Introduction

Breast cancer is the most commonly diagnosed cancer and second leading cause of cancer death among women globally [1, 2]. Breast cancer is characterized by significant heterogeneity, leading to variable genetic, phenotypic and behavioural characteristics, clinical manifestations, and treatment reactions [3,4,5,6,7,8].

More recently, gene expression analysis with complementary DNA microarrays has been used to classify breast cancer into four molecular subtypes: luminal-A, luminal-B, human epidermal growth factor receptor-2 positive (HER2+), and TNBC [9, 10]. These subtypes have been demonstrated to represent prognostic and predictive information in breast cancers.

Ultrasound is a technology with the advantages of safety, noninvasiveness, and cost effectiveness that is commonly used for breast cancer diagnosis and screening [11]. To improve the diagnostic accuracy, we adopted multimodal information from B-mode, real-time elastography (RTE), colour Doppler flow images (CDFI) and contrast-enhanced ultrasound (CEUS) videos [12]. RTE can reflect the stiffness of the tissue [13]. CDFI can detect blood flow, which often increases in the tumour areas [14]. CEUS is a quantitative kinetic imaging method that can measure blood flow in breast tumours down to the level of the capillaries [15,16,17,18,19]. A cutting-edge technique for CEUS video quantification, time-intensity curve (TIC) analysis, extracts quantitative parameters of tumour blood perfusion [20,21,22]. These features obtained from multimodal ultrasound are associated with biological biomarkers and molecular subtypes and can help patients manage using precision treatment.

Regarding tumour growth and invasion, the microenvironment of tumours is recognized to play a critical role [23, 24], and peritumour tissue has been proven to provide useful information for diagnosis and prognosis prediction [25,26,27,28,29]. However, how peritumour tissue should be analysed has received relatively little attention [28].

In our study, we explored the correlation between breast lesion multimodal ultrasound features and specific breast cancer molecular subtypes to aid in the rapid diagnosis and early treatment of specific breast cancer molecular subtypes.

Materials and methods

Patient collection and breast cancer classification

The medical ethics committee of the First Affiliated Hospital of Zhejiang Chinese Medical University approved this retrospective study and complied with the Declaration of Helsinki. Informed consent was waived for this retrospective research.

From March to November 2021, 85 breast cancer patients from our hospital underwent ultrasound examinations (B-mode, RTE, CDFI, and CEUS), and images were recorded. We investigated the pathology reports of the patients by reviewing the clinical records.

The criteria for inclusion were as follows: (1) pathology confirmation of a newly diagnosed breast cancer; (2) initial diagnosis of unilateral invasive breast cancer with a single lesion; (3) no history of other organ cancers; (4) no neoadjuvant chemotherapy or endocrine therapy before surgery; (5) mass-forming breast lesions.

The criteria for exclusion were as follows: (1) breast cancer history; (2) multiple lesions in bilateral breast cancer; (3) ultrasound images of poor quality; (4) no postsurgical pathological reports; and (5) non-mass forming breast lesions.

Based on the Immunohistochemistry (IHC) results of oestrogen receptor (ER), progesterone receptor (PR), HER2, Ki-67, and fluorescence in situ hybridization (FISH), all breast cancers were divided into the following four molecular subtypes [9, 10]: (1) luminal A: ER and/or PR positive, and HER2-negative, and Ki67 < 14%; (2) luminal B: ER and/or PR positive, and HER2-negative, and Ki67 ≥ 14% or ER and/or PR positive and HER2-positive; (3) HER2+: HER2-positive, and ER and PR negative; and (4) TNBC: ER and PR negative, and HER2-negative.

Ultrasound Examinations

All examinations were performed by an experienced sonographer (H. HL) with more than 15 years of breast technical and diagnosis experience so that the images were consistent. We employed a Techno MyLab Twice US system (Esaote, Genoa, Italy).

The LA523 probe with a frequency of 4–13 MHz was used for B-mode, RTE, and CDFI examinations. The protocol for breast scanning was as follows: The patient was placed in the supine position with the hands raised to fully expose the breast and armpits. The breast was scanned with continuous cross-section and longitudinal section. US images were all recorded on the largest transverse plane of the breast lesion.

RTE image acquisition was immediately performed and was conducted according to the World Federation for Ultrasound in Medicine & Biology guidelines for performing US elastography of the breast [46]. Briefly, the probe was vertically placed on the skin. An ultrasound probe was used to perform a light repetitive compression motion on the lesion. To obtain the best strain elastography, the pressure indicator on the screen was kept in green; that is, at least five of the seven pressure squares were displayed.

