Optimizing S-detect classification accuracy for BI-RADS 4 breast nodules using multimodal ultrasound parameters

Introduction

Breast cancer is the most common malignant tumor affecting women worldwide, posing a serious threat to their life and health (1). According to the latest cancer statistics report released by the International Agency for Research on Cancer (IARC) of the World Health Organization in 2022, there are up to 2.3 million new cases of breast cancer every year in the world, and its prevalence is increasing annually (2). The Breast Imaging Reporting and Data System (BI-RADS) ultrasound classification enables risk assessment of breast nodules and differential diagnosis of the benign and malignant nature of the masses. Among them, BI-RADS category 4 lesions are a kind of lesion with a large malignant risk span, and the malignant probability is 2–95%. The diagnosis is easily affected by the subjective factors of the operator, and it is easy to miss and misdiagnose (3). Therefore, how to improve the accuracy of the diagnosis of BI-RADS category 4 lesions has always been a hot issue.

With the development of science and technology, the screening and diagnosis methods of breast tumors are increasing day by day, mainly including mammography, magnetic resonance, and ultrasound (4). Although mammography can identify small calcifications, it is not the first choice for screening and diagnosis because it is heavily influenced by gland density and radiation (5). Magnetic resonance examination has good resolution of soft tissue, no radiation, no trauma, and high sensitivity, but it is not the first choice because of long inspection time, many restrictions, and high cost (6). Due to its convenience, no radiation, high resolution, and real-time guided interventional therapy, breast ultrasound has been widely used in the clinical screening and diagnosis of breast tumors (7).

In recent years, the auxiliary application of artificial intelligence (AI) technology in the medical field is increasing; in particular, the combination of AI and ultrasound technology has become an important direction of medical image AI research (8). The computer-aided diagnosis (CAD) system can extract image features for analysis, which has the advantages of objectivity, stability, and high repeatability (9). As an AI software built into the ultrasound device, S-detect uses deep learning (DL) technology to automatically identify breast nodules in grayscale ultrasound images, and performs a comprehensive analysis from the aspects of shape, direction, edge, rear features, echo pattern, and so on, to give a diagnosis of “possibly benign” or “possibly malignant” (10). S-detect technology is less affected by the subjective factors of the instrument operator, the evaluation of the nature of breast tumors is more objective, and the automatic evaluation shortens the evaluation time and is more convenient and faster (11). However, current evidence shows inconsistent performance of DL in breast ultrasound diagnosis, with no conclusive evidence that it reliably outperforms radiologists or effectively enhances diagnostic accuracy. Further high-quality studies are needed to validate its clinical value (12). Some studies have shown that S-detect has a certain risk of false positives in the identification of benign and malignant BI-RADS category 4 breast nodules. In the study of He et al. (13), 76 breast nodules classified by conventional ultrasound as BI-RADS category 4 were detected by S-detect alone, and the diagnostic specificity was only 59.57%. In addition, S-detect’s analysis of ultrasound images lacked information on the blood flow and hardness of the mass and was unable to identify calcification, which was its main limitation.

Tumor angiogenesis is a key feature of breast cancer development. Studies have shown that the abundance of neovascularization in tumors can be used to assist in the identification of tumor properties, and malignant tumors often have an unusually rich blood supply (14,15). Therefore, hemodynamic information is essential to distinguish between the activity and potential malignancy of a mass. Another biological behavior of a malignant breast mass is rapid growth. Malnutrition caused by the rapid growth of a breast mass can lead to calcium salt deposition in the breast mass (16). Most studies have found that malignant tumors usually present with an irregular shape and dense distribution of microcalcifications, whereas benign lesions are mostly round or oval, comprising coarse calcifications with clear edges (17,18). Therefore, calcification information is also an important clue for the identification of benign and malignant breast nodules. In addition, the evaluation of tissue hardness is one of the key factors in determining the nature of the mass (19). Previous studies have shown that malignant lumps tend to be hard and inflexible, whereas benign lumps are relatively soft (20). Therefore, in the diagnosis process, comprehensive consideration of hemodynamics, calcification information, tissue hardness, and other factors can significantly improve the accuracy of benign and malignant breast nodules and provide a more reliable basis for clinical decision-making.

