to those of normal tissue, leading to the creation of organized and non-tortuous vessel shapes [
4,
5].
Conventional Doppler methods with differentiating potential in breast masses [
6‐
10] are sensitive only to fast flows, leading to highly fragmented and patchy images of the underlying vessels, preventing structural analysis of microvessels. The utility of photoacoustic imaging approaches has been shown for microvessel architectural differences in superficial breast lesions [
11], but has limited use in deep-seated tumors. Contrast-enhanced ultrasound (US) has been investigated for increasing the specificity of ultrasound for differentiation of benign and malignant breast masses [
12,
13]. Acoustic angiography and ultrasound localization microscopy [
14‐
16], with the help of contrast agents, could resolve microvessels in preclinical studies [
17].
Recently, obtaining fine vascular features of breast tumors at super-resolution scales was possible in a spontaneous mouse model of breast cancer [
18] and in humans [
19], but this approach is associated with inconvenience and increased cost associated with injection of contrast agents. Contrast-free ultrasound imaging of tumor microvessels for differentiation of malignant from benign breast masses has been investigated; however, these efforts were limited to a pixel count method and visual inspection of images for the assessment of vessel shapes and distribution [
20‐
22]. To address these research gaps, we have previously developed a contrast-free ultrasound-based technology to visualize small submillimeter vessels (as small as 300 μm) and quantify tumor microvessel morphological structures, named quantitative high-definition microvasculature imaging (qHDMI) [
23,
24]. The objective of this research is to complement the gray scale morphology–based assessment of conventional ultrasound with the microvasculature features of breast tumor for increased accuracy in cancer detection.
Recently, the basic principles of 4 new quantitative biomarkers based on microvessel images, as well as some simulations and limited patient study results, were presented to illustrate the role of each biomarker [
25]. The four biomarkers are (1) microvessel fractal dimension, (2) Murray’s deviation, (3) bifurcation angle (BA), and (4) spatial vascular pattern [
25]. The goal of this study is to investigate the performance of the four newly developed HDMI quantitative biomarkers on a relatively large population. Thus, more lesion categories allowed us to investigate the performance of HDMI individual biomarkers and the combination of them in a multivariable analysis for different pathologies and different lesion size groups. The study also tests the performance of multiple prediction models, using only new biomarkers, only initial biomarkers, and a combination of new and initial with or with or without Breast Imaging Reporting and Data System (BI-RADS) scores. Furthermore, the correlation of HDMI biomarkers with cancer grades has been investigated. As such, the current validation study substantially expands the previous works. The proposed method objectively classifies the tumor as benign or malignant, which makes this method operator independent and eliminates the observer/reader variability for a reliable clinical use.
Discussion
This study investigated the performance of the novel quantitative biomarkers of contrast-free high-definition microvessel imaging (HDMI) for differentiating malignant and benign breast masses. Our findings show that four new HDMI biomarkers, SVP calculated by VDR, mvFD, BA, and MD, provided meaningful separation between malignant and benign lesion groups and outperformed our initial biomarkers (vessel diameter, vessel density, tortuosity, number of vessel segments, and number of branch points) [
24]. The multivariable analysis using a logistic regression classification method with all new biomarkers provided consistently better discrimination performance than any individual biomarker alone. Additionally, the discrimination power improved as tumors grow. The addition of BI-RADS scores based on US descriptors to the multivariable analysis using all biomarkers remarkably increased the sensitivity, specificity, and AUC in all size groups. This finding suggests that new quantitative HDMI offers complementary diagnostic information to conventional ultrasound for increased accuracy in breast cancer diagnosis. Moreover, an important advantage of this new tool is that it does not require injection of a contrast agent for better vessel enhancement. The envisioned strategy for the clinical use of the quantitative HDMI technique includes the following steps: (1) the ultrasound machine with the associated HDMI processing technique automatically processes the data providing the quantitative biomarkers. (2) The quantitative HDMI biomarkers will be further input to the prediction model implemented in the ultrasound machine to calculate the malignancy probability. (3) Then, the radiologist reads the malignancy probability and compares the value with the threshold for decision-making. The clinical application value of this HDMI technology is as follows: (1) if the malignancy probability calculated with the prediction equation is lower than the threshold, the algorithm would be more supportive of follow-up. As such, this model could help clinical decision-making, possibly downgrading a presumptive BI-RADS 4a lesion to a BI-RADS 3 with recommendation for follow-up. (2) If the malignancy probability is higher than the threshold, the algorithm would be supportive of breast biopsy. With additional validation, refinement, and testing with multi-center large-population studies, the threshold would be further validated.
Few studies have proposed ultrasound microvessel imaging for differentiation of breast masses, either with [
19,
22] or without [
37] contrast agents, with limited patient studies using a few morphological biomarkers. The current quantitative HDMI study includes a wide range of tumor microvessel morphological biomarkers tested on a relatively large group of patients. An additional advantage is that the enhancement and visualization of tumor vessels at the submillimeter level can be done without the need for contrast agents. Moreover, our method is capable of quantifying vessel diameter, which may be challenging in contrast-enhanced tracking approaches [
38].
