Peritumoral Radiomic Deviation on DCE-MRI Exceeds the Bilateral Normal-Appearing Parenchymal Distance in Breast Cancer: A Within-Patient Multi-Metric Study

Background: The imaging-defined peritumoral region may contain quantitative imaging information beyond that captured by the tumor core, but its radiomic deviation from normal-appearing breast parenchyma remains insufficiently characterized. This retrospective single-center study investigated whether peritumoral-to-normal radiomic distances exceed the within-patient bilateral normal-appearing parenchymal reference distance. Fifty-six patients with histopathologically confirmed breast cancer and preoperative bilateral dynamic contrast-enhanced MRI were included. Methods: Single-slice radiomic analysis was performed for four predefined regions: tumor core (T), peritumoral microenvironment (M), remote ipsilateral normal-appearing parenchyma (I), and contralateral normal-appearing parenchyma (C). After quality control, 101 radiomic features were standardized and analyzed using standardized Euclidean, Manhattan, cosine, and Ledoit–Wolf Mahalanobis distances. Peritumoral-to-ipsilateral (dMI) and peritumoral-to-contralateral (dMC) distances were compared with the within-patient bilateral normal-appearing parenchymal distance (dIC). Results: Across all four metrics, both dMI and dMC exceeded dIC, and all eight primary paired Wilcoxon comparisons remained significant after Benjamini–Hochberg correction (q≤1.07×10−6), with matched-pairs rank-biserial correlations ranging from 0.749 to 0.989. The principal pattern was broadly preserved across alternative Mahalanobis covariance formulations, reduced-dimensional PCA analyses, particularly when three or four principal components were retained, exclusion of all 13 retained shape features, separate radiomic feature-family analyses, and replacement of selected non-normalized non-uniformity features with normalized counterparts. Conclusions: These findings show that, in the present cohort, peritumoral-to-normal radiomic distances exceeded the within-patient bilateral normal-appearing parenchymal reference distance. Because the four analyzed regions differed in size, shape, and sampling geometry, the reported distances should be interpreted as region-level radiomic differences under the present ROI definitions rather than as geometry-independent tissue properties. Larger multi-institutional studies using full-volume three-dimensional radiomics and standardized imaging and preprocessing strategies are needed to assess reproducibility, clarify the biological correlates of this radiomic pattern, and evaluate its potential clinical relevance.