Conventional ultrasound can effectively make up for the deficiency of S-detect in the interpretation of blood flow information and calcification information. With color Doppler ultrasound, the clinician can evaluate the blood flow status of the lump. Among them, the Adler grading method, as a common evaluation means, uses a semi-quantitative method to describe the blood flow of the mass (21). In addition, by obtaining the spectrum of arterial blood flow, the resistance index (RI) of the blood flow in the mass can be calculated, which is of great significance for evaluating the blood supply and vascular characteristics of the mass (22). In general, a higher RI (>0.7 indicates a high risk of malignancy) indicates greater vascular resistance, which may be associated with abnormal tumor-associated angiogenesis (23,24). In addition, multi-sectional examination of the tumor with a high-frequency probe can clearly observe whether there is calcification inside the tumor, and evaluate the size, distribution, and morphology of calcification (25). Stöblen et al. (26) showed that the detection rate of microcalcification by high-frequency ultrasound could reach 98%.

Ultrasound elastography (UE), including strain elastography (SE) and shear wave elastography (SWE), is based on the elastic coefficient differences of tissues for diagnosis (27). SE reflects tissue hardness by assessing strain under pressure and is widely used in breast disease examination. Elasticity score (ES) and strain ratios (SR) are key indicators. A score ≥4 suggests high malignancy risk, and SR quantifies relative hardness by calculating the SR of the lesion to surrounding tissue (28,29). SWE calculates the elastic modulus by measuring shear wave speed, offering more objective quantitative data (30). SE, known for its ease of use, is often used for initial breast nodule screening, and combining ES and SR values helps to differentiate between benign and malignant lesions.

Microvascular flow imaging (MV-Flow), as a new generation of ultrasonic flow imaging technology, has significant advantages over color Doppler flow imaging (CDFI) (31,32). This technology uses advanced signal processing algorithms to detect extremely low speed blood flow signals at higher spatial resolution and frame rate, thus achieving fine display of the microvascular network of breast nodules (33). With the quantitative analysis function, the operator can delineate the region of interest (ROI) in the mass area, and the system automatically calculates the vascularity index (VI). Previous studies have shown that this parameter has a good correlation with histopathological microvascular density (MVD), which provides a reliable imaging indicator for the quantitative evaluation of tumor angiogenesis (34). This combination of high sensitivity and quantitative analysis ability gives MV-Flow technology unique clinical application value in the evaluation of breast tumor blood flow characteristics.

Currently, histopathological biopsy (surgical and needle biopsy) remains the gold standard for breast cancer diagnosis. Surgical biopsy is more invasive and carries higher risks of complications such as infection and excessive bleeding, making needle biopsy the preferred clinical approach. Breast cancer needle biopsy primarily includes fine-needle aspiration (FNA) and core-needle biopsy (CNB). FNA is limited by its small sample size, making it difficult to distinguish between in situ and invasive carcinoma. Although CNB can obtain more complete tissue samples, it typically requires 2–6 repeated punctures to minimize the risk of missed diagnosis, increasing both the medical burden and psychological distress for patients. Overreliance on biopsy also exacerbates resource strain and prolongs diagnostic workflows. Moreover, existing AI-assisted diagnostic technologies remain prone to false-positive results, further compounding these issues. To sum up, there is an urgent need to establish objective and quantitative diagnostic criteria to reduce the inter-observer differences in the diagnosis of BI-RADS category 4 breast nodules. In view of the limited information provided by a single ultrasonic mode, the joint diagnostic mode integrating multi-modal ultrasonic parameters has gradually become a research consensus. In this study, multi-modal ultrasound parameters (including Adler blood flow grading, RI, SR, VI, and calcification information, etc.) were integrated to optimize the classification results of S-detect on the four categories of breast nodules of BI-RADS, in order to improve the diagnostic accuracy. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1092/rc).

Methods

Participants

In this study, the clinical and multimodal ultrasound (MUS) data of 231 female patients with breast nodules identified by surgical pathology in The First Affiliated Hospital of Shihezi University from June 2019 to March 2024 were retrospectively included. All lesions were independently evaluated by two sonographers according to the American College of Radiology (ACR) BI-RADS 5th edition standard. The S-detect classification results based on grayscale ultrasound images were obtained for all breast nodules. Preoperative ultrasound evaluation of all tumors was classified as BI-RADS 4, including 106 cases of class 4a, 92 cases of class 4b, and 33 cases of class 4c. The final pathological diagnosis of breast nodules was malignant in 119 cases and benign in 112 cases. The median age of patients was 51.0 (45.0, 57.0) years and ranged from 20 to 84 years. All patients did not receive any treatment before surgery, and preoperative ultrasound images and clinicopathological data were complete. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of The First Affiliated Hospital of Shihezi University (No. KJ2024-456-02), and all patients signed informed consent before surgery.

