Scholarlink

Chronic Kidney Disease (CKD) remains a major global health concern, often progressing undetected until advanced stages. Accurate early prediction is critical, yet traditional AI approaches rely on centralized data, raising privacy concerns and limiting collaboration across healthcare institutions. This research proposed a novel collaborative CKD prediction framework that leverages Federated Learning (FL) to enable decentralized model training over multiple hospitals except sharing susceptive patient data. To address the “black-box” nature of deep learning models, we integrate Explainable Artificial Intelligence (XAI) techniques-specifically SHAP (SHapley Additive exPlanations) to enhance interpretability and foster clinical trust in the predictive outcomes. The proposed model is evaluated using real-world CKD datasets in a simulated federated environment, measuring performance in terms of accuracy, precision, recall, and area under the curve (AUC). Our federated model achieves performance comparable to centralized models (accuracy: 97.3%, AUC: 0.98) while maintaining data privacy. The XAI integration provides meaningful insights into key predictive features such as serum creatinine, blood pressure, and albumin levels, aligning with clinical understanding. This work demonstrates that privacy-preserving, interpretable AI is not only possible but essential for the future of collaborative healthcare diagnostics.

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 Collaborative and Interpretable AI Model for Chronic Kidney Disease Prediction Using Federated Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • P. Marimuktu,
  • Dinesh Dattatray Patil,
  • Manoj Narhar Behere

摘要

Chronic Kidney Disease (CKD) remains a major global health concern, often progressing undetected until advanced stages. Accurate early prediction is critical, yet traditional AI approaches rely on centralized data, raising privacy concerns and limiting collaboration across healthcare institutions. This research proposed a novel collaborative CKD prediction framework that leverages Federated Learning (FL) to enable decentralized model training over multiple hospitals except sharing susceptive patient data. To address the “black-box” nature of deep learning models, we integrate Explainable Artificial Intelligence (XAI) techniques-specifically SHAP (SHapley Additive exPlanations) to enhance interpretability and foster clinical trust in the predictive outcomes. The proposed model is evaluated using real-world CKD datasets in a simulated federated environment, measuring performance in terms of accuracy, precision, recall, and area under the curve (AUC). Our federated model achieves performance comparable to centralized models (accuracy: 97.3%, AUC: 0.98) while maintaining data privacy. The XAI integration provides meaningful insights into key predictive features such as serum creatinine, blood pressure, and albumin levels, aligning with clinical understanding. This work demonstrates that privacy-preserving, interpretable AI is not only possible but essential for the future of collaborative healthcare diagnostics.