Prediction of pathologic complete response in breast cancer patients undergoing neoadjuvant chemotherapy using baseline 18F-FDG PET/CT-derived CT radiomics and clinicopathological biomarkers.
TL;DR
Combining CT radiomics and SUVmax with clinicopathological biomarkers yielded the highest numerical discrimination for pCR, but improvement over the Clinical model was modest.
Abstract
Objective
The study aimed to assess whether baseline fluorine-18 fluorodeoxyglucose (18F-FDG) PET/computed tomography (CT)-derived CT radiomics adds predictive value to clinicopathological biomarkers and PET metabolic assessment for pathologic complete response (pCR) after neoadjuvant chemotherapy (NAC) in breast cancer.
Methods
This retrospective study included 104 patients with stage I-III breast cancer who underwent baseline 18F-FDG PET/CT, NAC, and surgery. Clinicopathological variables and tumor maximum standardized uptake value (SUVmax) were recorded; 107 CT radiomic features were extracted using PyRadiomics. Clinical, Metabolic, Radiomics, and Combined models were developed using L2-regularized logistic regression and evaluated by nested five-fold cross-validation. Discrimination, calibration, and decision curve analysis were assessed; 95% confidence intervals (CIs) were estimated using 1000 bootstrap resamples of pooled out-of-fold predictions.
Results
Forty-six patients (44.2%) achieved pCR. Area under the curves (AUCs) were 0.784 (95% CI: 0.686-0.875) for Clinical, 0.595 (0.470-0.704) for Metabolic, and 0.636 (0.526-0.739) for Radiomics models. The Combined model had the highest numerical AUC (0.797; 95% CI: 0.705-0.880), with accuracy 0.731, sensitivity 0.630, specificity 0.810, and F1-score 0.674.
Conclusion
Combining CT radiomics and SUVmax with clinicopathological biomarkers yielded the highest numerical discrimination for pCR, but improvement over the Clinical model was modest. Larger multicenter external validation is needed before clinical implementation.