Cross-Spatial-Scale Radiomics for Predicting Pathologic Complete Response to Neoadjuvant Chemotherapy in Breast...

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Pathologic complete response (pCR) is crucial for neoadjuvant chemotherapy (NAC) in breast cancer.
• Whole-tumor radiomics may obscure heterogeneous treatment responses.
• Accurate pretreatment prediction of pCR is needed to personalize treatment.

Purpose of the Study:

• To develop and validate TriSpace, an interpretable MRI biomarker for predicting pCR.
• To integrate global tumor, intratumoral, peritumoral, and clinicopathologic information.
• To assess TriSpace's performance against existing methods.

Main Methods:

• Retrospective analysis of 658 breast cancer patients undergoing pretreatment MRI and NAC.
• Development of TriSpace by fusing whole-tumor radiomics, intratumoral heterogeneity, peritumoral infiltration, and clinicopathologic features.
• External validation across four centers using AUC, calibration, and decision curve analysis.

Main Results:

• Intratumoral heterogeneity was the strongest predictor (external AUCs 0.813, 0.805).
• TriSpace achieved superior AUCs across all cohorts (0.946 to 0.815).
• SHAP analysis showed imaging features compensated for clinicopathologic shifts, maintaining discrimination across subgroups.

Conclusions:

• TriSpace offers an interpretable approach to enhance pretreatment pCR prediction.
• Integration of multi-scale tumor information improves prediction accuracy.
• Consistent external validation supports TriSpace's clinical utility.