CSV-Net predicted pathological complete response to neoadjuvant therapy in breast cancer with ROC-AUCs of 0.845 internal and 0.815 external, outperforming prior methods.
Does CSV-Net accurately predict pathological complete response to neoadjuvant therapy in breast cancer patients?
A novel graph learning framework integrating whole-slide images and clinicopathological variables accurately predicts pathological complete response to neoadjuvant therapy in breast cancer patients.
Tasa de eventos absoluta: 0% vs 0%
Abstract Predicting neoadjuvant therapy (NAT) response in breast cancer patients still remains challenging due to the high heterogeneity of tumors and modality differences in medical data. In this paper, we proposed a novel graph learning framework, a clinicopathological-guided spatial-variant network (CSV-Net), which integrates whole-slide images (WSIs) from biopsy samples with clinicopathological (CP) variables to predict pathological complete response (pCR) of patients receiving NAT. CSV-Net incorporates two novel components: (1) a spatial-variant graph convolution (SV-GraphConv) layer, which models the spatial-semantic heterogeneity inherent in whole-slide images (WSIs), and (2) a clinicopathological-guided graph pooling (CG-GraphPool) module that dynamically integrates clinicopathological (CP) variables with WSI features at a fine-grained level, enhancing both performance and interpretability. We evaluated CSV-Net on a retrospective dataset of 950 breast cancer patients from 5 medical centers. CSV-Net achieved ROC-AUCs of 0.845 (95% CI: 0.801-0.886) and 0.815 (95% CI: 0.755-0.873) in predicting pCR on the internal and external validation sets, respectively, outperforming state-of-the-art methods. Notably, subgroup analysis highlighted CSV-Net's robust performance in patients with different molecular subtypes (e.g. Luminal, HER2 overexpression and TNBC), underscoring its clinical utility. Furthermore, disease-free survival relevance and transcriptomic profiling experiments confirmed the model's prognostic value and biological interpretability, suggesting potential applications in personalized treatment planning and biomarker discovery. Citation Format: W. Hou, Z. Pu, Z. Xu, A. Wu, Z. Liu, K. Zhao, C. Duan, J. Guo, K. Chen, S. Qiu, Z. Du, X. Zhao, J. Bai, H. Zeng, G. Zhang. Interpretable Graph Learning on Preoperative Biopsy Predicts Pathological Complete Response to Neoadjuvant Therapy in Breast Cancer abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-06-14.
Hou et al. (Tue,) reported a other. CSV-Net predicted pathological complete response to neoadjuvant therapy in breast cancer with ROC-AUCs of 0.845 internal and 0.815 external, outperforming prior methods.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: