Development of a 18F-FDG PET/CT-based Radiomics Model for Predicting Axillary Lymph Node Metastasis in Breast Cancer

doi:10.21203/rs.3.rs-5297857/v1

Background

Axillary lymph node metastasis (ALNM) status is an important factor for the determination of the therapeutic strategies and breast cancer prognosis. In our study, we investigate whether radiomics features from ¹⁸F-fluorodeoxyglucose(¹⁸F-FDG) positron emission tomography /computed tomography (PET/CT), combined with clinical or pathological characteristics, provide a higher predictive value of ALNM.

Methods

A retrospective analysis was performed on 78 female patients who underwent preoperative ¹⁸F-FDG PET/CT scans at Jinhua Central Hospital from August 2015 to July 2024, with a mean age of 53.60 ± 12.49 years (range: 35–84 years). The cases were randomly divided into a training cohort (46 cases) and a testing cohort (32 cases) in a 6:4 ratio. All patients' PET/CT and clinical pathological features were analyzed, and radiomics features were extracted from the PET/CT images. Subsequently, we developed radiomics, clinical, and combined radiomics-clinical models. We also assessed the performance of these three models in predicting ALNM. The Python stats models package (version 0.13.2) was used for statistical analysis.

Results

For the three features radiomics model and combined model in the training cohort, the area under the curve (AUC) was 0.922 and 0.931, which were both higher than that of the traditional clinical feature model (AUC = 0.917). The AUC values for the three models in the testing cohort were 0.802, 0.821, and 0.778. For predicting ALNM across all cohorts, the radiomics model and the combined model showed clinical benefit in the decision curve analysis (DCA).

Conclusion

The PET/CT-based radiomics model demonstrated strong efficacy in predicting ALNM for breast cancer and has clinical application value.

Breast cancer

PET/CT

Radiomics

Axillary lymph node metastasis

Breast cancer is one of the leading causes of cancer-related mortality and the most common cancer among female patients. According to global cancer statistics, its incidence among female patients is higher than the incidence of lung cancer[1]. Breast cancer’s most common metastatic pathway is ALNM. For example, the 5-year survival rate is 98.8%, for patients with localized lesions confined to the breast, while it decreases to 85.8% for those with ALNM [2]. The treatment plan depends on the presence or absence of ALNM. Currently, ALNM is determined by sentinel lymph node biopsy (SLNB) and axillary lymph node dissection (ALND). Both procedures are invasive and may result in various adverse effects, like infection, sensory abnormalities, upper limb mobility issues, and lymphedema, while also possessing a certain false-negative rate. However, compared to ALND, SLNB is less invasive and has a lower incidence of adverse effects, and it has gradually replaced ALND in clinical practice[3]. The treatment guidelines from the American Medical Association also recommend that no further ALND is necessary when the SLNB result is negative[4]. Although SLNB has become the primary approach for clinically assessing ALNM, studies have shown that the negative rate of SLNB can reach 70%, with a complication rate of 41% post-SLNB[5]. Therefore, non-invasive evaluation of ALNM before treatment holds significant clinical importance.

The preoperative assessment of ALNM in clinical practice primarily relies on imaging examinations, including breast ultrasound, mammography, etc[6]. However, the sensitivity of these examinations for ALNM remains relatively low. In contrast to other imaging modalities, PET/CT combines the high-resolution anatomical localization of CT with the high sensitivity and specificity of metabolic imaging from PET, making it widely used for non-invasive diagnosis at the molecular imaging level[7].

Currently, nuclear medicine physicians commonly use maximum standardized uptake value (SUVmax) as a critical criterion for assessing the malignancy of lesions, with higher SUVmax values generally indicating a greater likelihood of malignancy. This assessment is often supplemented by other imaging characteristics of the lesions, such as morphology, volume, and density[7]. Numerous studies have shown a correlation between SUVmax and tumor malignancy[8][9]. However, the accuracy of these assessments can be significantly affected by image quality and subjective factors. In cases where lesions are small (< 0.5 cm), exhibit high differentiation, or grow slowly, the likelihood of false negatives increases.

