A Simple and Robust MRI Radiomics Feature for Predicting Pathological Complete Response: A Proof-of-Concept Study in...

Early prediction of pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer remains challenging. This study aimed to explore the value of a radiomics model based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) acquired after the second cycle of NAC for early prediction of pCR.

A retrospective analysis was conducted on 119 breast cancer patients who underwent NAC at our hospital between March 2020 and August 2023. Patients were categorized into pCR and non-pCR groups based on postoperative Miller-Payne pathological grading as the gold standard. Tumor regions of interest (ROIs) were manually delineated on phase-three DCE-MRI sequences. PyRadiomics extracted 851 features. A rigorous dimensionality reduction process-including stability screening, intergroup differential analysis, and decorrelation analysis-yielded 88 key features. LASSO regression (10-fold cross-validation) ultimately selected three optimal wavelet-based texture features that formed the core components of our radiomics signature: wavelet. LLH_glcm_Idn (inverse difference normalized), wavelet. LLH_glcm_MCC (maximum correlation coefficient), and wavelet. LHL_firstorder_Skewness. The dataset was randomly split into a training set (83 cases) and a validation set (36 cases) at a 7:3 ratio. A support vector machine (SVM) classifier was constructed, and model performance and clinical utility were evaluated using receiver operating characteristic (ROC) curves and decision curve analysis. Among 119 patients, 43(36.13%) achieved pCR. The constructed radiomics model demonstrated an area under the curve (AUC) of 0.667 and 0.647 in the training and validation sets, respectively, with accuracy rates of 66.27% and 73.49%. Decision curve analysis suggested potential clinial utility under hypothetical scenarios when the probability threshold exceeded 0.3, although this finding is exploratory and requires prospective validation.

This study developed and internally validated a minimalist radiomics model based on mid-treatment MRI, demonstrating moderate and stable predictive capability for pCR after NAC in breast cancer and showing potential for aiding clinical decision-making. As an exploratory proof-of-concept study, the findings underscore the necessity for future multicenter external validation and integration of multimodal features.