An XGBoost Machine Learning Based Model for Predicting Ki-67 Value ≥ 15% in T 2 N X M 0 Stage Primary Breast Cancer Receiving Neoadjuvant Chemotherapy Using Clinical Data and Delta-Radiomic Features on Ultrasound Images and Overall Survival Analysis: A 5-Year Postoperative Follow-Up Study

Abstract

Objective: To establish a model based on clinical and delta-radiomic features within ultrasound images using XGBoost machine learning to predict proliferation-associated nuclear antigen Ki-67 value ≥ 15% in T₂N_XM₀ stage primary breast cancer (BC). Method: Data were collected from 228 randomly selected BC patients who received ultrasound screening and postoperative pathologic assessment from April 2015 to September 2018. The patients were classified into the study group (n = 80) and control group (n = 148), and the data were apportioned into the training set and test set at a 7:3 ratio based on time intervals. In the training set, crucial factors were identified from clinical features and grayscale and delta-radiomic features within ultrasound images, by using the chi-square test, t-test, and rank-sum test. The clinical model, imaging model, and combined model were built using multivariate logistic regression, respectively. The model's predictive performance and clinical net benefit were assessed using DeLong's method and decision curve analysis. Meanwhile, an XGBoost algorithm is used to establish a prediction model to verify the above results. Results: The crucial factors affecting Ki-67 value ≥ 15% included BMI, lymph node metastases, BC volume, CA153, pathology type, tumor boundaries, tumor morphology, elastography score, and delta-radscore. The predictive performance of the combined model [AUC 0.857, OR 0.0290, 95% CI 0.793-0.908] was considerably improved on the training set than the clinical model [AUC 0.724, OR 0.0422, 95% CI 0.648-0.792] and the imaging model [AUC 0.798, OR 0.0355, 95% CI 0.727-0.857]. The decision curve analysis also confirmed that the combined model delivered a higher clinical net benefit, and the verification on the test set yielded similar results. The nomogram and the calibration curve plotted based on the combined model achieved satisfactory clinical effects. The SHAP value of the XGBoost algorithm also confirmed that lymph node metastasis, BC volume, elastography score, and delta-radscore are the best independent factors for predicting BC Ki-67 value ≥ 15%. Conclusion: The XGBoost machine learning-based combined model integrating clinical features and delta-radiomic features on ultrasound images was able to predict the Ki-67 value ≥ 15% in an efficient and noninvasive manner, providing important clues for clinical decision-making and follow-up in BC.

Keywords

ultrasound examination delta-radiomics breast cancer tumor proliferating cell nuclear antigen 67 XGBoost algorithm

Introduction

Recent years have witnessed an aggravation of environmental pollution alongside an intensification of life pressure in China, where breast cancer (BC) accounts for 24.2% of all female cancers, according to an estimation. The incidence of BC is 60 per 100 000, which is equivalent to twice the total incidence of cervical cancer, ovarian cancer, and endometrial cancer combined. The incidence of BC still keeps rising every year. Along with the rapid development of precision medicine, the combination of radiology and biomarker screening has played an increasingly important role in the individualized therapy of BC. Ultrasound (consisting of elastography, contrast-enhanced ultrasound, and ultrasonic blood flow imaging) is the preferred imaging technique for BC screening, with an accuracy rate of 95%. Ki-67 is a prominent proliferation marker of BC cells and is closely related to prognosis. A Ki-67 value ≥ 15% has been confirmed to be predictive of a high metastasis and recurrence rate and poor prognosis. Early prediction of Ki-67 value ≥ 15% provides an important basis for making treatment decisions in the mammary gland department. Predicting Ki-67 expression can inform the assessment of biological behaviors and prognosis of BC. Physicians may develop individualized treatment regimens for BC based on Ki-67 expression, which generally consist of neoadjuvant chemotherapy, targeted therapy, breast-conserving therapy, or radical resection, to improve the treatment efficacy while reducing side effects. Ki-67 is detected by immunohistochemistry which involves an aspiration biopsy or a pathologic evaluation of surgically resected specimens. However, immunohistochemistry is a subjective and invasive technique and may be disfavored by some patients or delayed after surgery. There is an urgent need for an accurate, noninvasive, and repeatable method for Ki-67 detection. Radiomics offers great potential in characterizing genetic heterogeneity of solid tumors of the liver and kidney and has the advantages of stability, noninvasiveness, and no additional cost.^1–5 Machine learning algorithms involved model can help screen reliable factors and test the stability of models, improving their predictive performance. The present study built a predictive model for the Ki-67 value ≥ 15% based on clinical and delta-radiomic features using the XGBoost algorithm, which achieved a high predictive performance and provides objective data for perioperative decision-making in BC (Figure 1).

