Introduction

According to the 2021 data released by the International Agency for Research on Cancer of the WHO,1 a staggering 2.26 million new cases of breast cancer were reported worldwide in 2020, surpassing lung cancer and establishing itself as the leading malignant tumour globally. Breast cancer subtypes are categorised based on molecular markers,2 including oestrogen receptor, progesterone receptor and human epidermal growth factor receptor 2 (HER2), leading to the classification of hormone receptor (HR)-positive, HER2-positive and triple-negative breast cancer (TNBC). TNBC, constituting approximately 15% to 20% of all breast carcinomas,3 stands out due to its aggressive nature, lack of targeted therapy options and poor overall prognosis.4 More than 50% of patients experience a relapse within the first 3–5 years after diagnosis,5 with a 5-year survival rate 8%–16% lower than HR-positive disease.6

In recent years, medical imaging, particularly ultrasound, has emerged as a valuable tool for characterising breast lesions and providing essential information for subtype classification.7 8 The distinctive ultrasound features of TNBC and non-TNBC lesions offer potential insights into their biological heterogeneity, thereby presenting an opportunity to establish a predictive model for subtype differentiation. Breast cancer diagnosis often involves a multistep process culminating in the determination of the tumour’s molecular subtype. In this context, the development of a predictive nomogram plays a vital role by aiding in the early differentiation of TNBC and non-TNBC. This expedited approach is poised to save crucial time, enabling the prompt design of a tailored and rational treatment strategy for TNBC patients. By incorporating this tool into clinical practice, we aim to streamline the diagnostic process and subsequently enhance the timeliness of therapeutic interventions, ultimately contributing to improved patient outcomes.

In this study, we conducted a comprehensive evaluation of ultrasound characteristics in TNBC and non-TNBC cases and subsequently developed a predictive nomogram based on these features. Through this approach, we aim to contribute to the refinement of non-invasive diagnostic modalities for breast cancer subtyping, ultimately enhancing personalised patient care and clinical decision-making.

Materials and methodsParticipants and study design

The study was conducted at the Quanzhou First Hospital over a 4-year period (September 2019 to August 2023), aiming to comprehensively investigate the ultrasound features and establish a predictive nomogram for TNBC and non-TNBC. A total of 779 patients were meticulously enrolled, comprising 205 cases of TNBC and 574 cases of non-TNBC. Each patient included in the study underwent thorough ultrasound examination, followed by a definitive pathological diagnosis, which was meticulously performed using H&E staining.

Patients were included in the study based on stringent criteria, ensuring the utmost standardisation and reproducibility of the data. Inclusion criteria primarily dictated that patients must have undergone ultrasound examination and received a confirmatory pathological diagnosis at the Quanzhou First Hospital. Additionally, all enrolled patients were required to possess a complete set of clinical and ultrasound images to ensure comprehensive and thorough data analysis. Conversely, exclusion criteria were carefully defined, targeting patients with absenteeism of clinical or ultrasound images, individuals who had received prior radiotherapy or chemotherapy, and those with a history of tumour recurrence. These clearly defined inclusion and exclusion criteria helped to fortify the integrity and robustness of the study. The fully detailed study flowchart, illustrated in figure 1, provides a comprehensive schematic representation of the systematic approach employed in patient enrolment and study progression. We recorded the following information for all patients: age, height, weight, clinical symptoms (including presence/absence of nipple discharge, breast tenderness and nipple retraction, etc), palpation (indicating whether the breast mass was palpable by a breast specialist), tumour size as well as ultrasound features such as shape, regularity, aspect ratio, margin, architectural distortion, microcalcifications, posterior echo, hyperechoic halo, internal flow, peripheral flow, resistive Index (RI) and the presence of axillary lymph node metastasis.

Figure 1

Flowchart of this study. TNBC, triple-negative breast cancer.

