bioimagesUnknownInna Averbukhhttps://www.ebi.ac.uk/biostudies/studies/S-BIAD2708bioimagesSpecimenFundingStudy ComponentBiosampleorganisationAssociationsAnnotationImage acquisitionSean BendallMichael AngeloInna AverbukhHadeesha PiyadasafalseTopological encoding of risk in ductal carcinoma in situOver 50,000 women in the US are diagnosed each year with ductal carcinoma in situ (DCIS), a preinvasive breast lesion of which only a subset will progress to invasive cancer. Because non-progressors cannot be identified at diagnosis, most patients undergo aggressive treatment with significant side effects. Here, we developed a high-dimensional spatial proteomic panel to comprehensively characterize epithelial and myoepithelial cell states and their surrounding microenvironment in a clinically annotated DCIS cohort (n = 162). Combined with deep learning–based segmentation, we profiled over one million single cells and mapped spatial niche organization across epithelial, stromal, and immune compartments. A composite spatial confidence score integrating epithelial organization, stromal desmoplasia, collagen branching and alignment, cell composition, and cell phenotypes, and compartmentalized ductal marker expression identified a high-confidence non-progressor group (46% of cases) with a negative predictive value of 100%. These data revealed conserved protective architectures and coordinated immune–stromal-tumor niches that maintain ductal integrity, establishing a framework for DCIS risk stratification and treatment de-escalation.2026-01-19T00:00:00Z2026-04-29T15:31:43.206Z2026-01-19T19:49:56.123ZS-BIAD2708