Figure 1.

According to Pubmed, in recent years, the hot topics of breast cancer research have focused on molecular mechanisms, surgery, radiotherapy, and chemotherapy management. There are rare reports on the prediction of Ki-67 expression by ultrasound radiomics combined with machine learning.

Materials and Methods

A retrospective analysis was performed for clinical and radiographic data of 275 BC patients who received surgeries at the Central Hospital of XiaoGan from April 2015 to September 2018. Inclusion criteria: (1) female patients who received treatments at our hospital and had T₂N_XM₀ stage BC lesions (breast invasive ductal carcinoma, infiltrating lobular carcinoma, ductal carcinoma in situ [DCIS] with microinvasion [DCIS-MI]) confirmed by aspiration biopsy or postoperative pathology; (2) having a complete set of preoperative ultrasound images; (3) definite detection results of Ki-67 expression. Exclusion criteria: (1) poor ultrasound image quality, making the delineation of lesion boundaries difficult; (2) incomplete clinical or follow-up data; (3) having previously received surgery or radiochemotherapy; and (4) secondary BC or combined with other tumors, triple-negative breast cancer (TNBC). Based on the above inclusion and exclusion criteria, 228 BC patients were finally included and split into the training set and the test set at the 7:3 ratio based on time intervals. The patients were aged 29 to 65 years old, with an average of 49.31. The clinical baseline data included age, BMI, diabetes, hypertension, family history of BC, lymph node metastasis, NLR, PLR, BC volume, BC location, CA153, CA125, CEA, HCG, pathology type (breast invasive ductal carcinoma, infiltrating lobular carcinoma, DCIS with DCIS-MI), the clear boundary of BC, tumor morphology, elastography, radscore, ultrasound enhancement pattern, and several others.^6,7 The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board at the Central Hospital of XiaoGan Cancer Center (Issue XGYY20230091) and all patients provided written informed consent. Patients were able to leave the study at any time at their own request or were able to withdraw at the discretion of the investigator for safety, behavioral, or administrative reasons. The reasons for discontinuation were documented. The reporting of this study conforms to STROBE guidelines⁷ (Figure 2).

Figure 2.

The case registration and grouping diagram of this study.

Ultrasound Image Collection, Delineation of Regions of Interest, and Delta-Radiomics Score Extraction

Ultrasound scans were performed using Philips IU22 and Siemens S2000 ultrasound systems attached with high-frequency linear array probes with a probe frequency of 7 to 12 MHz and 6 to 15 MHz, respectively, in the breast ultrasound exam mode. The patients took a supine position during the scans, with 2 arms uplifted to fully expose the breasts and armpits. A multisection scan was performed for each quadrant of bilateral breasts. A focused scan was performed for the lesions, and the scan depth and harmonic wave were adjusted based on lesion size. To reduce the differences between scans of different ultrasound instruments, the harmonic, grayscale, velocity scale, elastic imaging, etc parameters of the different equipment should be calibrated and co-ordinated before scanning. Static images of the maximum 2D ultrasound section of BC were saved. The ultrasound images were imported into the 3D Slicer software (V5.03). Two breast sonographers with over 10 years of experience were invited to manually delineate the regions of interest (ROIs) on the ultrasound images without knowing the pathology results. A controlled analysis was performed. An intraclass/interclass correlation coefficient > 0.7 indicated a good consistency, and it was required that ICC > 0.75 in the present study. The open-source R software was used to preprocess all ultrasound images and extract radiomic features. Image preprocessing consists of the following steps: (1) image normalization to ensure that all features are of the same scale and scope, thereby avoiding the influence of differences in factor or proportion between different categories of data on feature extraction; (2) preprocessing: The original images were cropped to cover an additional 10% to 20% of the area along the margins of the ROI. The purpose was to reduce the computational load involved in feature extraction and preclude the effects of tumor-adjacent tissues. From the ROI in each patient, 859 radiomic features were extracted in total, including 18 first-order statistical features, 276 texture features, and 396 wavelet features. Texture features are divided into gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), gray level dependence matrix (GLDM) and neighboring gray-tone difference matrix (NGTDM). The original image is decomposed into 4 components: low pass/high pass (LH), high pass/low pass (H), high pass (HH), and low pass (LL) using a wavelet filter, and the first-order statistics and texture features are calculated for each component, resulting in 97 features for each component. In this study, the neoadjuvant chemotherapy scheme for breast cancer was the AC-T scheme: 4 courses of chemotherapy were given every 3 weeks, first with doxorubicin and cyclophosphamide, and then paclitaxel. Our team extracted radiomics features before and after neoadjuvant chemotherapy, with a delta feature value = ultrasonic radiomics feature value after neoadjuvant chemotherapy − ultrasonic radiomics feature value before neoadjuvant chemotherapy. To reduce overfitting, we performed LASSO dimension reduction plus 10-fold cross-validation for radiomic features on the training set. Next, the dataset with the minimum binomial bias in cross-validation was considered the optimal feature set, and the radiomic score (delta-radscore) was constructed^8–10 (Figure 3).