Inclusion criteria:

• Complete ultrasound examinations (CUS, SE, and MV-Flow) were performed before surgery.
• Nodules classified into 4 categories according to BI-RADS.
• Complete ultrasound data and S-detect information available.
• After a clear pathological diagnosis, the nature of the mass was confirmed.
• Agreement to participate in this study.

Exclusion criteria:

• Preoperative biopsy, radiotherapy, chemotherapy, endocrine therapy, or breast surgery.
• No obvious breast nodules were found in ultrasound examination.
• Metastatic or non-primary breast lesions.
• No definite pathological diagnosis.
• Incomplete clinical or ultrasound data.

Equipment and methods

A Samsung RS85 color Doppler ultrasound diagnostic device (Samsung, Seoul, South Korea) was used, equipped with MV-Flow, LumiFlow, SWE, ultrasound contrast, quantitative, and S-detect capabilities. An LA2-14A linear array probe with a frequency of 2–14 MHz was used. Multisection scanning was performed with emphasis on the tumor area. Images were stored in the ultrasound diagnostic system for subsequent analysis. Clinical and ultrasound image data of patients with breast nodules were collected. Two sonographers with over 5 years of experience in breast ultrasound diagnosis independently reviewed the data without knowing the pathological results (Figures 1-5).

Figure 1 Flowchart of study participant enrollment. BI-RADS, Breast Imaging Reporting and Data System; MV, microvascular; SE, strain elastography.

Figure 2 An ultrasound image of a 62-year-old female patient with a malignant breast mass. The mass is located at the 2 o’clock position on the left breast, measuring approximately 2.01 cm × 1.12 cm. (A,B) Left: B-mode ultrasound (2D grayscale imaging); right: strain elastography. Strain A: the strain value of the target lesion; strain B: the strain value of the reference normal tissue. (C,D) MV-Flow velocity. 2D, two-dimensional; MV, microvascular.

Figure 3 The malignant mass in Figure 2 was judged as “Probably Benign” by S-detect.

Figure 4 An ultrasound image of a 51-year-old female patient with a benign breast mass. The mass is located at the 9 o’clock position on the left breast, measuring approximately 1.37 cm × 1.05 cm. (A,B) Left: B-mode ultrasound (2D grayscale imaging); right: strain elastography. Strain A: the strain value of the target lesion; strain B: the strain value of the reference normal tissue. (C,D) MV-Flow velocity. 2D, two-dimensional; MV, microvascular.

Figure 5 The benign mass in Figure 4 was judged as “Probably Malignant” by S-detect.

The patient lay supine with both arms raised to fully expose both breasts. The patient was instructed to breathe calmly. Scanning was performed sequentially from the outer upper quadrant, outer lower quadrant, inner lower quadrant, and inner upper quadrant of the breast. An overlapping scanning mode was used to comprehensively cover the breast tumor area, and conventional ultrasound parameters of the breast were recorded simultaneously, including maximum tumor diameter, Adler blood flow grading, RI, and calcification status.

Grayscale ultrasound images of breast nodules were respectively obtained in four sections: horizontal transverse section, maximum oblique long-axis section, vertical longitudinal section, and maximum oblique short-axis section. After freezing the image, the S-detect mode was entered. In this mode, the center of the mass was clicked, and the system automatically delineated the lesion boundary and defines it as the ROI. Manual correction was allowed if the difference between the automatically delineated boundary and the visually observed boundary exceeded 10%. After confirming the boundary, the system automatically identified and comprehensively analyzed the mass characteristics, then output a diagnostic result of “Probably Benign” or “Probably Malignant”. These labels were dummy-coded as 0 and 1 respectively, yielding four binary predictors named Q1, Q2, Q3, and Q4.

During conventional ultrasound dynamic scanning, a switch was made to SE mode. The ROI included the mass and surrounding normal tissue. The probe was kept perpendicular to the skin to clearly display the maximum diameter section of the mass. When the right indicator mark turned green (≥1 grid), indicating the pressure index was ideal, the image was frozen. The hardest area of the mass was selected as Zone A and normal breast tissue at the same depth as Zone B, ensuring equal areas for A and B. The system automatically calculated SR. Each mass was measured three times, and the average value was recorded.