Radiomics is a rapidly evolving research field that extracts a vast array of radiomics features from high-throughput imaging data. It offers richer imaging information, thereby reducing the limitations of subjective interpretation by physicians and enhancing the quality of medical imaging diagnosis. Research based on PET/CT radiomics is still scarce, despite the fact that several studies have used ultrasonography, CT, and MRI radiomics to predict ALNM in breast cancer[10][11]. The purpose of this study is to investigate the predictive usefulness of preoperative ¹⁸F-FDG PET/CT radiomics for breast cancer ALNM, offering more data to improve clinical diagnosis and guide treatment decisions.

Patient

Since this is a retrospective analysis, Jinhua Central Hospital's Ethics Committee, which is connected to Zhejiang University, has waived the need for patient-informed consent. A retrospective collection of pre-treatment ¹⁸F-FDG PET/CT imaging data, along with clinical and pathological data, was conducted for breast cancer patients confirmed by pathology at Jinhua Hospital affiliated with Zhejiang University, covering the period from August 5, 2015, to July 25, 2024. Data were rigorously screened according to inclusion and exclusion criteria. The patient selection process is illustrated in Fig. 1 .

Inclusion criteria: 1) No treatment prior to PET/CT examination, including surgery, radiotherapy, chemotherapy, and immunotherapy; 2) Both the primary breast cancer lesion and ALNM lesions were confirmed by pathology; 3) Complete clinical and pathological information.

Exclusion criteria: 1) Individuals who were on anticancer medication prior to PET/CT scanning; 2) Inadequate clinical and pathological information; 3) Existence of more cancerous tumors; 4) Poor image quality of PET/CT, affecting interpretation of results.

A training cohort and a testing cohort were randomly selected from the dataset in a 6:4 ratio. Instances from the testing cohort were used for an impartial assessment of the prediction model's performance, while all instances from the training cohort were used to train the model.

PET/CT Scanning Technology

All enrolled patients underwent scanning using the Siemens PET/CT-Biograph mCT. Prior to the examination, patients were told to fast for 6–8 hours, and their height, weight, and blood glucose levels were measured. An intravenous injection of the ¹⁸F-FDG imaging agent was administered at a dose of 0.1–0.15 mCi/kg, ensuring that the patient's fasting blood glucose level was below 11.1 mmol/L prior to injection. The patient was then instructed to rest quietly for 1 hour before undergoing PET/CT scanning.

The following were the CT scan parameters: 120 kV tube voltage and a tube current automatic mAs technology, the slice thickness is 3 mm, pitch is 0.8, and rotation duration is 0.5 s/r. The PET scan parameters were as follows: 1.5 min/bed, with 6 to 8 bed positions, and the slice thickness of 3 mm. For post-processing, the scanned photos are uploaded to the imaging workstation.

Image Assessment

The images are evaluated by two radiologists at our center, each with over three years of diagnostic experience, who are kept unaware of the patients' clinical information and pathology results. The evaluation criteria are as follows: A higher uptake of the radiopharmaceutical in breast tissue compared to surrounding areas indicates the presence of breast cancer, while an increased uptake in lymph nodes compared to adjacent muscle tissue suggests lymph node metastasis. A semi-automated delineation of the region of interest (ROI) is performed based on a 40% SUVmax threshold, measuring PET metabolic parameters, including SUVmax, minimum standardized uptake value (SUVmin), average standardized uptake value (SUVavg), metabolic tumor volume (MTV), and total lesion glycolysis (TLG). Additionally, the maximum diameter of the lesion in the axial plane is measured. In cases with multiple lesions, only the largest lesion by volume is selected for measurement.