Figure 3.

Schematic diagram of the ultrasound delta-radiomics score in this study.

Statistical Process

All statistical analyses were conducted using the R software. The measurement data obeying a normal distribution were expressed as X ± S and compared using the independent-samples t-test. Otherwise, the measurement data were expressed as M(Q, Q) and compared using the Mann-Whitney U test. The enumeration data were expressed as rates and ratios and analyzed by the chi-square test. The clinical and imaging features with statistically significant intergroup differences in Ki-67 expression (Ki-67 value ≥ 15% or otherwise) were included in the logistic regression analysis. The clinical model, radiomic model, and combined model were constructed to identify independent risk factors for Ki-67 expression. The performance of the 3 models was compared using Delong's method. Moreover, the predictive performance of each model was assessed on the training set and the test set using the area under the ROC curve (AUC). The 95% confidence interval of AUC was calculated by repeating the resampling for 1000 times. The calibration curve was used to assess the consistency between the predicted and observed Ki-67 values. The clinical net benefit of each model was calculated using the decision curve analysis under different threshold probabilities to assess the clinical applicability of each. All of the graphs and curves were plotted using R software. P < .05 was considered statistically significant (2-sided). In addition, we used Python software to establish an XGBoost algorithm predictive model, screened reliable independent risk factors, and calculated their SHAP values.^11,12

Results

The univariate analysis confirmed that BMI, lymph node metastases, BC volume, CA153, pathology type, tumor boundaries, tumor morphology, elastography score, and delta-radscore were risk factors predicting the Ki-67 value ≥ 15%. The age, family history of BC, NLR, PLR, BC location, CA125, CEA, HCG, and ultrasound enhancement pattern were not of statistical significance between the 2 groups. The multivariate analysis confirmed that lymph node metastasis, BC volume, elastography score, and delta-radscore were the risk factors predicting the Ki-67 value ≥ 15%. The combined model based on the above-identified factors exhibited a much higher predictive performance [AUC 0.857, OR 0.0290, 95％ CI 0.793-0.908] on the training set compared with the clinical model [AUC 0.724, OR 0.0422, 95％ CI 0.648-0.792] and the imaging model [AUC 0.798, OR, 0.0355, 95％ CI 0.727-0.857]. The decision curve analysis demonstrated a high clinical net benefit of the combined model, and the result was verified on the test set [combined model-AUC 0.845, OR 0.0461, 95％ CI 0.736-0.921] versus [clinical model-AUC 0.727, OR 0.0606, 95％ CI 0.606-0.828] versus [AUC 0.771, OR 0.0583, 95％ CI 0.683-0.864]. The nomogram and the calibration curve plotted using the combined model achieved satisfactory clinical effect. Then, the XGBoost algorithm is used to predict BC Ki-67 ≥ 15% based on the above factors and perform ROC analysis, which confirmed an AUC of 0.965, 95％ CI (0.924-0.988). We used XGBoost algorithms to analyze the best factors for predicting BC Ki-67 ≥ 15% and calculate correspondent SHAP values, confirming that “DeltaRadscore,” “BCvolume,” “elastography,” “lymph node metastasis,” and “CEA” were the most valuable independent risk factors, correspondent to logistic regression analysis. We conducted a Kaplan-Meier analysis on the risk factors selected by the XGBoost algorithm and obtained better overall survival analysis prediction results (Tables 1 to 3 and Figures 4 to 9).^11,13,14

Figure 4.

The 3 models in this study were analyzed using the Delong curve. The left figure shows the training set and the right figure shows the test set. (A) The clinical model, (B) the imaging model, and (C) the combined model.

Figure 5.

The decision curve analysis of the combined model with the training (left) and testing sets (right) in this study confirmed that the combined model had a higher clinical net benefit.