A five-point scale was used for qualitative assessment of breast lesions: 1 point means that the lesion is soft; 2 points mean that the lesion has both soft and hard components; 3 points mean that the lesion is hard and smaller on the elastogram than on the two-dimensional (2D) image; 4 points mean that the lesion is hard and the same size on the elastogram as on the 2D image; and 5 points mean that the lesion is hard and larger on the elastogram than on the 2D image.

The MV-Flow software was applied to observe the blood vessels inside the breast nodule, freeze the section with the most abundant blood flow signal, and manually outline the edge of the breast nodule. The software would automatically identify and measure the VI value, measure each lesion three times, and then take the average value and record it.

Statistical method

Using R 4.4.1 statistical analysis software (R Foundation for Statistical Computing, Vienna, Austria), all the data were randomly divided into the training group (161 cases) and the verification group (70 cases) at a ratio of 7:3. The measurement data of normal distribution were expressed as mean ± standard deviation, and that of non-normal distribution were expressed as median (P25, P75). Counting data were expressed as examples (units) and percentages (%). Chi-squared test was used for comparison between groups. Univariable and multivariable logistic regression analyses were employed to identify risk factors for breast cancer; variables exhibiting P<0.05 in univariable logistic regression were entered into a multivariable logistic regression model constructed with a backward stepwise procedure (entry α=0.05, removal α=0.10). On the basis of these analyses, three predictive models were developed: (I) a MUS parameter model; (II) an S-detect classification model; and (III) a combined model integrating both sets of features. Then, we calculated the sensitivity (SE), specificity (SP), and positive predictive value (PPV) of the three models for the diagnosis of benign and malignant breast nodules. PPV, negative predictive value (NPV), accuracy (ACC), and receiver operating characteristic (ROC) curves of each model were drawn. At the same time, the area under the curve (AUC) was calculated, and the Delong test was used to compare the AUC differences of the models in pairwise. At the test level α=0.05, P<0.05 was considered statistically significant.

Results

Patient basic information

A total of 231 BI-RADS 4 category breast nodules were included in this study based on the inclusion and exclusion criteria. Patients were divided into a benign group (112 cases) and a malignant group (119 cases) based on pathological diagnosis. The median age of patients in the benign group was 49.5 years (interquartile range, 39.3–54.0, age range, 20–80 years), whereas the median age of patients in the malignant group was 54.0 years (interquartile range, 48.0–58.0, age range, 34–84 years). The nodules were randomly divided into a training set (161 nodules, including 83 malignant nodules, with a malignancy rate of 51.55%) and a validation set (70 nodules, including 36 malignant nodules, with a malignancy rate of 51.43%) at a ratio of 7:3.

Univariate analysis of the training set

The univariate analysis of the training set showed that the four planes of S-detect were significant for the diagnosis of benign and malignant breast masses. Univariate analysis of MUS parameters indicated that the SR of quantitative parameters related to hardness information was significantly higher in malignant lesions than it was in benign lesions [median 3.55 (2.39, 4.95) vs. 2.13 (1.47, 2.71), P<0.01], and the VI of quantitative parameters related to blood flow information was also significantly higher in malignant lesions [median 13.20 (6.80, 22.20) vs. 4.10 (0.00, 11.08); P<0.01]. The optimal cutoff values of SR and VI for malignant and benign breast nodules were 3.08 and 5.15, respectively. Additionally, the maximum tumor diameter, calcification, ES, RI, and Adler blood flow grade are also related factors of breast cancer (Table 1).

Multivariate analysis and model construction in the training set

Multivariate logistic regression analysis was performed on S-detect classification plane and multimodal parameters that were meaningful for univariate analysis. The results showed that the maximum oblique long axial plane and the maximum oblique short axial plane of S-detect, RI, calcification, SR, and VI were independent risk factors for malignant breast tumors (P<0.05). The S-detect model, MUS parameters model, and combined prediction model were constructed based on independent risk factors. The AUC values of the combined model in the training set and validation set were 0.788 [95% confidence interval (CI): 0.701–0.874] and 0.802 (95% CI: 0.672–0.932), respectively, which were significantly better than those of the single model (Tables 2,3, Figures 6-9).

Figure 6 ROC curves of the three models in the training and validation groups. MUS, multimodal ultrasound; ROC, receiver operating characteristic.

Figure 7 Calibration curves of the combined model in the training and validation groups.

Figure 8 Decision curves of the combined model in the training and validation groups.

Figure 9 Visualization of 10-fold cross-validation results.