Radiomics

Data Collection: Export the raw DICOM data from the "Medex" platform. Image Segmentation: Using the 3D Slicer software (version 5.6.1), select the "Draw" tool to manually delineate the ROI layer by layer on the PET/CT fused images, taking care to avoid necrotic areas. Perform radiomics feature extraction on the segmented ROI using the "Radiomics" package for both PET and CT images. Figure 2 displays the radiomics analysis workflow used in this investigation. Three groups of radiomics features can be distinguished: (I) geometry, (II) intensity, and (III) texture. Tumors are characterized by geometric features in three dimensions. The first-order statistical distribution of voxel intensities inside the tumor is described by intensity characteristics. The higher-order spatial distribution of the intensity patterns is described by the texture features. Z-score normalization is used to address the issue of scale variation in manually extracted radiomics features.

Using a double-blind technique, two nuclear medicine doctors separately segmented each lesion. The intraclass correlation coefficient (ICC) was used to assess the stability of feature extraction; an ICC > 0.75 indicates strong consistency in feature extraction.

Clinical and Pathological Characteristics

Clinical information were recorded for all included patients, including age, body mass index (BMI), menstrual status, tumor T stage, carcinoembryonic antigen (CEA) levels, carbohydrate antigen 125 (CA125) levels, and carbohydrate antigen 153 (CA153) levels. Additionally, the following pathological and immunohistochemical data were recorded: estrogen receptor (ER) status, progesterone receptor (PR) status, human epidermal growth factor receptor 2 (HER-2) status, tumor cell proliferation marker 67 (Ki-67) levels, and molecular subtypes of breast cancer. ER positivity is defined as ≥ 10% of tumor cell nuclei staining positive; PR positivity is defined as ≥ 10% of tumor cell nuclei staining positive, with ≥ 20% indicating high expression, and lower percentages indicating low expression; HER-2 positivity is defined as an immunohistochemical score of ≥ 2 + or positive FISH amplification; Ki-67 expression index ≥ 14% indicates high expression, while lower percentages indicate low expression. According to the 2023 NCCN guidelines[12], breast cancer molecular subtyping is determined based on immunohistochemical results, specifically: luminal A, luminal B, HER-2 positive, and triple-negative.

Feature Selection and Model Construction

A Mann-Whitney U test was run on all radiomics characteristics, preserving only those with a p-value < 0.05. The correlation between characteristics with high redundancy was determined using Pearson's rank correlation coefficient, which kept features with a correlation coefficient of at least 0.9 between any two features. For additional feature selection, the Least Absolute Shrinkage and Selection Operator (LASSO) was utilized. To find the ideal λ, 10-fold cross-validation was utilized, keeping features with non-zero coefficients. Finally, the Rad score was calculated by summing the products of the LASSO-Logistic regression coefficients and their corresponding values. This study also compared the stability and reliability of seven machine learning algorithms in predicting ALNM, including Logistic Regression (LR), Naive Bayes, K-Nearest Neighbors (KNN), Decision Tree, Random Forest (RF), Extra Trees, and Extreme Gradient Boosting (XGBoost). The diagnostic performance of the optimal predictive model was evaluated using the Receiver Operating Characteristic (ROC) curve.

Building the clinical model follows a nearly identical procedure to that of building the radiomics model. First, statistical variables with a p-value < 0.05 were selected from the clinical baseline data, and a clinical model was constructed using the optimal machine learning algorithm, incorporating 5-fold cross-validation and a fixed testing cohort. Finally, a nomogram was constructed based on logistic regression analysis by integrating radiomics features and clinical characteristics, and the ROC curve and calibration curve were plotted to evaluate its performance. The clinical utility was evaluated using DCA.

Statistical Analysis

The Python statsmodels module (version 0.13.2) was utilized for statistical analysis, and a p-value < 0.05 was considered statistically significant. For continuous variables, the Mann-Whitney U test was utilized to examine intergroup differences; for categorical variables, the χ2 test was employed.

Clinical Feature

This study included a total of 78 pathologically confirmed female breast cancer patients, with an average age of 53.60 ± 12.49 year. Among them, 57 patients had positive ALNM, while 21 had negative results. The subjects were randomly allocated into a training cohort (46 instances) and a testing group (32 cases) in a 6:3 ratio. The training cohort included 34 cases with positive ALNM and 12 cases with negative results, while the testing cohort comprised 23 positive cases and 9 negative cases. The participants in our cohort's baseline clinical characteristics are displayed in Table. 1. Age, BMI, tumor markers, MTV, TLG, menstrual status, and immunohistochemical indicators showed no significant differences between two cohorts (P > 0.05), while tumor diameter, SUVmax, SUVmin, SUVavg, T staging, and breast cancer molecular subtypes exhibited significant differences (P < 0.05).

