Scholarlink

Background/aims <p>Accurate prediction of the lymph node status is essential for clinical decision-making in breast cancer. This work aimed to develop a multimodal deep learning model that integrates gene expression data with hematoxylin and eosin (H&amp;E)-stained whole slide images (WSIs) to predict lymph node metastasis (LNM) and assess the tumor microenvironment in breast cancer.</p> Methods <p>The Cancer Genome Atlas (TCGA) database was used to gather the clinicopathological data, transcriptome information, and WSIs of patients with breast cancer. WSIs from Xuzhou Central Hospital were used as external validations. Four deep learning models were used to build the LNM pathological model. The pathogenic model and hub genes provided the foundation for constructing the nomogram. The potential mechanism was evaluated through the performance of functional enrichment. The calibration, discrimination, and clinical usefulness of the pathological model, multi-gene model, and nomogram were evaluated. Furthermore, the correlation between the nomogram and prognosis, clinicopathological features, and the quantity of immune cells was evaluated. Using immunohistochemistry and popliteal lymph node metastasis tests, the expression of beta-1,3-galactosyltransferase-4 (B3GALT4) in tumor tissues and its association with LNM and CD8<sup>+</sup> T cells were examined.</p> Results <p>The findings of the present study demonstrated that in comparison to any single model, the nomogram’s area under the curve (AUC) was better, at 0.99 [95% CI: 0.98–1.00]. The calibration curve demonstrated a high degree of agreement. The bulk of the threshold probabilities in this model were linked to favorable net benefits. Compared to the low-risk group, patients with high-risk breast cancer had more CD8<sup>+</sup> T cells, a more advanced stage, and worse overall outcomes (<i>P</i> &lt; 0.05). Breast cancer lymph node metastasis was facilitated by B3GALT4-induced CD8<sup>+</sup> T cell exhaustion.</p> Conclusions <p>This model may provide a higher predictive value for LNM. B3GALT4 could serve as a risk biomarker of lymph node metastatic breast cancer.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel model for assessing lymph node metastasis and the immune microenvironment in breast cancer by integrating digital pathology images and transcriptomics

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月20日

查看全文
  • Yongliang Sha,
  • Huijie Zhuang,
  • Yiqiu Wang,
  • Hao Guo,
  • Hui Cheng

摘要

Background/aims

Accurate prediction of the lymph node status is essential for clinical decision-making in breast cancer. This work aimed to develop a multimodal deep learning model that integrates gene expression data with hematoxylin and eosin (H&E)-stained whole slide images (WSIs) to predict lymph node metastasis (LNM) and assess the tumor microenvironment in breast cancer.

Methods

The Cancer Genome Atlas (TCGA) database was used to gather the clinicopathological data, transcriptome information, and WSIs of patients with breast cancer. WSIs from Xuzhou Central Hospital were used as external validations. Four deep learning models were used to build the LNM pathological model. The pathogenic model and hub genes provided the foundation for constructing the nomogram. The potential mechanism was evaluated through the performance of functional enrichment. The calibration, discrimination, and clinical usefulness of the pathological model, multi-gene model, and nomogram were evaluated. Furthermore, the correlation between the nomogram and prognosis, clinicopathological features, and the quantity of immune cells was evaluated. Using immunohistochemistry and popliteal lymph node metastasis tests, the expression of beta-1,3-galactosyltransferase-4 (B3GALT4) in tumor tissues and its association with LNM and CD8+ T cells were examined.

Results

The findings of the present study demonstrated that in comparison to any single model, the nomogram’s area under the curve (AUC) was better, at 0.99 [95% CI: 0.98–1.00]. The calibration curve demonstrated a high degree of agreement. The bulk of the threshold probabilities in this model were linked to favorable net benefits. Compared to the low-risk group, patients with high-risk breast cancer had more CD8+ T cells, a more advanced stage, and worse overall outcomes (P < 0.05). Breast cancer lymph node metastasis was facilitated by B3GALT4-induced CD8+ T cell exhaustion.

Conclusions

This model may provide a higher predictive value for LNM. B3GALT4 could serve as a risk biomarker of lymph node metastatic breast cancer.