142 DCE-MRI radiogenomics predicts immune microenvironment subtypes and neoadjuvant chemotherapy response in HR + /HER2 − breast cancer
Year: 2026
Type: conference-abstract
Abstract: Background Accurate pretreatment prediction of pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) remains an unmet clinical need for HR +/HER2− breast cancer. We aimed to build an immune-derived radiogenomic signature to predict NAC response non-invasively.Methods We quantified the infiltration of 28 immune cell subsets from transcriptomic data of 625 patients with HR +/HER2- breast cancer and performed unsupervised clustering to define two immune clusters with distinct tumor microenvironments. Differentially expressed genes between clusters were used to develop a 3-gene immune-related score (IRS). We compared immune phenotypes, chemosensitivity and pCR rates between immune clusters and IRS subgroups. Intra- and peritumoral DCE-MRI radiomic features were extracted from a training cohort of 378 patients (I-SPY2) to build an XGBoost classifier for IRS discrimination, which was validated externally across five independent cohorts (n = 575).Results Two discrete immune clusters were identified ( figure 1). Cluster 2 exhibited significantly higher pCR rates (17.3% vs 6.5%, P = 3.06e-05), elevated ESTIMATE immune scores and immunophenoscore (IPS), abundant CD8+ T cells and M1 macrophages infiltration, and better sensitivity to paclitaxel, docetaxel, epirubicin and cyclophosphamide. The 3-gene IRS (CXCL9, TTK, COL10A1) stratified patients into high- and low- score subgroups (figure 2). The high-IRS group recapitulated the favorable immune and chemotherapeutic phenotypes of Cluster 2, with significantly higher pCR rates in both training (21.1% vs 2.2%) and independent TCGA validation cohorts (28.6% vs. 5.3%, P < 0.001). Pathway enrichment showed the high IRS subgroup was enriched in cell cycle and chemokine signaling pathways, with CXCL9 and TTK serving as core functional mediators. Multivariate logistic regression confirmed high IRS as an independent predictor of pCR (OR = 13.127, P < 0.001). The optimized 7-feature XGBoost radiomic classifier accurately distinguished IRS subgroups (training AUC = 0.926) and stably predicted pCR across all four external cohorts (all AUC > 0.88).Conclusions We developed an 3-gene IRS that independently predicts pCR in HR +/HER2- breast cancer receiving NAC. A high-performance DCE-MRI radiogenomics classifier was further constructed to non-invasively identify IRS subgroups and directly predict pCR. This biopsy-free imaging signature supports reliable pretreatment risk stratification to guide individualized neoadjuvant treatment strategies.Abstract 142 Figure 1Unsupervised immune cluster identification and phenotypic characterization in HR+/HER2− breast cancer. (A)Unsupervised clustering plot based on 28 immune cell infiltration scores; (B) Heatmap showing the relative abundance of 28 immune cell subsets across two immune clusters; (C) Box plot comparing pathological complete response (pCR) rates between ClusterAbstract 142 Figure 2Construction, validation and radiogenomic translation of the 3-gene immune-related score (IRS). (A) Box plot of pCR rates stratified by high versus low IRS in transcriptomic training and TCGA external cohorts; (B) Bubble plot of top enriched biological pathways in high- versus low-IRS patients; (C) Forest plot of multivariate logistic regression ide
Authors: Ruiliang Chen, Yushuai Yu, Chuangui Song
Institutions: Fujian Medical University, Fujian Provincial Cancer Hospital
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $2,490
OpenAlex ID: W7225165598