Scholarlink

<p> <!--Query ID="Q1" Text="Please check and confirm if the authors and their respective affiliations have been correctly identified. Amend if necessary." Resolved="yes"-->Recently, ensuring data privacy has become a critical concern in the healthcare sector. This study presents GenDiabSec, a collaborative solution that integrates fully homomorphic encryption (FHE), a federated recurrent neural network, and blockchain-based smart contracts to securely and effectively detect diabetes. In order to maintain data privacy, GenDiabSec introduces a novel FHE scheme called DNAGen, which leverages the N-nucleotide representation, multistage genetic algorithms, homomorphic DNA operations (ADD, SUB, XOR), and the Arabidopsis thaliana DNA database. For local models, each hospital applies a new Patches-Based Bidirectional Recurrent Neural Network (PBRNN) on encrypted learning data, incorporating data augmentation to address the data imbalance issues in the Pima Indian Diabetes Dataset (PIDD). We use Federated learning (FL) for global model to securely aggregate local models without sharing raw data. Experimental evaluations using the PIDD demonstrate the superior performance of our solution, achieving an accuracy of 95%, surpassing other methods such as SVM (86.2%), KNN (82%), CNN (82%), XGBoost (82%), ResNet14 (79.6%), ResNet50 (78.7%), K-means (64.8%), GA-MLP (80%), GRNN (80.2%), RF (79.5%), DT (77.6%), LR (80.2%), MLFNN (81.73%), LDA(72.2%), Bagging(69%), SMO(71.3%) and LSTM (85%). Notably, with a CEE of 0.09 and an RMSE of 0.09, the DNAGen scheme improves detection precision and security level while reducing execution time compared to existing FHE schemes such as CKKS and BGV. Finally, by integrating average federated (AvgFed) with blockchain smart contracts, GenDiabSec facilitates collaborative training between healthcare institutions and decentralized model updates, efficiently scaling to millions of users.</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

GenDiabSec: A Federated Recurrent Neural Network with Multistage DNAGen and Blockchain for Secure Collaborative Diabetes Detection

  • 期刊论文

  • 发表日期：

    2025年7月1日

查看全文
  • Hadj Ahmed Bouarara,
  • Amine Rahmani,
  • Mohamed Elhadi Rahmani,
  • Mohamed Guendouz

摘要

Recently, ensuring data privacy has become a critical concern in the healthcare sector. This study presents GenDiabSec, a collaborative solution that integrates fully homomorphic encryption (FHE), a federated recurrent neural network, and blockchain-based smart contracts to securely and effectively detect diabetes. In order to maintain data privacy, GenDiabSec introduces a novel FHE scheme called DNAGen, which leverages the N-nucleotide representation, multistage genetic algorithms, homomorphic DNA operations (ADD, SUB, XOR), and the Arabidopsis thaliana DNA database. For local models, each hospital applies a new Patches-Based Bidirectional Recurrent Neural Network (PBRNN) on encrypted learning data, incorporating data augmentation to address the data imbalance issues in the Pima Indian Diabetes Dataset (PIDD). We use Federated learning (FL) for global model to securely aggregate local models without sharing raw data. Experimental evaluations using the PIDD demonstrate the superior performance of our solution, achieving an accuracy of 95%, surpassing other methods such as SVM (86.2%), KNN (82%), CNN (82%), XGBoost (82%), ResNet14 (79.6%), ResNet50 (78.7%), K-means (64.8%), GA-MLP (80%), GRNN (80.2%), RF (79.5%), DT (77.6%), LR (80.2%), MLFNN (81.73%), LDA(72.2%), Bagging(69%), SMO(71.3%) and LSTM (85%). Notably, with a CEE of 0.09 and an RMSE of 0.09, the DNAGen scheme improves detection precision and security level while reducing execution time compared to existing FHE schemes such as CKKS and BGV. Finally, by integrating average federated (AvgFed) with blockchain smart contracts, GenDiabSec facilitates collaborative training between healthcare institutions and decentralized model updates, efficiently scaling to millions of users.