During CDFI examination, the blood flow in and around the lesions was observed, and a default equipment setting was implemented: a scale of 7 cm/s, a medium wall filter, and a pulse repetition frequency of 1.0 kHz.

Finally, we performed a CEUS examination using a 3–90 MHz linear transducer (LA522) probe. The machine was set as the default condition of breast CEUS. SonoVue (Bracco SpA, Milan, Italy) contrast agent was used. For contrast-tuned imaging, a bolus of 4.8 mL of contrast agent mixed with saline solution was injected via an antecubital vein, followed by aflush with 10 mL of 0.9% normal saline solution. Simultaneously, the dynamic pictures were recorded from the start of the injection and viewed for 120 s. The whole video was saved on the US machine for subsequent analyses.

Ultrasound Image Analysis

Ultrasound images were assessed retrospectively by two sonographers (B. HW and J. XC) with more than 20 years of breast imaging and breast tumour diagnosis experience using a blinded study design, indicating that sonographers were not informed of the molecular subtype of each tumour throughout the analysis. The following features were evaluated: B-mode ultrasound image features (based on BI-RADS: shape, orientation, margin, echo pattern, posterior characteristic, and calcification), RTE score, and CDFI score. When the two sonographers reached opposing opinions, a chief physician sonographer (C. F, with more than 25 years of breast tumour diagnosis experience, recognized by the medical community) was consulted to make a definitive decision.

RTE has a five-point scale. Itoh’s approach defined tumour stiffness as follows [13]: (1) score of 1: the entire lesion was evenly shaded in green; (2) score of 2: the hypoechoic lesion had a mosaic pattern of green and blue; (3) score of 3: the peripheral part of the lesion was green, and the central part was blue; (4) score of 4: the entire lesion was blue, but its surrounding area was not included; (5) score of 5: both the entire hypoechoic lesion and its surrounding area were blue (Fig. 1).

Fig. 1
figure 1

Examples of CDFI and RTE score. (a) CDFI grade 1; (b) CDFI grade 2; (c) CDFI grade 3; (d) RTE score of 2; (e) RTE score of 3; (f) RTE score of 4; (g) RTE score of 5. CDFI grade 0 and RTE score of 1 were not showed in our dataset.

Full size image

Adler’s approach defined tumour blood flow as follows [14]: (1) grade 0: no blood flow; (2) grade 1: minimum blood flow (1–2 dot-like signals or short-line-like signals); (3) grade 2: moderate blood flow (3–4 dot-like signals or 1 blood vessel longer than the lesion radius); and (4) grade 3: significant blood flow (3 or more blood vessels) (Fig. 1).

The region of interest (ROI) in each CEUS was manually drawn along the tumour by a sonographer (Z. JY) who was blinded to the clinical and histological data of the patients, and the tissue 3–5 mm surrounding the tumour was defined as peritumour. Three ROIs were sketched in the CEUS of the same lesions—ROI1 (corpus mamma ROI), ROI2 (intratumour ROI) and ROI3 (peritumour ROI)—where the time-intensity curve (TIC) of the contrast transit was recorded by using QontraXt (Esaote, Genoa, Italy). Additionally, using analysis software, the quantitative parameters were computed. Two quantitative parameters of breast lesions on CEUS were observed: peak and time to peak (TTP) (Fig. 2).

Fig. 2
figure 2

Contrast-enhanced ultrasound (CEUS) time intensity curve (TIC) analysis. The figure showed the Peak and Time to Peak (TTP). (a) ROI1: corpus mamma ROI; (b) ROI2: intratumour ROI; (c) ROI3: peritumour ROI.

Full size image

Peak (%): the maximum intensity of the enhancing curve during the bolus given by the formula [(postcontrast signal—precontrast signal)/precontrast signal] ×100%.

TTP(s): the time from the appearance of the first microbubbles in the lesion to its maximum peak intensity.

Data and statistical analysis

We focused on the relationship between ultrasound characteristics and specific subtypes of breast cancer. All the features were screened by chi-squared test and Fisher’s exact test and then were subjected to binary logistic analysis. Each subtype was subjected to a separate binary logistic regression analysis. Breast cancer molecular subtype was a binary variable in this study, with 1 indicating that a tumour is a subtype of interest, such as HER2+, and 0 indicating any other subtype. Binary logistic regression was performed for the four specific breast cancer molecular subtypes. The data were analysed by SPSS 26.0 (IBM, International Business Machines Corp., New York, US). P < 0.05 indicated a statistically significant difference.

