Predictive value of a combined model for lymph node metastasis in NSCLC based on primary lesion radiomics from 18F-FDG PET/CT
10.3760/cma.j.cn371439-20241113-00022
- VernacularTitle:基于 18F-FDG PET/CT原发灶影像组学的联合模型预测NSCLC淋巴结转移的价值
- Author:
Ruihe LAI
1
;
Yue TENG
;
Jian RONG
;
Dandan SHENG
;
Yuzhi GENG
;
Jianxin CHEN
;
Chong JIANG
;
Chongyang DING
;
Zhengyang ZHOU
Author Information
1. 南京医科大学鼓楼临床医学院 南京鼓楼医院核医学科，南京 210008
- Keywords:
Carcinoma, non-small-cell lung;
Positron emission tomography computed tomography;
Radiomics;
Lymph node metastasis
- From:
Journal of International Oncology
2025;52(3):144-151
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To evaluate the value of a combined model based on primary lesion 18F-fluorodeoxyglucose ( 18F-FDG) PET/CT radiomics for predicting lymph node metastasis in non-small cell lung cancer (NSCLC) . Methods:A retrospective analysis was conducted on the clinical data of 203 NSCLC patients who underwent pre-treatment PET/CT imaging at Nanjing Drum Tower Hospital from June 2013 to July 2023. Patients were randomly assigned to the training set ( n=142) and the validation set ( n=61) at a ratio of 7∶3. A predictive model was developed in the training set, and its predictive performance and clinical application value were assessed in both the training and validation sets. Traditional PET/CT parameters and PET/CT radiomics features of the primary lesion were obtained by 3D-slicer software. Least absolute shrinkage and selection operator (LASSO), random forest, and extreme gradient boosting were performed to extract features. Support vector machine was used to construct a radiomics score (Radscore). Univariate and multivariate logistic regression analysis was used to predict the influencing factors of lymph node metastasis in NSCLC patients and to establish models. Predictive performance of the models was evaluated by receiver operator characteristic (ROC) curves and clinical application value was assessed by calibration curves and decision curve analysis (DCA) . Results:Among 203 NSCLC patients, 116 had lymph node metastasis, with 64 cases in the training set and 52 cases in the validation set. Three complementary classical machine learning methods were used for feature screening, and finally 10 radiomics features were obtained. The optimal threshold for Radscore-PET was 0.43 and the optimal threshold for Radscore-CT was 0.39. Univariate analysis showed that, sex ( OR=0.48, 95% CI: 0.24-0.95, P=0.036), tumor marker levels ( OR=3.81, 95% CI: 1.84-7.91, P<0.001), long diameter of tumor ( OR=2.56, 95% CI: 1.27-5.16, P=0.009), short diameter of tumor ( OR=3.73, 95% CI: 1.75-7.92, P=0.001), vacuolar sign ( OR=0.32, 95% CI: 0.12-0.86, P=0.024), ring-like metabolism ( OR=3.67, 95% CI: 1.33-10.13, P=0.012), maximum standardized uptake value (SUV max) ( OR=6.57, 95% CI: 3.03-14.25, P<0.001), metabolic tumor volume (MTV) ( OR=2.91, 95% CI: 1.43-5.92, P=0.003), total lesion glycolysis (TLG) ( OR=4.23, 95% CI: 2.08-8.59, P<0.001), Radscore-PET ( OR=21.93, 95% CI: 9.04-53.20, P<0.001) and Radscore-CT ( OR=13.72, 95% CI: 6.12-30.76, P<0.001) were all influencing factors for predicting lymph node metastasis in NSCLC patients. Multivariate analysis showed that, tumor marker levels ( OR=2.55, 95% CI: 1.11-5.90, P=0.028), vacuolar sign ( OR=0.26, 95% CI: 0.08-0.83, P=0.023), SUV max ( OR=5.94, 95% CI: 1.99-17.75, P=0.001), Radscore-PET ( OR=25.51, 95% CI: 5.92-110.22, P<0.001), and Radscore-CT ( OR=8.68, 95% CI: 2.73-27.61, P<0.001) were independent influencing factors for predicting lymph node metastasis in patients with NSCLC. Based on the above independent influencing factors, models were constructed: the traditional model (tumor marker levels, vacuolar sign, SUV max), the PET model (SUV max, Radscore-PET), the CT model (vacuolar sign, Radscore-CT), and the combined model (tumor marker levels, vacuolar sign, SUV max, Radscore-PET, Radscore-CT). ROC curve analysis showed that, the area under curve (AUC) of the traditional, PET, CT, and combined models in the training set were 0.75 (95% CI: 0.67-0.82), 0.90 (95% CI: 0.84-0.95), 0.85 (95% CI: 0.78-0.90), and 0.94 (95% CI: 0.88-0.97), respectively. The predictive value of the combined model was higher than that of the traditional model ( Z=5.01, P<0.001), the PET model ( Z=1.99, P=0.047), and the CT model ( Z=3.25, P=0.001). In the validation set, the AUCs for the traditional model, PET model, CT model, and combined model were 0.65 (95% CI: 0.52-0.77), 0.86 (95% CI: 0.74-0.93), 0.85 (95% CI: 0.73-0.93), and 0.90 (95% CI: 0.80-0.96), respectively. The predictive value of the combined model was superior to that of the traditional model ( Z=3.23, P=0.001). The sensitivity and specificity of the combined model in the training set were 84.37% and 91.03%, while in the validation set, the sensitivity and specificity were 82.61% and 94.74%, respectively. Calibration curves showed a good agreement between the predicted and actual probabilities in both the training and validation sets. DCA showed that the combined models had good discriminative ability in both the training and validation sets. Conclusions:Tumor marker levels, vacuolar sign, SUV max, Radscore-PET, and Radscore-CT are all independent influencing factors for predicting lymph node metastasis in patients with NSCLC. The combined model based on these factors demonstrates excellent predictive performance and clinical application value for predicting lymph node metastasis in NSCLC.
