Abstract
Purpose
To identify robust tumor habitat subregions using pre-treatment MRI and assess the prognostic value of the resulting phenotypes for recurrence risk stratification in stages II-III breast cancer.Methods
This multicenter retrospective study included 842 women with stages II-III breast cancer who received pre-treatment MRI scans and subsequent surgical resection from three centers (multicenter discovery cohort: n = 415; external test cohort: n = 427). An unsupervised clustering-based habitat characterization framework was applied to delineate tumor subregions and identify distinct structure habitat phenotypes (SH-phenotypes). Recurrence-free survival (RFS) was evaluated using Kaplan-Meier analysis and multivariate Cox regression. Fine-Gray competing risk regression was additionally used to account for death as a competing event.Results
Two structural subregions (central and peripheral) with unique radiomic patterns and three SH-phenotypes with distinct RFS were identified in the discovery cohort and validated in the test cohort (log-rank p Conclusions
Two structural habitat subregions were identified, and the derived SH-phenotypes provide additional prognostic value beyond clinicopathological factors for recurrence risk stratification in stages II-III breast cancer.Key points
Question Precise evaluation of recurrence risk is crucial for tailoring therapy for stages II-III breast cancer, but it remains challenging with classical clinicopathological factors. Findings Pre-treatment MRI-based SH analysis identified three phenotypes with distinct RFS and recurrence cumulative incidence, independent of clinicopathological factors. Clinical relevance The clinical utility of this phenotyping lies in its role as an adjunct to classical clinicopathological factors, enhancing recurrence risk stratification for optimized treatment planning in stages II-III breast cancer.