All experimental procedures were conducted in strict adherence to the principles outlined in the Declaration of Helsinki and other relevant guidelines.

Patient and Public Involvement (PPI) statement: We involved patients and members of the public in the design and conduct of this study. Patients were consulted to gain insights into their experiences with breast cancer diagnosis and treatment, and their perspectives were considered in the development of the research questions and study methodology. Additionally, we collaborated with relevant patient advocacy groups to ensure that the study objectives and outcomes were aligned with the needs and priorities of individuals affected by triple-negative and non-TNBC. Patient representatives also provided feedback on the comprehensibility and relevance of the study findings. Their input was valuable in shaping the research process and has contributed to the overall quality and applicability of the study.

We want to clarify that the calculation of interobserver coefficients was not performed. As our data collection spanned 4 years, changes in the pool of researchers and observers were possible. This situation, which could involve varying observers during patient examinations, is a common occurrence. It’s important to recognise that any assessment by specialists, including those interpreting ultrasound imaging, inherently carries subjective elements. Given the rigorous and meticulously described criteria in our study, we acknowledge that the evaluation of intra-rater reliability was not conducted. This information will provide transparency regarding the reproducibility of our study and the specific aspects of observer variability that were not addressed in our research.

Instruments and methods

The ultrasound imaging for this study was conducted using Lumify L12-4 scanners, which are high-resolution and portable devices manufactured by PHILIPS in the Netherlands. To ensure comprehensive assessment, patients were positioned in the supine position, facilitating the optimal exposure of both breasts and the axillae for detailed ultrasound examination. The obtained ultrasound images were systematically evaluated in accordance with the established standards outlined by the Breast Imaging-Reporting and Data System. Emphasis was placed on the meticulous retention of slices presenting notable nodular ultrasound characteristics. These included a thorough recording of nodule features encompassing size, shape, margin delineation, echogenicity relative to surrounding tissues, internal echoes, aspect ratio, posterior features, calcifications, echo edge, Color Doppler Flow Imaging (CDFI) grading to assess vascularity, and nodular resistance index. This methodological approach ensured a comprehensive and standardised assessment, serving as a solid foundation for subsequent analyses aiming to establish a predictive nomogram for distinguishing between triple-negative and non-TNBC based on ultrasound characteristics.

Ultrasound evaluation

The ultrasound assessments were carried out by ultrasound physicians with over 3 years of experience in breast ultrasound examinations. A retrospective analysis of all images was conducted by two ultrasound physicians with 10 years of experience in breast ultrasound diagnosis, respectively. They were blinded to the pathological results of the nodules to ensure unbiased evaluation. CDFI was used for semiquantitative assessment, using the Adler method9 to categorise the nodular blood flow signals into four levels: level 0 indicating absence of blood flow within the nodules, level 1 denoting minimal blood flow with 1–2 rod-shaped or punctate vessels within the nodules, level 2 representing moderate blood flow with one long or three to four punctate vessels, and level 3 indicating abundant blood flow with two long vessels or five or more punctate vessels within the nodules.

Statistical analysis

All statistical analyses were performed using the R statistical software, V.3.5.1. All quantitative data with a normal distribution and homogeneity of variance were presented as mean±SD and analysed using the t-test. For non-normal distributions or variances, the Mann-Whitney U test was employed. Categorical data were expressed as composition ratios or rate ratios and compared between groups using the χ2 test or Fisher’s exact test. Univariate logistic regression analysis was conducted to identify individual ultrasound features that were significantly associated with TNBC. Subsequently, a multivariate forward stepwise logistic regression analysis was performed to construct the predictive nomogram based on the selected ultrasound features. The receiver operating characteristic (ROC) curve was plotted to evaluate the model’s predictive ability using the area under the curve and 95% CIs. A calibration curve was constructed to assess the calibration of the model. Decision curve analysis (DCA) was used to examine clinical benefits. To gauge the goodness of fit of the established nomogram, the Hosmer-Lemeshow test was applied to assess the calibration of the model. The online dynamic nomogram was built with Shiny. A p value of less than 0.05 was deemed statistically significant.

