Radiomics signature and clinical parameters of 18F-FDG PET/CT predicting progression-free survival in patients with breast cancer: a preliminary study

Introduction: This study aimed to investigate the feasibility of predicting progression-free survival (PFS) of breast cancer with radiomics feature built from ¹⁸F-FDG PET/CT images and clinical parameters before treatment.

Methods: We retrospectively analyzed breast cancer patients who underwent ¹⁸F-FDG PET/CT examinations from January 2012 to December 2020. Inclusion criteria were as follows: (1) pathological diagnosis was breast cancer and molecular subtypes was given; (2) no treatment before PET/CT imaging; (3) complete clinical data. Exclusion criteria included: (1) no pathological molecular subtypes; (2) chemotherapy, local puncture biopsy, surgery or other treatments before PET/CT imaging; (3) history of other malignancies; (4) blood glucose level was more than 11.1 mmol/L; (5) missing the follow-up data. Clinical parameters, including age, tumor size, molecular subtypes, initial TNM staging, and pretreatment tumor biomarkers (CEA, CA125 and CA153), were collected. Radiomic features were extracted from preoperative PET/CT images. The least absolute shrinkage and selection operators were applied to shrink feature size and build a predictive radiomic signature. Univariate and multivariate Cox proportional hazards model and Kaplan-Meier analysis were used to assess the association of the rad-score and clinical parameter with PFS. Nomogram was built to visualize Cox proportional hazards model for years survival prediction. C-index and calibration curves were used to evaluate the performance of the nomogram.

Results: After screening, 112 patients were included and randomly divided into training (n=61), internal test (n=26), and external validation (n=25) sets. A total of 11 radiomics features were selected to generate rad-score. Clinical-score was calculated by clinical model, which consisted of three clinical parameters (initial M staging, CA125, and pathological N staging).

In the training set, the rad-score and clinical-score were significantly associated with PFS (P=0.00081, P<0.0001, respectively), but there was no significant difference in the test set (P=0.26, P=0.13, respectively), which may suggested much heterogeneity in breast cancers. The integrated clinical-radiomics (ICR) model was significantly associated with PFS in both the training and test sets (P < 0.0001, P = 0.0034, respectively).

The ICR model nomogram estimated PFS (C-index, 0.845, 95% confidence interval [CI], 0.793-0.912 in training set and 0.758, 95% CI, 0.723-0.801 in test set) was better than the clinical model (C-index, 0.790, 95% CI, 0.754-0.872 in training set and 0.714, 95% CI, 0.632-0.774 in test set) or rad-score-only nomogram (C-index, 0.777, 95%CI, 0.712-0.833 in training set and 0.626, 95% CI, 0.597-0.755 in test set).

The performance of ICR model was further confirmed in the external validation set, with C-index of 0.754 (95% CI, 0.726-0.812). The Kaplan-Meier analysis showed a significant difference between the two groups stratified by the nomogram model (P=0.0092). Calibration curve also indicated the highest clinical benefit of the ICR model.

Conclusions: In this study, the integrated clinical-radiomics model from three clinical data (initial M staging, CA125, pathological N staging) and ¹⁸F-FDG PET/CT radiomics signature could independently predict the PFS of breast cancer patients, but the clinical model and radiomics signature alone could not be used as independent predictors. Larger sample study and more outside data to be verified are needed.

• Download figure
• Open in new tab
• Download powerpoint