This research investigates the performance of MD, BA, mvFD, and SVP as new morphological biomarkers of tumor microvessels in contrast-free ultrasound microvessel imaging for differentiation of breast lesions. The diagnostic value of MD was demonstrated for different diseases [
39‐
42], indicating that the vascular network of diseased tissue could show a deviation from Murray’s law [
43]. Our study also showed a higher MD in malignant breast lesions. Moreover, our study found a statistically significant decrease in BA in malignant breast lesions. Similarly, a decreased BA in invasive carcinomas of the colon has been shown in a previous study [
30]. In our study, mvFD was found to have higher values in malignant compared to benign lesions for all size constraints. This finding is consistent with the results of other studies, indicating that microvascular complexity calculated by mvFD may provide important diagnostic and prognostic information as well as insight into tumor angiogenesis [
27,
28]. In our study, the SVP biomarker indicated that peripherally concentrated vascularity in larger tumors (diameter > 20 mm) is associated with malignancy; however, in smaller tumors (diameter ≤ 20 mm), a centrally concentrated vascularity is an indicator of malignancy. This finding is also consistent with other studies suggesting that small malignant tumors have few large vessels in the periphery, but as the tumor enlarges, the vessel density decreases in the central area and the microvessels tend to have more peripheral distribution [
4,
44,
45]. If there are no or few microvessels within the lesion, the quantitative HDMI could classify the lesion as benign.
In this study,
Dmax was statistically significantly higher in malignant lesions compared to benign masses. In fact, using
Dmax, one can test the possibility of a major feeding vessel that may be indicative of malignancy. This result is consistent with the fact that VEGF-A forms numerous larger blood vessels (presumably mother vessels) in the periphery of malignant tumors, but fewer and smaller vessels in the central part of the tumor [
4,
44,
46]. Therefore, maximum vessel diameter has a better discriminatory power than averaging the diameter of the vessels. A similar observation was also made for vessel tortuosity. With tortuosity averaged over the entire vascular bed, there were no statistically significant differences between malignant and benign in all size groups; however, the maximum vessel tortuosity was statistically significantly higher in the malignant cases compared to the benign cases. These findings concur with the fact that, as a malignant tumor enlarges, more tortuous vessels with increased diameter are seen at the tumor–host interface than in the central region [
4], indicating that averaging these biomarkers has less diagnostic value than determining their maxima. This indicates that vessel tortuosity analysis can offer information complementary to flow imaging and may offer additive value in discrimination when both benign and malignant tumors are hypervascular [
47]. The increased numbers of branch points and vessel segments in our study signify a greater level of vessel sprouting, endorsing them as discriminators of benign and malignant tumors [
48].
Additionally, HDMI biomarkers were statistically significantly different between higher and lower NG grades of malignant breast tumors. Higher values of mvFD, a marker of vessel complexity, NB, NV, and VD were seen in the higher grades of breast carcinomas. Previous studies reported a higher microvessel density, sprouting, and structural irregularity associated with higher pathological grades of breast carcinomas that may lead to higher incidences of metastasis and a poorer prognosis [
28,
49,
50].
One limitation in this study is that the quantitative biomarkers were estimated using 2D HDMI which may overlook some important 3-dimensional (3D) morphological features and the connectivity of tumor microvessels, potentially leading to either underestimation or overestimation of these features. To address these limitations, a complementary study would involve quantitative 3D HDMI imaging and morphometric analysis using either a mechanical scanning system equipped with a linear array [
51] or a matrix ultrasound transducer [
52] for volumetric imaging. Such approaches would enable a more comprehensive vessel morphological analysis. To keep a single gold reference standard for all patients, the pathology results of core-needle biopsy rather than the surgical pathology served as the gold reference standard. As surgical pathology is not available in benign lesions that do not normally have surgical excision for treatment and in the group of complete pathological responders to neoadjuvant therapy will be no cancer. However, the histological features of cancer in core needle biopsy were the same as with surgical pathology.
Future work should also focus on using the emerging radiomic analysis approach by incorporating a data characterization algorithm to extract numerous features from images. Although radiomic analysis has its own challenges [
53], it may have the potential to facilitate improved clinical decision-making [
54]. Another direction for improving diagnostic performance of ultrasound is to combine our microvasculature morphometric analysis with established conventional ultrasound metrics. Conventional ultrasound provides information about the shape and texture of a breast lesion to aid in cancer detection, while our quantitative microvasculature method provides information related to angiogenesis. Combining these two pieces of information may improve the overall diagnostic performance of ultrasound. The HDMI study was performed on patients with suspicious breast lesions detected by clinical ultrasound and scheduled for biopsy. Nearly all cases were classified as BI-RADS 4 and 5. Therefore, we believe it would not be fair to compare the sensitivity and specificity of our method to conventional ultrasound since limited numbers of BI-RADS categories lower than 4 were included in our study. A future study could include microvasculature morphometric analysis of breast lesions regardless of their BI-RADS category, with the caveat that cases in lower BI-RADS categories will not have pathology results for comparison. The focus of the present study is to validate the performance of the new HDMI biomarkers for breast lesion differentiation on a large patient population. For future studies, we would like to compare the performance of HDMI to other diagnostic methods, e.g., B-mode, color Doppler, and the contrast-enhanced ultrasound.
In conclusion, the efficacy of the four novel quantitative biomarkers of the HDMI method for breast cancer detection is promising. The fact that HDMI does not require injection of a contrast agent simplifies its use in routine clinical practice. In the future, the proposed method with new biomarkers can offer a new means of detecting breast cancer when used as a complementary imaging tool to conventional ultrasound.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.