ResultsCharacteristics of the training and validation sets

The overall study cohort comprised a total of 779 patients, with 523 patients designated to the training set and 256 patients to the validation set. The characteristics and outcomes of both TNBC and non-TNBC within the training and validation sets have been systematically enumerated and are presented in detail in table 1.

Table 1

Baseline characteristics in the training and validation sets

Ultrasonic characteristics in the training and validation sets

In the training set, TNBC and non-TNBC exhibited significant differences in clinical symptoms, size, margin, microcalcification, posterior echo, hyperechoic halo, internal flow and peripheral flow (p<0.05). In the validation set, TNBC and non-TNBC showed significant differences in microcalcification, hyperechoic halo, internal flow and peripheral flow (p<0.05).

Logistic regression analysis in the training set and construction of nomogram

Eight candidate variables, including margin, microcalcification, size, hyperechoic halo, internal flow, peripheral flow, posterior echo and clinical symptoms(p<0.05), were significantly associated with TNBC and non-TNBC in the univariate logistic regression analyses. Among them, margin(p=0.041), microcalcification(p=0.002), size(p=0.040), posterior echo (p=0.047), hyperechoic halo (p=0.001), internal flow and (p<0.001) and clinical symptoms (p=0.001) were identified as independent risk factors for breast cancer by the multivariate forward stepwise logistic regression analysis (table 2).

Table 2

Univariate and multivariate logistic regression analysis in the training set

Thereafter, a static nomogram and an online dynamic nomogram (https://suliyang.shinyapps.io/TNBCdynamic/) were developed by incorporating these seven predictors (figure 2).

Figure 2

Static nomogram model. The nomogram was developed based on logistic regression analysis results to predict the diagnosis of TNBC and non-TNBC. The dynamic nomogram can be accessed at https://suliyang.shinyapps.io/TNBCdynamic/. This model comprises margin, microcalcification, size (cm), posterior echo, hyperechoic halo, internal flow and clinical symptoms. TNBC, triple-negative breast cancer.

Evaluation of nomogram

The ROC curve was employed to assess the discriminatory ability of the predictive model. The pooled area under the ROC of the nomogram for the predictive model is 0.781 in the training set and 0.702 in the validation set, indicating moderately good performance (figure 3). The predictive model was calibrated using a calibration plot and the Hosmer–Lemeshow test. The calibration curves indicated a very good fit for the predictive model in the validation set. The Hosmer-Lemeshow test further demonstrated high consistency between predicted and actual probabilities (training set, p=0.891; validation set, p=0.567) (figure 4). DCA demonstrated that the model possessed clinical benefits (figure 5).

Figure 3

ROC curve of the training set (A) and validation set (B). The pooled area under the ROC of the nomogram for the predictive model is 0.781 in the training set and 0.702 in the validation set, indicating moderately good performance. AUC, area under the curve; ROC, receiver operating characteristic.

Figure 4

Calibration curve of the training set (A) and validation set (B). Calibration curves depict the correlation between the predicted probability (x-axis) and the actual probability (y-axis) of TNBC. The red line along the diagonal signifies where predicted probability equals actual probability, while the green line represents the calibration curve of the nomogram. Both the training and validation set curves closely align with the dashed line, demonstrating high calibration accuracy. TNBC, triple-negative breast cancer.

Figure 5

DCA of the training set (A) and validation set (B). The x-axis displays the threshold probability, while the y-axis quantifies the net benefit. The solid black line corresponds to the assumption that all patients are non-TNBC, the thin solid line corresponds to the assumption that all patients are TNBC, and the red line represents the risk nomogram. The DCA curve values lie above the lines for none and all, indicating acceptable model performance within this range. DCA, decision curve analysis; TNBC, triple-negative breast cancer.