Diagnostic performance of the models

Using pathological examination results as the gold standard, the diagnostic SE, SP, ACC, PPV, and NPV of S-detect were 85.54%, 74.36%, 80.1%, 78.0%, and 82.9%, respectively. For MUS, the corresponding values were 67.47%, 94.87%, 93.3%, 73.3%, and 80.7%. The combined assessment of MUS and S-detect demonstrated diagnostic SE of 86.75%, SP of 92.31%, ACC of 92.3%, PPV of 86.7%, and NPV of 89.4%, indicating more efficient performance for diagnosing benign and malignant breast masses, with AUC values of 0.93 (95% CI: 0.89–0.97) and 0.92 (95% CI: 0.83–0.97) (Tables 4,5, Figure 10).

Figure 10 Nomogram of the combined model. RI, resistance index; SR, strain ratio; VI, vascularity index.

Discussion

S-detect is an AI-assisted diagnostic system based on DL. By objectively analyzing the morphological features of breast nodules, it can reduce the impact of operator subjectivity in traditional ultrasound diagnosis and improve diagnostic efficiency. The software can be equipped in ultrasound devices and has high clinical applicability (10). However, this study found that although S-detect alone demonstrated high sensitivity (85.54%), its SP was relatively low (74.36%), a result consistent with previous studies, indicating that the system has a certain risk of false positives. In-depth analysis revealed that this limitation mainly stems from three aspects: First, S-detect analyzes only grayscale ultrasound images and lacks the assessment of hemodynamic features. Second, the system cannot obtain the important diagnostic indicator of tissue hardness. Finally, its ability to recognize calcification features (especially microcalcifications of significant diagnostic value) is insufficient.

Recent research has proposed innovative solutions to address these limitations through advanced architectures such as the 3MT-Net, a prospective DL model that leverages multimodal multi-task learning (35). This system dynamically weights high- and low-dimensional data from diverse modalities (including grayscale, Doppler, and elastography) while simultaneously performing binary classification (benign/malignant) and pathological subtyping. Retrospective multicenter validation demonstrated its superior performance, with AUC improvements of 1.4–3.8% over S-detect. The 3MT-Net’s success underscores the diagnostic value of integrating morphological, hemodynamic, and stiffness features—precisely the dimensions that S-detect lacks. Although this study integrated different dimensions of tumor information in ultrasound examination, only the CAD system and S-detect were compared, and it was not explored whether the fusion of these dimensions would be more expressive based on S-detect.

To address the technical limitations of S-detect, this study proposes a combined diagnostic strategy of multimodal parameters with S-detect. By integrating the morphological analysis advantages of S-detect and the comprehensive diagnostic information (including hemodynamic, tissue hardness, and calcification features) from MUS, a combined model for diagnosing the benign/malignant nature of BI-RADS 4 category breast nodules is built. Although S-detect’s DL algorithm enables objective and precise morphological analysis, MUS parameters offer additional functional assessment indicators. The results showed that the combined model maintains high SE (86.75%) but significantly improves SP to 92.31%, with an AUC of 0.93, indicating excellent diagnostic performance. This suggests that MUS parameters can effectively compensate for S-detect’s limitations. Moreover, the combined model surpasses single models (S-detect and MUS) in SE, ACC, and AUC, further confirming the strategy’s effectiveness. This approach not only offsets S-detect’s functional assessment limitations for a more comprehensive nodule evaluation but also reduces the reliance on operator experience in traditional ultrasound, providing a more objective and reliable diagnostic tool for clinical decisions.

This study deeply analyzed the correlation between ultrasound parameters and pathological results, finding significant differences in multiple dimensions for malignant masses. In hemodynamic features, the VI and RI of malignant masses were significantly higher than those of benign masses, aligning with the pathological features of abnormal angiogenesis in malignant tumors (34). In terms of tissue hardness, quantitative assessments via SR and ES confirmed that malignant masses generally have higher tissue hardness. Additionally, the difference in calcification characteristics provides vital clues for differential diagnosis, with a significantly higher proportion of microcalcifications in malignant masses. These findings not only validate the diagnostic value of these parameters in breast cancer but also lay a scientific foundation for constructing a combined model.