Figure 6.

The nomograms (left) and calibration curves (right) developed based on the combined model have been well evaluated in clinical trials.

Figure 7.

XGBoost algorithm is used to predict BC Ki-67 ≥ 15% based on the above factors and perform ROC analysis, which confirmed an AUC of 0.965, 95％ CI (0.924-0.988).

Figure 8.

XGBoost algorithms are applied to analyze the best factors for predicting BC Ki-67 ≥ 15% and calculate correspondent SHAP values, confirmed that “DeltaRadscore,” “BCvolume,” “elastography,” “lymph node metastasis,” and “CEA” were the most valuable independent risk factors, similar to logistic regression analysis.

Figure 9.

The Kaplan-Meier analysis was conducted using the risk factors selected by the XGBoost algorithm and better overall survival analysis prediction results were obtained.

Table 1.

Logistic Regression Analysis Results of Clinical Model Based on Clinical Characteristics for Predicting the BC Ki-67 ≥ 15%, *P < .05.

Clinical model	Univariate analysis		Multivariate analysis
Factors	P	Hazard ratio	P	Hazard ratio
Age	.82	0.99 (0.93-1.05)
BMI	.04*	1.09 (1.01-1.18)
Diabetes	.17	1.67 (0.81-3.45)
Hypertension	.74	0.89 (0.47-1.72)
Family history of breast cancer	.23	0.67 (0.35-1.29)
Lymph node metastasis	.04*	2.07 (1.01-4.26)	.01*	2.77 (1.23-6.22)
NLR	.06	1.40 (0.99-1.97)
PLR	.19	1.15 (0.93-1.42)
BC volume	.03*	1.01 (1.00-1.02)	.03*	1.01 (1.00-1.03)
BC location	.12	1.41 (0.92-2.16)
CA153	.04*	1.11 (1.01-1.23)
CA125	.31	0.98 (0.96-1.01)
CEA	.06	1.29 (0.99-1.67)
HCG	.17	1.71 (0.79-3.71)
Pathology type	<.05*	1.89 (1.19-3.02)	.01*	1.86 (1.13-3.06)

Abbreviations: BMI, body mass index; BC, breast cancer; CEA, carcinoembryonic antigen; NLR, neutrophil/lymphocyte ratio; PLR, platelet/lymphocyte ratio; CA125, cancer antigen 125; HCG, human chorionic gonadotropin.

* P < .05 indicates a statistically significant difference.

Table 2.

Logistic Regression Analysis Results of Imaging Model Based on Imaging Characteristics for Predicting the BC Ki-67 ≥ 15%, *P < .05.

Clinical model	Univariate analysis		Multivariate analysis
Factors	P	Hazard ratio	P	Hazard ratio
Tumor boundaries	.04*	1.98 (1.03-3.83)
Tumor morphology	.04*	1.18 (0.96-1.46)
Elastography score	.01*	1.01 (0.99-1.02)	.01*	1.90 (1.28-2.81)
Delta-radscore	<.05*	3.28 (1.98-5.42)	<.05*	3.32 (1.95-5.68)
Ultrasound enhancement pattern	.23	0.81 (0.59-1.14)

* P < .05 indicates a statistically significant difference.

Table 3.

Logistic Regression Analysis Results of Combined Model Based on Mentioned Above for Predicting the BC Ki-67 ≥ 15%, * P < .05.

Clinical model	Univariate analysis		Multivariate analysis
Factors	P	Hazard ratio	P	Hazard ratio
BMI	.04*	1.09 (1.01-1.18)
Lymph node metastasis	.04*	2.07 (1.01-4.26)	.01*	3.51 (1.35-9.10)
BC volume	.03*	1.01 (1.00-1.02)	.02*	1.02 (1.00-1.03)
CA153	.04*	1.11 (1.01-1.23)
Pathology type	<.05*	1.89 (1.19-3.02)
Tumor boundaries	.04*	1.98 (1.03-3.83)
Tumor morphology	.04*	1.18 (0.96-1.46)
Elastography score	.01*	1.01 (0.99-1.02)	.01*	1.72 (1.13-2.61)
Delta-radscore	<.05*	3.28 (1.98-5.42)	<.05*	3.77 (2.09-6.81)

Abbreviations: BMI, body mass index; BC, breast cancer; CA125, cancer antigen 125.

* P < .05 indicates a statistically significant difference.

