Scholarlink

<p>Papillary thyroid carcinoma (PTC) is the most common thyroid malignancy, and cervical lymph node metastasis significantly impacts patient prognosis. This study aimed to develop interpretable artificial intelligence models based on transcriptomics to predict PTC occurrence and cervical lymph node metastasis, while exploring the heterogeneity of risk factors across different regions. We obtained 419 samples from the GEO database, originating from Asia, Europe, and America, comprising 158 normal samples, 203 PTC samples, and 58 metastatic samples. After comparing multiple machine learning algorithms, deep neural networks (DNN) demonstrated superior performance and were used to construct the PTC diagnostic and metastasis predictive models. The optimized PTC diagnostic model achieved an AUC of 0.987 with an accuracy of 0.945, while the PTC metastasis predictive model reached an AUC of 0.998 with an accuracy of 0.987. Model interpretation using SHapley Additive exPlanations (SHAP) and Kolmogorov-Arnold Networks (KAN) methods identified SYT1, REN, CNTN5, and ADAM12 as critical features for PTC diagnosis, whereas COL9A1, CYP4F3, and GAD1 were key predictors for PTC metastasis. Stratification analysis revealed regional differences in risk factors for PTC occurrence, while factors promoting PTC metastasis exhibited commonalities across different regions. Pathway enrichment analysis indicated that regulation of hormone levels and cell population proliferation were common pathways involved in both PTC occurrence and metastasis. Finally, we developed online predictive platforms based on the Streamlit framework to facilitate transparent model exploration and risk estimation. These tools are publicly accessible research-use prototypes intended for model demonstration and interpretability visualization. Because they require standardized gene-expression inputs and have not undergone prospective clinical validation, they should not be used as standalone tools for clinical diagnosis, risk stratification, or treatment decision-making. Overall, our findings identify candidate transcriptomic markers associated with PTC occurrence and lymph node metastasis and provide a basis for future translational evaluation.</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

Predictive models for the occurrence and lymph node metastasis of papillary thyroid carcinoma with regional risk heterogeneity

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月18日

查看全文
  • Zhigang Zhang,
  • Hongyu Liu,
  • Zheng Zhao,
  • Guoyu Tan,
  • Yang Zhao,
  • Xun Liu

摘要

Papillary thyroid carcinoma (PTC) is the most common thyroid malignancy, and cervical lymph node metastasis significantly impacts patient prognosis. This study aimed to develop interpretable artificial intelligence models based on transcriptomics to predict PTC occurrence and cervical lymph node metastasis, while exploring the heterogeneity of risk factors across different regions. We obtained 419 samples from the GEO database, originating from Asia, Europe, and America, comprising 158 normal samples, 203 PTC samples, and 58 metastatic samples. After comparing multiple machine learning algorithms, deep neural networks (DNN) demonstrated superior performance and were used to construct the PTC diagnostic and metastasis predictive models. The optimized PTC diagnostic model achieved an AUC of 0.987 with an accuracy of 0.945, while the PTC metastasis predictive model reached an AUC of 0.998 with an accuracy of 0.987. Model interpretation using SHapley Additive exPlanations (SHAP) and Kolmogorov-Arnold Networks (KAN) methods identified SYT1, REN, CNTN5, and ADAM12 as critical features for PTC diagnosis, whereas COL9A1, CYP4F3, and GAD1 were key predictors for PTC metastasis. Stratification analysis revealed regional differences in risk factors for PTC occurrence, while factors promoting PTC metastasis exhibited commonalities across different regions. Pathway enrichment analysis indicated that regulation of hormone levels and cell population proliferation were common pathways involved in both PTC occurrence and metastasis. Finally, we developed online predictive platforms based on the Streamlit framework to facilitate transparent model exploration and risk estimation. These tools are publicly accessible research-use prototypes intended for model demonstration and interpretability visualization. Because they require standardized gene-expression inputs and have not undergone prospective clinical validation, they should not be used as standalone tools for clinical diagnosis, risk stratification, or treatment decision-making. Overall, our findings identify candidate transcriptomic markers associated with PTC occurrence and lymph node metastasis and provide a basis for future translational evaluation.