Table 1

Characteristics of patients in our cohorts.
Features	ALL(n = 87)	Non-Metastasis(n = 30)	Metastasis(n = 57)	P-value
Age	53.60 ± 12.49	56.52 ± 16.22	52.53 ± 10.78	0.379
BMI	22.57 ± 3.59	22.81 ± 4.25	22.48 ± 3.36	0.892
Tumor diameter	31.09 ± 19.11	21.01 ± 16.66	34.80 ± 18.73	< 0.001
Menstrual status				0.891
0	40(51.28)	10(47.62)	30(52.63)
1	38(48.72)	11(52.38)	27(47.37)
T stage				< 0.001
1	18(23.08)	14(66.67)	4(7.02)
2	41(52.56)	6(28.57)	35(61.40)
3	11(14.10)	Null	11(19.30)
4	8(10.26)	1(4.76)	7(12.28)
CEA	37.08 ± 100.61	18.93 ± 44.65	43.76 ± 114.18	0.461
CA125	92.73 ± 301.16	95.99 ± 362.67	91.54 ± 278.81	0.161
CA153	26.70 ± 53.65	15.02 ± 22.34	31.00 ± 60.90	0.226
ER				1
0	24(30.77)	6(28.57)	18(31.58)
1	54(69.23)	15(71.43)	39(68.42)
PR				0.751
0	45(57.69)	11(52.38)	34(59.65)
1	33(42.31)	10(47.62)	23(40.35)
HER2				0.983
0	24(30.77)	7(33.33)	17(29.82)
1	54(69.23)	14(66.67)	40(70.18)
Ki67				1
0	17(21.79)	5(23.81)	12(21.05)
1	61(78.21)	16(76.19)	45(78.95)
Mol-subtypes				0.27
1	2(2.56)	Null	2(3.51)
2	17(21.79)	4(19.05)	13(22.81)
3	35(44.87)	11(52.38)	24(42.11)
4	19(24.36)	3(14.29)	16(28.07)
5	5(6.41)	3(14.29)	2(3.51)
SUVmax	10.17 ± 6.12	6.82 ± 4.35	11.41 ± 6.25	0.001
SUVmin	4.32 ± 2.44	2.43 ± 1.54	5.02 ± 2.34	< 0.001
SUVavg	6.19 ± 3.70	4.24 ± 2.68	6.91 ± 3.79	0.003
MTV	16.46 ± 37.50	5.98 ± 6.97	20.32 ± 43.12	0.032
TLG	147.58 ± 443.00	31.20 ± 44.44	190.46 ± 512.04	0.004

Radiomics Features Selection and Model Establishment

Using the 3D Slicer program, 1,702 radiomics characteristics in total were retrieved. There were 246 features left after all features were first screened using the Mann-Whitney U test. Following that, a Pearson correlation analysis was carried out, yielding 65 residual characteristics. Finally, LASSO regression with an optimal λ of 0.1048 identified three non-zero coefficients for the radiomics features, establishing the Rad score(Fig. 3).

The following formula is used to determine the Rad score:

0.739130434782608 + 0.02911 * wavelet_LHL_glszm_SizeZoneNonUniformityNormalized_CT + 0.047979 * wavelet_LLL_glcm_SumAverage_CT − 0.107920 * wavelet_HHL_glcm_JointAverage_PET

This study compared predictive models established using seven different machine learning algorithms. The results indicated that the radiological model constructed using the LR method exhibited the best stability and reliability (Table. 2). Ultimately, the radiomics model based on LR was selected to develop the subsequent nomogram.