Discussion

BC manifests as uncontrolled cell proliferation under the joint action of several carcinogenic factors. Early BC is associated with the symptoms of breast lumps, nipple discharge, and axillary lymph node enlargement, and late-stage BC is usually complicated by distant metastases and multiorgan lesions that can be life-threatening. It has been reported that the prognosis is poor in patients with Ki-67 value ≥ 15%, and the 5-year survival is low. This study intended to predict Ki-67 value in BC by constructing a model based on clinical and radiomic features and to verify that lymph node metastasis, breast cancer volume, elastography score, and delta-radscore were an independent predictor. We found that the combination of delta-radscore and clinical features further improved the model's predictive performance and may help physicians make more correct decisions. In recent years, high resolution has been achieved by high-frequency ultrasound, and ultrasound elastography is undergoing rapid progress. The detection rate of ultrasound for minor lesions in breasts has been increasing, which makes early diagnosis and prognostic prediction of BC possible. At present, radiomics has been widely applied to differentiating between benign and malign lesions and to pathological typing, and prognostic prediction. According to some reports, radiomics based on magnetic resonance imaging (MRI) and computed tomography (CT) can predict the recurrence and prognosis of solid tumors. From the entire tumor area on the dynamic contrast-enhanced MRI and the apparent diffusion coefficient (ADC) maps for gliomas, researchers have extracted radiomic features and built support vector machine classifier models to predict the Ki-67 expression state with high efficiency.^15–18 However, MRI is expensive, and patients may need to wait for some time to make reservations to receive a scan. Ultrasound is the first-line diagnostic technique for BC screening in China. However, few researchers have discussed the prediction of Ki-67 value using ultrasound-based radiomics in BC. Compared with previous models, our predictive model integrating clinical and delta-radiomic features of BC patients using the XGBoost algorithm enjoyed the advantages of convenience, rapidity, economical efficiency, and noninvasiveness. The AUC of the proposed logistic regression model was 0.857 and 0.845 on the training set and the test set, respectively. This model provides reliable support for the clinical management of BC.^19,20

Our study confirmed that BMI, lymph node metastases, breast cancer volume, CA153, pathology type, tumor boundaries, tumor morphology, elastography score, and delta-radscore were predictive factors for the Ki-67 value ≥ 15%. Patients with BMI > 28.5 are considered overweight and tend to have lower immunity. These patients are more prone to liver and kidney dysfunction after neoadjuvant chemotherapy in BC. Internal environment disturbance is associated with a higher tumor proliferation and hence a higher Ki-67 value. The surgery is usually more difficult for BC patients with lymph node metastases, in whom tumor invasion is wider in extent and there is a higher possibility of distant metastases, which, in turn, gives rise to a higher Ki-67 expression. CA153 is a protein produced by a variety of cells, particularly BC cells. It has been reported that the higher the CA153 level, the more advanced the BC in the TNM staging system. A higher CA153 level also indicates an active proliferation of BC cells and represents the extent of tumor invasion and bioactivity. Therefore, the CA153 level is positively related to Ki-67. In this study, the pathological classification of the breast mainly includes breast invasive ductal carcinoma, infiltrating lobular carcinoma, and DCIS with DCIS-MI. DCIS-MI usually has a good prognosis, breast invasive ductal carcinoma and infiltrating lobular carcinoma have a high degree of malignancy, low differentiation, severe infiltration of surrounding tissues, and a high probability of recurrence and metastasis, which affects the expression of Ki-67. BC boundary is a commonly used cancer staging indicator in radiography. A well-defined cancer boundary usually indicates confined and slow cancer growth. Such cancers have an envelope, infiltrating into the surrounding soft tissues to a limited degree, and they are featured by low invasiveness and better prognosis. By contrast, those with blurred boundaries are mostly invasive, infiltrating into surrounding tissues over a broader scope. The patients may require a total mastectomy, partial pectoral major muscle resection, axillary lymph node dissection, and postoperative chemoradiotherapy. In light of the above, the BC boundary is also closely correlated with Ki-67. Tumor morphology is another commonly used radiographic feature differentiating benign lesions from malignant ones. Benign lesions are regular in morphology and oval and circular in shape, typically with smooth boundaries. On the contrary, malignant lesions are morphologically irregular and lobulated and have burrs, displaying expansive growth. Crab foot-like changes are characteristic of BC. Some highly malignant cancers may even contain local liquefaction necrosis and infiltrate into the surrounding nerves. For this reason, tumor morphology is indirectly correlated with Ki-67. Elastography score is an evaluation method for tumor tissue hardness unique to ultrasound. Receiving acoustic energy reflected from the soft tissues, the ultrasound probe obtains the composition changes and hardness gradient of soft tissues and analyzes the hardness distribution and changes of the tumor. Tumor tissues are usually composed of more densely packed cells and have higher fiber content and a harder texture than normal tissues. Besides, tumor cells are more proliferative and self-regenerative than normal cells. The more proliferative the tumor cells, the higher the hardness of the tumor. Therefore, malignant tumors with a higher Ki-67 value have higher elastography scores and hardness.^21–25 Radscore, or radiomics score, was first proposed by Lambin, a Dutch scholar, in 2011. Radscore is a radiographic method for texture quantification and can dramatically improve radiographic diagnosis and prognostic evaluation of tumors. The high-order features selected in the present study adequately reflected the heterogeneity of BC. Patients with different Ki-67 values were significantly different in grayscale and texture features of ultrasound images. The tiny differences manifested by these high-order features are hardly discernible to the human eye on ultrasound images. Radscore proved to be highly predictive of Ki-67 in BC and has great application potential in this respect. In particular, the delta-radiomics extracted in this study well integrate the differences in tumor imaging changes before and after neoadjuvant chemotherapy, representing the chemotherapy response of the tumor reflecting its invading nature, and playing a decisive role in predicting Ki-67 expression. Based on the above-identified risk factors, a combined model was constructed on the training set. The model's predictive performance was significantly higher than that of the clinical model and radiomic model. The decision-making analysis also confirmed the high clinical net benefit of the combined model, and the result was verified on the test set. The nomogram tool developed based on the combined model achieved good clinical effect, as the nomogram successfully captured the differential biological properties of BC. This nomogram is useful for the clinical management and prognostic improvement of BC patients.^26–28 Compared with previous reports, the novelty of this study is to establish the prediction model by integrating clinical pathology and imaging data using the XGBoost algorithm. The prediction efficiency of the high-order algorithm often surpasses the logistic regression method, so the AUC value of the combined model in this study is higher than that in previous studies [0.857 vs 0.812].^29,30