Discussion

According to previous studies, nomograms have been widely used in predictive models for breast cancer, such as using nomograms to predict TNBC and breast fibroadenomas,10 benign and malignant breast tumours,11 12 fibrocystic changes and invasive ductal carcinoma.13 Nomograms can also be used to improve the diagnostic efficiency of malignant breast tumours.14 15 However, to the best of our knowledge, there has been no large-scale nomogram prediction for TNBC and non-TNBC. In this study, we conducted a retrospective analysis of breast cancer patients in our hospital over the past 4 years, comparing the clinical presentations and ultrasound images of TNBC and non-TNBC, and we established conventional and dynamic nomograms, providing clinicians with a more convenient tool.

Our study findings revealed that the size of tumours in TNBC is larger than those in non-TNBC, and TNBC is associated with fewer clinical symptoms. The correlation between these two characteristics could be explained by the fact that patients with fewer clinical symptoms might be less likely to seek medical attention for breast examination, potentially leading to undiagnosed breast cancer. Furthermore, the aggressive nature and rapid growth rate of TNBC tumours may contribute to their larger volume compared with non-TNBC. Our results align with other research findings; an Surveillance, Epidemiology, and End Results(SEER) analysis4 from 2010 to 2013 showed that TNBC accounted for only 8.2% of stage I cancers but exceeded 15% in high-stage disease at diagnosis. Additionally, triple-negative tumours represented only 8.5% of tumours diagnosed with a size smaller than 2 cm. These findings underscore the clinical relevance of our study and contribute to the growing body of evidence on the distinct characteristics and behaviour of TNBC, shedding light on the importance of early detection and screening for this aggressive breast cancer subtype. Further investigation into the underlying mechanisms driving the larger tumour size and fewer clinical symptoms in TNBC is warranted, providing insights that could ultimately impact patient outcomes and clinical management strategies.

Our study revealed that TNBC demonstrated fewer microcalcifications, fewer hyperechoic halos, less posterior acoustic shadowing and more internal flow on ultrasound compared with non-TNBC. Corroborating the essence of our findings, the meticulous meta-analysis conducted by Tian and colleagues,16 which incorporated data from 10 scholarly articles and the clinical narratives of 620 patients afflicted with TNBC, unveiled the tendency for TNBC to display fewer malignant ultrasound characteristics compared with non-TNBC lesions. TNBC lesions often exhibit ostensibly benign attributes; they tend towards regularity in shape, devoid of angular or spiculated margins, and frequently demonstrate posterior acoustic enhancement, an abrupt interface with surrounding tissues, a parallel alignment to the skin and an absence of calcific deposition, bearing a deceptive resemblance to benign tumours. Echoing these observations, the retrospective literature review by Schopp and team17 also revealed that TNBCs may masquerade as innocuous masses, eluding the typical ultrasound imagery associated with primary breast malignancies. The chameleon-like behaviour of TNBC can lead to its mischaracterisation as a simplistic or complex cyst, manifesting as an anechoic or near-anechoic lesion with potential concurrent posterior acoustic empowerment—yet, echogenic disparity in itself lacks specificity. Our research underscores the critical importance of a meticulous diagnostic approach in the evaluation of breast masses that may surreptitiously mimic benign entities in ultrasonic imagery.

Within the repository of literature, scant emphasis has been placed on the vascular intricacies of TNBC, a lacuna our study seeks to fill. Venturing into the uncharted territories of intratumoral haemodynamics, our research unveiled a revelation: TNBC is characterised by a more profuse internal blood flow signal as compared with non-TNBC counterparts. This could be interpreted as a sonographic whisper of the aggressive nature and the propensity for expedited growth that TNBC harbours. The observations we harvested suggest a manifold tapestry of vascularisation within TNBC, setting it apart from the non-TNBC landscapes. It hints at a divergence in the paths of angiogenesis and the architecture of the tumour microenvironment that underlies these two disparate subtypes of breast cancer.18 19Our findings revealed that TNBC exhibits a richer internal blood flow signal compared with non-TNBC. This observation suggests that the vascularisation pattern of TNBC may differ from non-TNBC, indicating potential differences in angiogenesis and tumour microenvironment between these two breast cancer subtypes. Further research into the underlying mechanisms driving the enhanced internal blood flow in TNBC could provide valuable insights into the pathophysiology of this aggressive breast cancer subtype and may offer new opportunities for targeted therapeutic interventions.20 21