Table 2

Performance of different machine learning models in predicting ALNM in both cohorts.
Model	Accuracy	AUC	95%CI	Sen	Spe	PPV	NPV
LR-train	0.891	0.922	0.8347–1.0000	0.912	0.833	0.939	0.769
LR-test	0.719	0.802	0.6497–0.9542	0.609	1	1	0.5
Naïve Bayes-train	0.87	0.892	0.7840–1.0000	0.882	0.833	0.937	0.714
Naïve Bayes-test	0.688	0.787	0.6298–0.9451	0.565	1	1	0.474
KNN-train	0.761	0.93	0.8611–0.9992	0.676	1	1	0.522
KNN-test	0.688	0.72	0.5164–0.9232	0.696	0.667	0.842	0.462
Decision Tree-train	0.261	0.985	0.9650–1.0000	0	1	0	0.261
Decision Tree-test	0.281	0.684	0.4892–0.8779	0	1	0	0.281
Random Forest-train	0.913	0.988	0.9660–1.0000	0.882	1	1	0.75
Random Forest-test	0.531	0.804	0.6539–0.9548	0.348	1	1	0.375
Extra Trees-train	0.935	0.98	0.9489–1.0000	0.941	0.917	0.97	0.846
Extra Trees-test	0.656	0.751	0.5824–0.9200	0.522	1	1	0.45
XG Boost-train	0.978	1	1.0000–1.0000	0.971	1	1	0.923
XG Boost-test	0.656	0.645	0.4072–0.8826	0.609	0.778	0.875	0.437

Clinical Features Selection and Model Establishment

A predictive model was developed using the clinical features of patients in the training cohort, with the univariate and multivariate analysis results for 46 breast cancer patients presented in Table. 3. The results of the univariate logistic regression analysis indicated that tumor diameter, SUVmax, SUVmin, SUVavg, and tumor T stage are risk factors for ALNM. The creation of the clinical model took into account the findings of further multivariate logistic regression analysis, which showed that SUVmin (OR = 1.175, P < 0.05) and tumor T stage (OR = 1.246, P < 0.05) are independent risk factors for ALNM.

Table 3

Logistic regression analysis of clinical data in the training cohort.
Feature	Univariate analysis		Multivariate analysis
Feature	OR (95% CI)	P-value	OR (95% CI)	P-value
Tumor diameter	1.010(1.004–1.016)	0.009	1.004(0.989–1.018)	0.67
T stage	1.317(1.188–1.461)	0	1.246(1.051–1.477)	0.035
SUVmax	1.022(1.005–1.049)	0.035	3.026(0.355–25.842)	0.389
SUVavg	1.034(1.006–1.062)	0.043	0.569(0.14–2.316)	0.502
SUVmin	1.083(1.043–1.124)	0.001	1.175(1.06–1.303)	0.012

Predictive Performance of Different Models

Three models were used to assess the prediction accuracy for ALNM: the clinical model, the radiomics model, and the combination model. The training cohort's AUCs were 0.917, 0.922, and 0.931, while the testing cohorts were 0.778, 0.802, and 0.821 (Fig. 4). In both cohorts, the combined model's AUC was found to be significantly higher by the DeLong test than by the clinical and radiomics models (Table. 4). Ultimately, a nomogram was created by merging the extremely stable and dependable Rad score derived from logistic regression with clinical prognostic criteria (Fig. 5). The calibration curve showed that the combined model fit well to forecast the spread of breast cancer to the axillary lymph nodes (Fig. 6). Furthermore, the DCA showed that there is a substantial net advantage offered by the combined approach (Fig. 7).

