Prediction of clinicopathological features, multi-omics events and prognosis based on digital pathology and deep learning in HR+/HER2− breast cancer

• DOI: 10.21037/jtd-23-445
• Publication Date: 2023-05-30T03:12:51Z
• Authors: Jia Hu; Hong Lv; Shen Zhao; Cai-Jin Lin; Guan-Hua Su; Zhi-Ming Shao
• Keywords: Concordance; Progesterone receptor

Background: Breast cancer has the highest incidence and mortality rates among women worldwide. Hormone receptor (HR)+/human epidermal growth factor 2 (HER2)− breast is most common molecular subtype, accounting for 50–79% of cancers. Deep learning been widely used in image analysis, especially predicting targets related to precise treatment patient prognosis. However, studies focusing on therapeutic target prognosis HR+/HER2− are lacking. Methods: This study retrospectively collected hematoxylin eosin (H&E)-stained slides patients between January 2013 December 2014 at Fudan University Shanghai Cancer Center (FUSCC) scanned generate whole-slide images (WSIs). Then, we built a deep-learning-based workflow train validate model predict clinicopathological features, multi-omics features prognosis; area under curve (AUC) receiver operating characteristic (ROC) concordance index (C-index) test set were assess effectiveness. Results: A total 421 included our study. Regarding grade III could be predicted with an AUC 0.90 [95% confidence interval (CI): 0.84–0.97]. somatic mutations, TP53 GATA3 mutation AUCs 0.68 (95% CI: 0.56–0.81) 0.47–0.89), respectively. gene enrichment analysis (GSEA) pathways, G2-M checkpoint pathway was 0.79 0.69–0.90). markers immunotherapy response, intratumoral tumor-infiltrating lymphocytes (iTILs), stromal (sTILs), CD8A, PDCD1 0.78 0.55–1.00), 0.76 0.65–0.87), 0.71 0.60–0.82), 0.74 0.63–0.85), In addition, found that integration clinical prognostic variables deep can improve stratification Conclusions: Using workflow, developed models using pathological WSIs. work may contribute efficient promote personalized management cancer.