This single-center retrospective study has the following limitations: (I) sample size and subgroup analysis: The single-center retrospective design with only 231 cases precluded meaningful subgroup analyses of BI-RADS 4a, 4b, and 4c categories or molecular subtypes, potentially affecting result reliability. (II) The retrospective design may have introduced case selection bias, which may not fully reflect the distribution of BI-RADS 4 lesions. Prospective validation is needed. (III) Emerging technologies such as contrast-enhanced ultrasound (CEUS) and SWE were not included. CEUS provides detailed microcirculation perfusion information, and SWE enables absolute quantitative measurement of tissue hardness, both of which could add diagnostic value. (IV) Validation constraints: the study lacks external and multicenter validation, limiting generalizability across different clinical settings and populations. (V) Ultrasound equipment and elastography techniques from a single manufacturer were used in this study. Since technologies such as MV-Flow and S-detect are proprietary to Samsung, the generalizability of the findings to systems from other manufacturers (e.g., GE, Philips) may be limited. (VI) S-detect’s reliance on 2D grayscale image analysis makes it sensitive to image quality, which can be affected by factors such as patient body type and breast density, potentially preventing ideal image acquisition in some cases. (VII) The current semi-automated mode requires physicians to first acquire images and then manually input multimodal parameters, increasing workflow complexity. Future studies should address these limitations through multicenter prospective validation, incorporation of advanced imaging modalities, standardized protocols, automated workflow solutions, and diverse population testing.

In view of the above limitations, future studies should focus on exploring the following aspects to further enhance the clinical application value of the diagnostic model. First, a more refined diagnostic model was established for each subgroup of BI-RADS category 4, especially for category 4a nodules with lower risk of malignancy, and a more accurate risk stratification tool was developed. Secondly, the imaging characteristics of different molecular types of breast cancer were studied to provide imaging basis for individualized treatment. Finally, although the combined model proposed in this study has demonstrated excellent diagnostic efficiency in the identification of benign and malignant BI-RADS category 4 breast nodules, the further development of AI-assisted diagnostic systems still faces many challenges. Among them, the most critical is how to directly integrate key parameters such as hemodynamic information, tissue hardness characteristics, and calcification performance into the S-detect system to realize the optimization of the whole process from screening to diagnosis, and realize the true sense of multi-modal intelligent diagnosis.

Conclusions

S-detect demonstrates significant diagnostic value in distinguishing benign and malignant BI-RADS 4 breast nodules, particularly when analyzing the maximum oblique long and short axial planes. MUS parameters, including microcalcification, RI >0.7, SR, and VI, serve as independent risk factors for malignancy. Malignant lesions exhibit significantly higher SR (cut off: 5.15) and VI (cut off: 3.08) values compared to benign lesions. The combined S-detect + MUS model significantly improves diagnostic performance, achieving a sensitivity of 86.75%, specificity of 92.31%, and AUC of 0.93. This model enhances specificity while maintaining high sensitivity. This combined diagnostic strategy enhances precision in classifying BI-RADS 4 breast nodules while reducing unnecessary biopsies, and future research should focus on intelligent multimodal AI integration and multicenter validation to optimize automated diagnosis.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1092/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1092/dss