Table 4

Delong test.
	Cohort	Combined Vs Clinic	Combined Vs Rad	Rad Vs Clinic
P-value	Train	0.016	0.027	0.927
P-value	Test	0.027	0.026	0.831

As one of the most prevalent cancers in females, breast cancer affects a younger and younger population globally each year in terms of incidence [13]. Precise determination of lymph node staging in breast cancer is essential for choosing therapeutic treatment strategies and forms a significant foundation for prognostic assessment. The primary method for assessing ALNM in breast cancer is SLNB; however, this procedure may result in postoperative complications and impact recovery. A non-invasive method for forecasting the metastases of axillary lymph nodes is provided by radiomics. Radiomics provides a non-invasive means of predicting metastases from axillary lymph nodes by supplying features from images that are not visible through conventional imaging assessment methods. These features are often linked to genomic, cellular, and metabolic information relevant to tumor biology[14, 15]. Some studies have utilized MRI radiomics to predict pelvic lymph node metastasis in cervical cancer, reporting AUC values of 0.82, 0.84, and 0.9 [16–18]. A meta-analysis using ultrasound radiomics has also been used in papers summarizing prognostic research on cervical lymph node metastasis in thyroid cancer, with a sensitivity value of 0.81 (95% CI, 0.73–0.86) and a specificity value of 0.87 (95% CI, 0.83–0.91) [19]. Zhao et al.'s[20] study, which had an AUC of 0.91, showed how CT radiomics analysis might improve preoperative prediction of cervical lymph node metastasis in laryngeal squamous cell carcinoma. The aforementioned studies indicate that conventional radiomics demonstrates a strong predictive capability for lymph node metastasis in tumors.

This work developed a hybrid model that uses ¹⁸F-FDG PET/CT radiomics and clinical variables to predict the metastasis of axillary lymph nodes in breast cancer. Independent validation with a testing cohort confirmed the model's strong predictive capability. This study included five metabolic parameters from PET/CT in the clinical characteristics. Statistical analysis indicated that SUVmin demonstrated a strong predictive capability within our cohort. In related studies on PET/CT radiomics, various metabolic parameters have frequently shown significant predictive abilities. For instance, research by Moazemi et al.[21] revealed that the SUVmin from pre-treatment ⁶⁸Ga-PSMA PET/CT may serve as a potential predictor of overall survival in patients with advanced prostate cancer undergoing ¹⁷⁷Lu-PSMA therapy. This may be attributed to the fact that low SUV values can indicate a certain level of tumor heterogeneity, which may be associated with a broader spectrum of activity values within the tumor.

This study extracted a total of 1,702 radiomics features through data analysis. After conducting differential analysis and LASSO regression, three radiomics features were selected for model construction. These features are closely related to the regional texture, gray values, contrast, and frequency domain characteristics of breast cancer lesions. They exhibit partial similarity to the radiomics features used in the comprehensive predictive model developed by Li et al.[22], suggesting that PET/CT imaging manifestations of different primary breast cancer lesions present various subtle differences.

The combined model in this investigation did well in terms of predictive power. In the training cohort, the AUC for predicting the metastasis of axillary lymph nodes was 0.931, whereas in the testing cohort, it was 0.821. Prior research has demonstrated that conventional clinical characteristics can forecast the ALNM of breast cancer. However, compared to models relying on SUVmin and tumor T staging clinical indicators, the radiomic model demonstrated higher accuracy in both cohorts. This suggests that radiomics may possess a superior ability to identify tumor heterogeneity in breast cancer. We created a combined predictive model to assess whether clinical factors could improve the radiomics model's performance even further. According to the results, the combined model's prediction capacity outperformed the radiomics model alone in both cohorts, according to the DeLong test. This indicates that the combined model can integrate both types of information, capturing a more comprehensive set of features and thereby improving predictive accuracy.

One of the study's limitations is that it is a retrospective study conducted at a single location, which raises the possibility of selection bias. Furthermore, because of the small sample size, it's possible that several indicators that revealed differences in univariate analysis won't reach statistical significance in the cohort's multivariate analysis. Finally, this study did not exclude patients with multifocal breast cancer. Future multicenter, large-sample, prospective studies are expected to improve the treatment of breast cancer by further validating the prognostic usefulness of ¹⁸F-FDG PET/CT imaging for ALNM.

The multivariable model developed from ¹⁸F-FDG PET/CT radiomics and clinical features demonstrates strong predictive performance for ALNM in breast cancer. This model is expected to serve as a reference for personalized precision treatment decisions in clinical practice.