Limitations

Firstly, the present study was conducted at a single center and only included a limited number of BC patients, which might cause biases in predictions. In the future, multicenter, large-sample data will be included to further improve the predictive performance and generalization capacity of the model. Secondly, our study was retrospective in nature, and the ultrasound images were collected by several ultrasound machines and sonographers. Variables might exist between the sonographers and ultrasound machines. It is necessary that we conduct a prospective study on the automated segmentation of BC images.^7,31,32

Conclusion

To conclude, the combined model integrating clinical and delta-radiomic features of BC patients displayed an excellent performance in predicting the Ki-67 value ≥ 15%. The XGBoost algorithm-based model is effective in identifying patients with a high Ki-67 expression, informing individualized clinical decision-making, and improving survival.

Footnotes

Abbreviations

Acknowledgments

Not applicable.

Author's Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis, and interpretation, or in all these areas; took part in drafting, revising, or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work. Lu, Yang, and Tao contributed equally to this work.

Data Availability

All data discussed in this article are included in the main manuscript text or Supplemental materials. The article data are available through the web link:

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Ethics Approval

This study was approved by the ethics committee of Central Hospital of Xiaogan (Issue XGYY20230091) and signed informed consent forms with patients or their families.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The “323” Public Health Project of the Hubei Health Commission and the Central Hospital of Xiaogan (XYY2022-323).Science and Technology Bureau of Xiaogan City, Natural Science Plan Project of Xiaogan City in 2023, Application of multimodal ultrasonography combined with lateral thoracic flap in breast-conserving plastic surgery for breast cancer (XGKJ2023010008-DaiPan).

ORCID iD

Pang An

References

Polat

Hulsey

Lewis

Dogan

. Radiomics analysis of contrast-enhanced breast MRI for optimized modelling of virtual prognostic biomarkers in breast cancer. Eur J Breast Health. 2024 Apr 1;20(2):122-128.

Hussain

Ali

Naseem

, et al. Breast cancer risk prediction using machine learning: a systematic review. Front Oncol. 2024 Mar 20;14:1343627.

Jin

Wang

Huang

. Clinical and radiomics integrated nomogram for preoperative prediction of tumor-infiltrating lymphocytes in patients with triple-negative breast cancer. Front Oncol. 2024 Mar 19;14:1370466.

Wang

Shang

Guo

, et al. Clinical study on the prediction of ALN metastasis based on intratumoral and peritumoral DCE-MRI radiomics and clinico-radiological characteristics in breast cancer. Front Oncol. 2024 Mar 19;14:1357145.

