Abstract
OBJECTIVES: The tumor-node-metastasis (TNM) staging system is vital for evaluating treatment efficacy in breast cancer patients undergoing neoadjuvant chemotherapy (NACT). However, the prognostic significance of longitudinal TNM changes remains unclear. This study aimed to develop radiomic signatures associated with TNM downstaging (dTNM) and to evaluate their utility in prognostic stratification for breast cancer patients. MATERIALS AND METHODS: The prognostic analysis included a development cohort (n = 292) and two external validation sets (n = 180, n = 61), all with DCE-MRI data and follow-up information. A random forest-based multitask model was developed using radiomic features from DCE-MRI to predict recurrence, pathological complete response (pCR), and dTNM in the development cohort, stratifying patients into distinct groups. The model's discriminative performance was assessed with the area under the curve (AUC). In the external validation set, a multivariable Cox proportional hazards model evaluated the prognostic significance of the groups stratified by the radiomic signatures. RESULTS: The multitask model, incorporating 17 imaging features, achieved AUCs of 0.905, 0.795, and 0.818 for predicting recurrence, pCR, and dTNM, respectively, in the inner validation set from the development dataset. External validation showed that, after adjusting for clinicopathological factors, the dTNM-related radiomic signatures were independently associated with better overall survival (OS) and recurrence-free survival (RFS) (p < 0.001 and p = 0.004, respectively). Furthermore, group stratification by radiomic signatures associated with pCR and dTNM demonstrated significant differences in survival (all p < 0.001) in both external validation datasets. CONCLUSION: Radiomic signatures of dTNM can be a prognostic indicator for survival outcomes in breast cancer. CRITICAL RELEVANCE: This study demonstrates that radiomic signatures associated with dTNM offer valuable prognostic insights for survival and recurrence outcomes. KEY POINTS: The imaging-based model can predict longitudinal TNM staging. Radiomic signatures of dTNM demonstrate significant prognostic value. dTNM-associated radiomic signatures provide better prognostic stratification than pCR.