Funding: This work was supported by the Tianshan Young Talent Scientific and Technological Innovation Team: Innovative Team for Research on Prevention and Treatment of High-incidence Diseases in Central Asia (No. 2023TSYCTD0020), the Tianshan Talent Training Program for High-level Medical and Health Professionals (No. TSYC202401A006), the National Natural Science Foundation of China (Nos. 82460353 and 82060318), The First Affiliated Hospital of Shihezi University School of Medicine Youth Fund Project (Nos. QN202107 and QN202126).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1092/coif). All authors report that this work was supported by the Tianshan Young Talent Scientific and Technological Innovation Team: Innovative Team for Research on Prevention and Treatment of High-incidence Diseases in Central Asia (No. 2023TSYCTD0020), the Tianshan Talent Training Program for High-level Medical and Health Professionals (No. TSYC202401A006), the National Natural Science Foundation of China (Nos. 82460353 and 82060318), The First Affiliated Hospital of Shihezi University School of Medicine Youth Fund Project (Nos. QN202107 and QN202126). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of The First Affiliated Hospital of Shihezi University (No. KJ2024-456-02), and all patients signed informed consent before surgery.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Giaquinto AN, Sung H, Newman LA, Freedman RA, Smith RA, Star J, Jemal A, Siegel RL. Breast cancer statistics 2024. CA Cancer J Clin 2024;74:477-95. [Crossref] [PubMed]
2. Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
3. Spak DA, Plaxco JS, Santiago L, Dryden MJ, Dogan BE. BI-RADS(®) fifth edition: A summary of changes. Diagn Interv Imaging 2017;98:179-90.
4. McDonald ES, Clark AS, Tchou J, Zhang P, Freedman GM. Clinical Diagnosis and Management of Breast Cancer. J Nucl Med 2016;57:9S-16S. [Crossref] [PubMed]
5. Malherbe K, Tafti D. Breast Ultrasound. StatPearls. Treasure Island (FL): StatPearls Publishing; 2025.
6. Monticciolo DL, Newell MS, Moy L, Lee CS, Destounis SV. Breast Cancer Screening for Women at Higher-Than-Average Risk: Updated Recommendations From the ACR. J Am Coll Radiol 2023;20:902-14. [Crossref] [PubMed]
7. Guo R, Lu G, Qin B, Fei B. Ultrasound Imaging Technologies for Breast Cancer Detection and Management: A Review. Ultrasound Med Biol 2018;44:37-70. [Crossref] [PubMed]
8. Fruchtman Brot H, Mango VL. Artificial intelligence in breast ultrasound: application in clinical practice. Ultrasonography 2024;43:3-14. [Crossref] [PubMed]
9. Ling Y, Wang Y, Dai W, Yu J, Liang P, Kong D. MTANet: Multi-Task Attention Network for Automatic Medical Image Segmentation and Classification. IEEE Trans Med Imaging 2024;43:674-85. [Crossref] [PubMed]
10. Bartolotta TV, Orlando AAM, Di Vittorio ML, Amato F, Dimarco M, Matranga D, Ienzi R. S-Detect characterization of focal solid breast lesions: a prospective analysis of inter-reader agreement for US BI-RADS descriptors. J Ultrasound 2021;24:143-50. [Crossref] [PubMed]
11. Yu ZH, Hong YT, Chou CP. Enhancing Breast Cancer Diagnosis: A Nomogram Model Integrating AI Ultrasound and Clinical Factors. Ultrasound Med Biol 2024;50:1372-80. [Crossref] [PubMed]
12. Dan Q, Xu Z, Burrows H, Bissram J, Stringer JSA, Li Y. Diagnostic performance of deep learning in ultrasound diagnosis of breast cancer: a systematic review. NPJ Precis Oncol 2024;8:21. [Crossref] [PubMed]
13. He F, Xiao JD, Guo Y. Application value of artificial intelligence S-detect technology in ultrasound BI-RADS category 4 breast masses. Medical Information 2023;36:115-8.
14. Mou J, Li C, Zheng Q, Meng X, Tang H. Research progress in tumor angiogenesis and drug resistance in breast cancer. Cancer Biol Med 2024;21:571-85. [Crossref] [PubMed]
15. Zhang Q, Lu S, Li T, Yu L, Zhang Y, Zeng H, Qian X, Bi J, Lin Y. ACE2 inhibits breast cancer angiogenesis via suppressing the VEGFa/VEGFR2/ERK pathway. J Exp Clin Cancer Res 2019;38:173. [Crossref] [PubMed]
16. Hadhazi D, Vetro M, Hullam G, editors. Detection of Microcalcifications in Mammograms using Algorithmic Filtering and Region-based Neural Networks. IEEE International Symposium on Medical Measurements and Applications (MeMeA); 2024 Jun 26-28; Eindhoven Univ Technol, Eindhoven, Netherlands. New York: IEEE; 2024.
17. Yu LF, Dai CC, Zhu LX, Xu XJ, Yan HJ, Jiang CX, Bao LY. Detection and diagnosis of automated breast ultrasound in patients with BI-RADS category 4 microcalcifications: a retrospective study. BMC Med Imaging 2024;24:126. [Crossref] [PubMed]