Results

Baseline characteristics

Among 85 patients, 36 were luminal A, 20 were luminal B, 16 were HER2 + and 13 were TNBC. The mean participant age ± standard deviation was 52.9 years ± 9.2 (range, 30–77 years), and the mean tumour size was 2.33 ± 1.14 (range, 0.35–5.28 cm). The patient characteristics and tumour histopathologic features are summarized in Table 1.

Table 1 Patient and tumour characteristics
Full size table

Correlation between the B-mode、 RTE、 CDFI ultrasound features and specific breast cancer molecular subtypes

The luminal A cancer subtype was significantly correlated with tumour nonparallel orientation (p = 0.014), a spiculated margin (p = 0.022), no calcification (p = 0.004), high stiffness (p < 0.001), and a weak blood flow signal (p = 0.004).

The luminal B cancer subtype was significantly associated with tumour low stiffness (p = 0.003).

The HER2 + cancer subtype significantly corresponded with the tumour microlobulated margins (p = 0.024), isoechoic echo patterns (p = 0.003), and a rich blood flow signal (p = 0.039).

The TNBC subtype was significantly associated with a rich blood flow signal of the tumour (p = 0.007).

The tumour ultrasound shape and posterior features did not significantly correspond to any molecular subtypes. The results are shown in Table 2.

Table 2 The tumour B-mode、RTE、CDFI ultrasound features per molecular subtype
Full size table

Correlation between the CEUS features and specific breast cancer molecular subtypes

Regarding CEUS features, in our study, we used the peak ratio among corpus mamma (ROI 1), intratumour (ROI 2) and peritumour (ROI 3) for analysis. The results are shown in Table 3.

Table 3 The tumour CEUS features per molecular subtype
Full size table

The peak ROI2/peak ROI1 was significantly associated with the luminal B cancer subtype (p = 0.050). Compared with the peak of corpus mamma, low enhancement of the intratumour was associated with luminal B.

Peak ROI3/Peak ROI2 was significantly correlated with HER2+ (p = 0.014) and TNBC (p = 0.009) subtypes. Compared with the intratumour peak, the low peritumour enhancement is associated with HER2+, and the high peritumour enhancement is associated with TNBC.

The analysis of Peak ROI3/Peak ROI1 and TTP did not significantly correspond with any molecular subtypes.

Binary logistic regression analysis of specific breast cancer molecular subtypes

The binary logistic regression analysis results are summarized in Table 4.

Table 4 Binary logistics regression analysis of molecular subtypes with ultrasound feature
Full size table

Binary logistic regression showed that RTE (P < 0.001) and CDFI (p = 0.036) were predictive of the luminal A cancer subtype (C-index: 0.741). RTE (P = 0.016) and the peak ratio between intratumour and corpus mamma (P = 0.036) were predictive of the luminal B cancer subtype (C-index: 0.788). The peak ratio between peritumour and intratumour (P = 0.039) was independently predictive of the HER2 + subtype (C-index: 0.859); CDFI (P = 0.002) demonstrated excellent discrimination for predicting the subtype of TNBC (C-index: 0.847).

Discussion

Breast cancer has traditionally been regarded as a heterogeneous disease [30, 31]. The identification of human breast cancer subtype-specific molecular features has substantial implications for clinical treatment options, disease progression, and ultimately patient prognosis [32, 46, 47]. Tumour blood flow differs across breast cancer tumour subtypes. Previous studies have demonstrated that tumour blood flow and metabolism differ across breast cancer tumour subtypes, a finding that is congruent with the molecular heterogeneity across tumour types identified by gene profiling [48, 49].

Our study had the following limitations: we based our analysis on breast tumours from a single centre, and it was a retrospective analysis. Furthermore, multiple data could help make more accurate models that can predict the molecular subtype of breast cancer.

Conclusions

Our investigation suggested that high stiffness and insufficient blood supply of breast tumours is related to luminal A. Low stiffness of tumour and low enhancement of intratumour relative to the corpus mamma is related to the Luminal B cancer subtype. High intratumour enhancement compared with peritumour enhancement is related to the HER2 + cancer subtype. Tumour with an abundant blood supply and high peritumour enhancement are related to TNBC. These multimodal ultrasound features, especially intra- and peritumour CEUS features, may help noninvasively predict specific subtypes.