Integrated collagen architecture and composition improve risk stratification in triple-negative breast cancer.

Triple-negative breast cancer (TNBC) exhibits substantial clinical heterogeneity, with some patients experiencing early recurrence and poor survival despite similar clinicopathologic features. Here, we investigated whether quantitative assessment of intratumoral collagen architecture and composition could improve risk stratification in TNBC. We analyzed a retrospective cohort of 79 TNBC tumors assembled into tissue microarrays using a multimodal computational pathology framework integrating Masson Trichrome staining with COL1 and COL3 immunohistochemistry. Collagen architecture was quantified using fiber-based image analysis and unsupervised clustering, while collagen composition was assessed using a normalized COL3:COL1 ratio. Unsupervised analysis identified four distinct collagen architectural states, which were consolidated into low-risk and high-risk groups based on recurrence patterns. High-risk collagen architecture was associated with shorter recurrence-free interval (log-rank p = 0.025; restricted mean survival time difference=10.1 months). Independently, a higher COL3:COL1 ratio was associated with improved overall survival (log-rank p = 0.042; restricted mean survival time difference=9.4 months). Integration of collagen architecture and composition further refined risk stratification, with patients demonstrating high-risk architecture and low COL3:COL1 ratios exhibiting the poorest outcomes. Notably, collagen-defined phenotypes identified patients with divergent outcomes not readily apparent from tumor stage alone. Together, these findings demonstrate that quantitative assessment of intratumoral collagen architecture and composition provides clinically meaningful prognostic information in TNBC and supports extracellular matrix phenotyping as a practical computational pathology approach for refining risk assessment.