18. Aleem J, Rehman S, Shafqat M, Zahra H, Ashraf J, Niazi IK. Recurrence Yield of Stereotactic Biopsy of Suspicious Calcifications After Breast Conservation Therapy. Cureus 2022;14:e24318. [Crossref] [PubMed]
19. Calzado-Martín A, Encinar M, Tamayo J, Calleja M, San Paulo A. Effect of Actin Organization on the Stiffness of Living Breast Cancer Cells Revealed by Peak-Force Modulation Atomic Force Microscopy. ACS Nano 2016;10:3365-74. [Crossref] [PubMed]
20. Egorov V, Kearney T, Pollak SB, Rohatgi C, Sarvazyan N, Airapetian S, Browning S, Sarvazyan A. Differentiation of benign and malignant breast lesions by mechanical imaging. Breast Cancer Res Treat 2009;118:67-80. [Crossref] [PubMed]
21. Zhang F, Li G, Jin L, Jia C, Shi Q, Wu R. Diagnostic value of Doppler imaging for malignant non-mass breast lesions: with different diagnostic criteria for older and younger women: first results. Clin Hemorheol Microcirc 2022;81:123-34. [Crossref] [PubMed]
22. Mehdikhani B, Benam M, Moradkhani A, Roostaee A, Bahman SS, Barmayoon P, Dezyani G, Safarpour Lima Z. Evaluation of diagnostic value of Doppler ultrasound in the diagnosis of malignant breast masses. Eur J Transl Myol 2024;34:12372. [Crossref] [PubMed]
23. Choi HY, Kim HY, Baek SY, Kang BC, Lee SW. Significance of resistive index in color Doppler ultrasonogram: differentiation between benign and malignant breast masses. Clin Imaging 1999;23:284-8. [Crossref] [PubMed]
24. Chinese Anti-Cancer Association Committee of Breast Cancer Society, Chinese Society of Oncology Breast Cancer Study Group. Chinese Anti-Cancer Association guidelines and standards for breast cancer diagnosis and treatment (2024 edition). China Oncology 2023;33:1092-187.
25. Ouyang YL, Zhou ZH, Wu WW, Tian J, Xu F, Wu SC, Tsui PH. A review of ultrasound detection methods for breast microcalcification. Math Biosci Eng 2019;16:1761-85. [Crossref] [PubMed]
26. Stöblen F, Landt S, Ishaq R, Stelkens-Gebhardt R, Rezai M, Skaane P, Blohmer JU, Sehouli J, Kümmel S. High-frequency breast ultrasound for the detection of microcalcifications and associated masses in BI-RADS 4a patients. Anticancer Res 2011;31:2575-81.
27. Cantisani V, David E, Barr RG, Radzina M, de Soccio V, Elia D, De Felice C, Pediconi F, Gigli S, Occhiato R, Messineo D, Fresilli D, Ballesio L, D'Ambrosio F. US-Elastography for Breast Lesion Characterization: Prospective Comparison of US BIRADS, Strain Elastography and Shear wave Elastography. Ultraschall Med 2021;42:533-40. [Crossref] [PubMed]
28. Youk JH, Son EJ, Gweon HM, Kim H, Park YJ, Kim JA. Comparison of strain and shear wave elastography for the differentiation of benign from malignant breast lesions, combined with B-mode ultrasonography: qualitative and quantitative assessments. Ultrasound Med Biol 2014;40:2336-44. [Crossref] [PubMed]
29. Li X, Sun W, Zhang H. The value of elastography strain rate ratio in benign and malignant BI-RADS-US 3-4 breast masses. Biomol Biomed 2024;24:625-32. [Crossref] [PubMed]
30. Evans A, editor. Breast Shear Wave Elastography in Clinical Practice. IEEE International Ultrasonics Symposium (IUS); 2019 Oct 06-09; Glasgow, UK. New York: IEEE; 2019.
31. Zhu YC, Zu DM, Zhang Y, Shan J, Shi XR, Deng SH, Jiang Q. A comparative study on superb microvascular imaging and conventional ultrasonography in differentiating BI-RADS 4 breast lesions. Oncol Lett 2019;18:3202-10. [Crossref] [PubMed]
32. Zhang X, Cheng F, Song X, Wang P, Tian S, Zhao X, Wang Q, Zhang M. Superb microvascular imaging for evaluation of microvascularity in breast nodules compared with conventional Doppler imaging. Quant Imaging Med Surg 2023;13:7029-40. [Crossref] [PubMed]
33. Park AY, Seo BK, Han MR. Breast Ultrasound Microvascular Imaging and Radiogenomics. Korean J Radiol 2021;22:677-87. [Crossref] [PubMed]
34. Park AY, Kwon M, Woo OH, Cho KR, Park EK, Cha SH, Song SE, Lee JH, Cha J, Son GS, Seo BK. A Prospective Study on the Value of Ultrasound Microflow Assessment to Distinguish Malignant from Benign Solid Breast Masses: Association between Ultrasound Parameters and Histologic Microvessel Densities. Korean J Radiol 2019;20:759-72. [Crossref] [PubMed]
35. Duan Y, Pang PC, He P, Wang R, Sun Y, Liu C, Zhang X, Yuan X, Song P, Lam CT, Cui L, Tan T. 3MT-Net: A Multi-Modal Multi-Task Model for Breast Cancer and Pathological Subtype Classification Based on a Multicenter Study. IEEE J Biomed Health Inform 2025;29:4680-91. [Crossref] [PubMed]