In breast cancer imaging diagnostics, MRI and mammography are crucial tools; however, distinguishing TNBC from non-TNBC differs significantly between these modalities and ultrasound. Recent research22 has revealed the utility of synthetic MRI in identifying predictors specific to TNBC—showcasing T2 values and cystic/necrotic lesions as independent biomarkers. This research addresses the demand for precise non-invasive identification methods before surgery. Additionally, TNBCs typically appear as high-density masses on mammograms, most commonly displaying oval or round shapes (51%–88% of cases).17

We observed instances where the predictive output did not coincide with actual clinical diagnoses. This discordance, manifested in the form of false positives and negatives, warrants a meticulous dissection to unravel the underlying discrepancies. A detailed examination of false-positive cases, where our nomogram potentially indicated a diagnosis of TNBC in patients ultimately diagnosed with a non-triple-negative subtype, suggests that ultrasound features may overlap significantly between different subtypes. This overlap could be attributed to variations in tumour biology that are not captured purely by imaging characteristics.23 24 Conversely, the false-negative cases spotlight a set of patients for whom the nomogram underestimated the likelihood of triple-negative pathology. These instances could hint at the possibility of more subtle ultrasonographic nuances or the influence of interobserver variability in the assessment of ultrasound features. To enhance the predictive accuracy of our nomogram, it is imperative to delve into these discrepant cases and identify commonalities that may inform the refinement of our model. Potential factors could include the tumour size at the time of imaging, associated lesions or even intrinsic tumour heterogeneity.

We are keenly aware of the need for prospective validation to strengthen the reliability and broaden the applicability of our proposed predictive model. Prospective validation is crucial in the translational pathway of diagnostic tools, serving as a litmus test to ensure that a model’s performance is not limited to the retrospective population on which it was developed. We advocate for future research to focus on prospective validation of our predictive model in diverse clinical settings to verify its generalisability and clinical utility. In discussing the limitations of this study, it is important to acknowledge that our research is confined to the assessment of ultrasound features and the development of a predictive nomogram. As such, this study does not encompass other diagnostic modalities or genetic analysis that could further enhance the comprehensive understanding of TNBC and non-TNBC. Additionally, while our predictive nomogram shows promise, its generalisability should be further validated across diverse clinical settings and populations to ascertain its broader applicability.

The nomogram model developed in our study has shown excellent predictive performance, calibration and clinical application value in both the training and validation sets. This indicates that the nomogram could be a valuable tool for clinicians in accurately predicting the likelihood of TNBC based on ultrasound features. Our findings contribute to the growing body of evidence supporting the utility of ultrasound in the classification of breast cancer subtypes. Predicting TNBC using ultrasound features has important clinical implications, as it can assist in making informed treatment decisions and improving patient outcomes. Further research and validation of our nomogram model in larger patient cohorts are needed to confirm its reliability and applicability in clinical practice.

Conclusion

The results of our study demonstrated that ultrasound features can be valuable in distinguishing between TNBC and non-TNBC. Specifically, the presence of posterior echo, size, clinical symptoms, margin, internal flow, halo and microcalcification was identified as predictive factors for this differentiation. Microcalcification, hyperechoic halo, internal flow and clinical symptoms emerged as the strongest predictive factors, indicating their potential as reliable indicators for identifying TNBC and non-TNBC.