Huang

Luo

, et al. Photoacoustic imaging radiomics to identify breast cancer in BI-RADS 4 or 5 lesions. Clin Breast Cancer. 2024 Feb 29:S1526–8209(24)00055–7.

Lin

, et al. Ultrasound-based radiomics-clinical nomogram for noninvasive prediction of residual cancer burden grading in breast cancer. J Clin Ultrasound. 2024 Mar 27. doi: 10.1002/jcu.23666. Epub ahead of print. PMID: 38538081

von Elm

Altman

Egger

, et al. The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. Ann Intern Med. 2007;147(8):573-577. [published correction appears in Ann Intern Med. 2008 Jan 15;148(2):168. doi:10.7326/0003-4819-147-8-200710160-00010

Wang

Zhang

Guo

, et al. Prediction of lymphovascular space invasion in endometrial cancer based on multi-parameter MRI radiomics model. Curr Med Imaging. 2024 Mar 19. doi: 10.2174/0115734056266366231219111246. Epub ahead of print. PMID: 38529651

Aksu

Güç

Küçüker

Alacacıoğlu

Turgut

. Intra and peritumoral PET radiomics analysis to predict the pathological response in breast cancer patients receiving neoadjuvant chemotherapy. Rev Esp Med Nucl Imagen Mol (Engl Ed). 2024 Mar 23:500002. doi: 10.1016/j.remnie.2024.500002. Epub ahead of print. PMID: 38527731

10.

Liu

Leng

Liu

, et al. An ultrasound-based nomogram model in the assessment of pathological complete response of neoadjuvant chemotherapy in breast cancer. Front Oncol. 2024 Mar 4;14:1285511. doi: 10.3389/fonc.2024.1285511. PMID: 38500656; PMCID: PMC10946249

11.

Zuo

Zhou

Cai

. Prediction of the benign and malignant nature of masses in COPD background based on Habitat-based enhanced CT radiomics modeling: a preliminary study. Technol Health Care. 2024 Feb 29. doi: 10.3233/THC-231980. Epub ahead of print. PMID: 38517821

12.

Song

, et al. Exploring the role of modified vascular anatomical molding (MVAM) in prenatal diagnosis teaching and prognosis prediction of fetal complex congenital heart disease (CCHD): A preliminary study. Int J Gen Med. 2023 Jul 31;16:3229-3245. doi: 10.2147/IJGM.S421751. PMID: 37546241; PMCID: PMC10403051

13.

Wang

Song

Liu

. Prognostic modeling for nasopharyngeal carcinoma (NC) undergoing concurrent chemoradiotherapy using clinical and enhanced MRI-delta radiomics data: a preliminary study. Technol Health Care. 2024 Feb 29. doi: 10.3233/THC-231173. Epub ahead of print. PMID: 38517817

14.

Liu

Zou

, et al. Deep learning radiomics based prediction of axillary lymph node metastasis in breast cancer. NPJ Breast Cancer. 2024 Mar 12;10(1):22. doi: 10.1038/s41523-024-00628-4. PMID: 38472210; PMCID: PMC10933422

15.

Luo

Zhang

, et al. Prediction of lymphovascular invasion status in breast cancer based on magnetic resonance imaging radiomics features. Magn Reson Imaging. 2024 Mar 10;109:91-95. doi: 10.1016/j.mri.2024.03.008. Epub ahead of print. PMID: 38467265

16.

Yan

. Predictive value of background parenchymal enhancement on breast magnetic resonance imaging for pathological tumor response to neoadjuvant chemotherapy in breast cancers: a systematic review. Cancer Imaging. 2024 Mar 11;24(1):35. doi: 10.1186/s40644-024-00672-0. PMID: 38462607; PMCID: PMC10926651

17.

Huang

Chen

Sun

Liu

. Conditional generative adversarial network driven radiomic prediction of mutation status based on magnetic resonance imaging of breast cancer. J Transl Med. 2024 Mar 2;22(1):226. doi: 10.1186/s12967-024-05018-9. PMID: 38429796; PMCID: PMC10908206

18.

Deng

Liang

Yan

, et al. Development and validation of ultrasound-based radiomics model to predict germline BRCA mutations in patients with breast cancer. Cancer Imaging. 2024 Feb 29;24(1):31. doi: 10.1186/s40644-024-00676-w. PMID: 38424620; PMCID: PMC10905812

19.