ALNM	Axillary lymph node metastasis
¹⁸F-FDG	¹⁸F-fluorodeoxyglucose
PET/CT	Positron emission tomography /computed tomography
AUC	Area under the curve
DCA	Decision curve analysis
SLNB	Sentinel lymph node biopsy
ALND	Axillary lymph node dissection
ROI	Region of interest
SUVmax	Maximum standardized uptake value
SUVmin	Minimum standardized uptake value
SUVavg	Average standardized uptake value
MTV	Metabolic tumor volume
TLG	Total lesion glycolysis
ICC	Intraclass correlation coefficient
BMI	Body mass index
CEA	Carcinoembryonic antigen
CA125	Carbohydrate antigen 125
CA153	Carbohydrate antigen 153
ER	Estrogen receptor
PR	Progesterone receptor
HER-2	Human epidermal growth factor receptor 2
Ki-67	Tumor cell proliferation marker 67
LASSO	Least absolute shrinkage and selection operator
LR	Logistic Regression
KNN	K-Nearest Neighbors
RF	Random Forest
XGBoost	Extreme Gradient Boosting
ROC	Receiver operating characteristic

Ethical Approval

This retrospective study was approved by the Biomedical Ethics Committee of Jin Hua Municipal Central Hospital, which waived the requirement for patient informed consent.

Funding

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This research was funded by Zhejiang Province Public Welfare Technology Application Research Project (LGF21H180007).

Author Contribution

ZY: Conceptualization, Data curation & analysis, Writing – original draft. KD: Funding acquisition, Resources, Software, Writing – original draft. QH: Project administration, Supervision, Validation, Writing – review & editing. All authors participated in the revision and improvement of the manuscript to ensure the accuracy and completeness of the content.

Acknowledgement

We thank all the participants and all the researchers and collaborators who participated in this study.

