Scholarlink

Objective <p>To systematically evaluate the applications of machine learning for rapid and noninvasive diagnosis of gastric cancer using serum-based biomarkers, and to propose recommendations, in the form of a conceptual model, derived from the review to address identified barriers to clinical translation.</p> Methods <p>A systematic search was conducted on May 10, 2025, in PubMed, Scopus, Embase, and Web of Science for studies from 2014 to 2023 using machine learning with serum-based biomarkers for the diagnosis or prognosis of gastric cancer. Following the PRISMA 2020 guidelines, 12 studies were selected after screening 416 articles. A conceptual model was developed that included the integration of standard data collection, preprocessing, feature engineering, machine learning hybrid modeling, and clinical decision support.</p> Results <p>The reviewed studies demonstrated high performance of machine learning with AUCs of 0.862–1.0, sensitivities of 81.8–100%, and specificity of 90.67–100%. Supervised models (e.g., random forests, support vector machines) and deep learning, with feature selection (e.g., LASSO, Boruta) that enhances interpretability, performed exceptionally well. Challenges included small sample sizes, inconsistent preprocessing, limited model transparency, and poor clinical integration. Based on these findings the recommendations in the form of a proposed conceptual model that standardizes protocols, improves generalizability and interpretability using explainability tools, and enables electronic health record integration, providing a scalable framework for early detection of gastric cancer.</p> Conclusion <p>Machine learning with serum-based biomarkers has transformative potential for gastric cancer diagnosis, but barriers limit its adoption. The recommendations outlined in the conceptual model provide a reproducible and interpretable pipeline suitable for resource-limited settings. Prospective, multicenter validation is needed to integrate these recommendations into clinical practice and advance precision oncology.</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 conceptual model for machine learning based noninvasive gastric cancer diagnosis using serum biomarkers

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月15日

查看全文
  • Vahideh Zolfaghari,
  • Solmaz Sohrabei,
  • Majid Khadem Rezaiyan,
  • Amin Dalili,
  • Masoumeh Gharib

摘要

Objective

To systematically evaluate the applications of machine learning for rapid and noninvasive diagnosis of gastric cancer using serum-based biomarkers, and to propose recommendations, in the form of a conceptual model, derived from the review to address identified barriers to clinical translation.

Methods

A systematic search was conducted on May 10, 2025, in PubMed, Scopus, Embase, and Web of Science for studies from 2014 to 2023 using machine learning with serum-based biomarkers for the diagnosis or prognosis of gastric cancer. Following the PRISMA 2020 guidelines, 12 studies were selected after screening 416 articles. A conceptual model was developed that included the integration of standard data collection, preprocessing, feature engineering, machine learning hybrid modeling, and clinical decision support.

Results

The reviewed studies demonstrated high performance of machine learning with AUCs of 0.862–1.0, sensitivities of 81.8–100%, and specificity of 90.67–100%. Supervised models (e.g., random forests, support vector machines) and deep learning, with feature selection (e.g., LASSO, Boruta) that enhances interpretability, performed exceptionally well. Challenges included small sample sizes, inconsistent preprocessing, limited model transparency, and poor clinical integration. Based on these findings the recommendations in the form of a proposed conceptual model that standardizes protocols, improves generalizability and interpretability using explainability tools, and enables electronic health record integration, providing a scalable framework for early detection of gastric cancer.

Conclusion

Machine learning with serum-based biomarkers has transformative potential for gastric cancer diagnosis, but barriers limit its adoption. The recommendations outlined in the conceptual model provide a reproducible and interpretable pipeline suitable for resource-limited settings. Prospective, multicenter validation is needed to integrate these recommendations into clinical practice and advance precision oncology.