Wang

Chen

Jiang

, et al. Intra- and peritumoral radiomics features based on multicenter automatic breast volume scanner for noninvasive and preoperative prediction of HER2 status in breast cancer: a model ensemble research. Sci Rep. 2024 Feb 29;14(1):5020. doi: 10.1038/s41598-024-55838-4. PMID: 38424285; PMCID: PMC10904744

20.

Hua

Mai

, et al. A multiparametric MRI-based radiomics model for stratifying postoperative recurrence in luminal B breast cancer. J Imaging Inform Med. 2024 Feb 29. doi: 10.1007/s10278-023-00923-9. Epub ahead of print. PMID: 38424277

21.

Qin

Duan

Xie

Zhou

Zhang

. Potential added value of computed tomography radiomics to multimodal prediction models for benign and malignant breast tumors. Transl Cancer Res. 2024 Jan 31;13(1):317-329. doi: 10.21037/tcr-23-1042. Epub 2024 Jan 15. PMID: 38410225; PMCID: PMC10894355

22.

Yang

, et al. Value of magnetic resonance imaging radiomics features in predicting histologic grade of invasive ductal carcinoma of the breast. Technol Health Care. 2024 Feb 7. doi: 10.3233/THC-230671. Epub ahead of print. PMID: 38393931

23.

Zhang

Yin

Zhou

, et al. Identification of ipsilateral supraclavicular lymph node metastasis in breast cancer based on LASSO regression with a high penalty factor. Front Oncol. 2024 Feb 2;14:1349315. doi: 10.3389/fonc.2024.1349315. PMID: 38371618; PMCID: PMC10869533

24.

Zheng

Yang

, et al. Discrimination between HER2-overexpressing, -low-expressing, and -zero-expressing statuses in breast cancer using multiparametric MRI-based radiomics. Eur Radiol. 2024 Feb 16. doi: 10.1007/s00330-024-10641-7. Epub ahead of print. PMID: 38363315

25.

Hacking

Windsor

Cooper

Jiao

Lourenco

Wang

. A novel approach correlating pathologic complete response with digital pathology and radiomics in triple-negative breast cancer. Breast Cancer. 2024 Feb 13. doi: 10.1007/s12282-024-01544-y. Epub ahead of print. PMID: 38351366

26.

Youn

Biswas

Hippe

, et al. Diagnostic performance of point-of-care apparent diffusion coefficient measures to reduce biopsy in breast lesions at MRI: clinical validation. Radiology. 2024 Feb;310(2):e232313. doi: 10.1148/radiol.232313. PMID: 38349238; PMCID: PMC10902596

27.

Wang

Lin

Ding

, et al. Multi-modality radiomics model predicts axillary lymph node metastasis of breast cancer using MRI and mammography. Eur Radiol. 2024 Feb 10. doi: 10.1007/s00330-024-10638-2. Epub ahead of print. PMID: 38337068

28.

Zhang

, et al. The emergence of the potential therapeutic targets: ultrasound-based radiomics in the prediction of human epidermal growth factor receptor 2-low breast cancer. Acad Radiol. 2024 Feb 2:S1076-6332(24)00023-0. doi: 10.1016/j.acra.2024.01.023. Epub ahead of print. PMID: 38309977

29.

Zhang

Niu

Chen

, et al. Radiomics signatures for predicting the Ki-67 level and HER-2 status based on bone metastasis from primary breast cancer. Front Cell Dev Biol. 2024;11:1220320. Published 2024 Jan 8. doi:10.3389/fcell.2023.1220320

30.

Tabnak

HajiEsmailPoor

Baradaran

Pashazadeh

Aghebati Maleki

. MRI-based radiomics methods for predicting Ki-67 expression in breast cancer: a systematic review and meta-analysis. Acad Radiol. 2024;31(3):763-787. doi:10.1016/j.acra.2023.10.010

31.

Chen

Lin

Kong

, et al. Non-invasive prediction model of axillary lymph node status in patients with early-stage breast cancer: a feasibility study based on dynamic contrast-enhanced-MRI radiomics. Br J Radiol. 2024 Feb 2;97(1154):439-450. doi: 10.1093/bjr/tqad034. PMID: 38308028

32.

Zhang

Wang

, et al. Ultrasound-based radiomics nomogram for predicting axillary lymph node metastasis in early-stage breast cancer. Radiol Med. 2024 Feb;129(2):211-221. doi: 10.1007/s11547-024-01768-0. Epub 2024 Jan 27. PMID: 38280058