Sung, H, Ferlay, J, Siegel, RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA-CANCER J CLIN. 2021; 71 CA-CANCER J CLIN.
Chang JM, Leung JWT, Moy L, Ha SM, Moon WK. Axillary Nodal Evaluation in Breast Cancer: State of the Art. Radiology. 2020 Jun;295(3):500-515.
Giuliano AE, Ballman KV, McCall L, Beitsch PD, Brennan MB, Kelemen PR, Ollila DW, Hansen NM, Whitworth PW, Blumencranz PW, Leitch AM, Saha S, Hunt KK, Morrow M. Effect of Axillary Dissection vs No Axillary Dissection on 10-Year Overall Survival Among Women With Invasive Breast Cancer and Sentinel Node Metastasis: The ACOSOG Z0011 (Alliance) Randomized Clinical Trial. JAMA. 2017 Sep 12;318(10):918-926.
Lyman GH, Somerfield MR, Bosserman LD, Perkins CL, Weaver DL, Giuliano AE. Sentinel Lymph Node Biopsy for Patients With Early-Stage Breast Cancer: American Society of Clinical Oncology Clinical Practice Guideline Update. J Clin Oncol. 2017 Feb 10;35(5):561-564. doi: 10.1200/JCO.2016.71.0947. Epub 2016 Dec 12.
Zhang X, Liu M, Ren W, Sun J, Wang K, Xi X, Zhang G. Predicting of axillary lymph node metastasis in invasive breast cancer using multiparametric MRI dataset based on CNN model. Front Oncol. 2022 Dec 6;12:1069733.
Cömert, D, van Gils, CH, Veldhuis, WB, et al. Challenges and Changes of the Breast Cancer Screening Paradigm. J MAGN RESON IMAGING. 2023; 57 J MAGN RESON IMAGING.
Farwell, MD, Pryma, DA, Mankoff, DA. PET/CT imaging in cancer: current applications and future directions. CANCER-AM CANCER SOC. 2014; 120 (22): 3433-45.
Niu, R, Gao, J, Shao, X, et al. Maximum Standardized Uptake Value of 18F-deoxyglucose PET Imaging Increases the Effectiveness of CT Radiomics in Differentiating Benign and Malignant Pulmonary Ground-Glass Nodules. Front Oncol. 2021; 11 Front Oncol.
de Koster EJ, Noortman WA, Mostert JM, Booij J, Brouwer CB, de Keizer B, de Klerk JMH, Oyen WJG, van Velden FHP, de Geus-Oei LF, Vriens D; EfFECTS trial study group. Quantitative classification and radiomics of [¹⁸F]FDG-PET/CT in indeterminate thyroid nodules. Eur J Nucl Med Mol Imaging. 2022 Jun;49(7):2174-2188. doi: 10.1007/s00259-022-05712-0. Epub 2022 Feb 9.
Li Y, Han D, Shen C. Prediction of the axillary lymph-node metastatic burden of breast cancer by ¹⁸F-FDG PET/CT-based radiomics. BMC Cancer. 2024 Jun 7;24(1):704.
Li, X, Yang, L, Jiao, X. Comparison of Traditional Radiomics, Deep Learning Radiomics and Fusion Methods for Axillary Lymph Node Metastasis Prediction in Breast Cancer. ACAD RADIOL. 2023; 30 ACAD RADIOL.
Gradishar, WJ, Moran, MS, Abraham, J, et al. Breast Cancer, Version 3.2024, NCCN Clinical Practice Guidelines in Oncology. J NATL COMPR CANC NE. 2024; 22 (5): 331-357.
Giaquinto, AN, Sung, H, Newman, LA, et al. Breast cancer statistics 2024. CA-CANCER J CLIN. 2024; CA-CANCER J CLIN.
Lee, G, Lee, H, Ko, E, et al. Radiomics and imaging genomics in precision medicine PRECIS FUTURE MED. 2024; 1 PRECIS FUTURE MED
Lee, G, Lee, HY, Park, H, et al. Radiomics and its emerging role in lung cancer research, imaging biomarkers and clinical management: State of the art. EUR J RADIOL. 2016; 86 297-307.
Song, J, Hu, Q, Ma, Z, et al. Feasibility of T2WI-MRI-based radiomics nomogram for predicting normal-sized pelvic lymph node metastasis in cervical cancer patients. EUR RADIOL. 2021; 31 (9): 6938-6948.
Wang, T, Li, YY, Ma, NN, et al. A MRI radiomics-based model for prediction of pelvic lymph node metastasis in cervical cancer. World J Surg Oncol. 2024; 22 (1): 55.
Xia, X, Li, D, Du, W, et al. Radiomics Based on Nomogram Predict Pelvic Lymph node Metastasis in Early-Stage Cervical Cancer. Diagnostics (Basel). 2022; 12 (10):
Mu, J, Cao, Y, Zhong, X, et al. Prediction of cervical lymph node metastasis in differentiated thyroid cancer based on radiomics models. BRIT J RADIOL. 2024; 97 BRIT J RADIOL.
Zhao, X, Li, W, Zhang, J, et al. Radiomics analysis of CT imaging improves preoperative prediction of cervical lymph node metastasis in laryngeal squamous cell carcinoma. EUR RADIOL. 2023; 33 EUR RADIOL.
Moazemi, S, Erle, A, Lütje, S, et al. Estimating the Potential of Radiomics Features and Radiomics Signature from Pretherapeutic PSMA-PET-CT Scans and Clinical Data for Prediction of Overall Survival When Treated with 177Lu-PSMA. Diagnostics (Basel). 2021; 11 (2):
Li, Y, Han, D, Shen, C, et al. Construction of a comprehensive predictive model for axillary lymph node metastasis in breast cancer: a retrospective study. BMC Cancer. 2023; 23 BMC Cancer.

No competing interests reported.

Development of a ¹⁸F-FDG PET/CT-based Radiomics Model for Predicting Axillary Lymph Node Metastasis in Breast Cancer

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Background

Methods

Statistical Analysis

Results

Clinical Feature

Radiomics Features Selection and Model Establishment

Clinical Features Selection and Model Establishment

Predictive Performance of Different Models

Discussion

Conclusion

Abbreviations

Declarations

Funding

Author Contribution

Acknowledgement

References

Additional Declarations

Status:

Version 1