Khan Muhammad Amir, Alsulami Musleh, Yaqoob Muhammad Mateen, Alsadie Deafallah, Saudagar Abdul Khader Jilani, AlKhathami Mohammed, Farooq Khattak Umar
Department of Computer Science, COMSATS University Islamabad Abbottabad Campus, Abbottabad 22060, Pakistan.
Information Systems Department, Umm Al-Qura University, Makkah 21961, Saudi Arabia.
Diagnostics (Basel). 2023 Jul 11;13(14):2340. doi: 10.3390/diagnostics13142340.
Healthcare professionals consider predicting heart disease an essential task and deep learning has proven to be a promising approach for achieving this goal. This research paper introduces a novel method called the asynchronous federated deep learning approach for cardiac prediction (AFLCP), which combines a heart disease dataset and deep neural networks (DNNs) with an asynchronous learning technique. The proposed approach employs a method for asynchronously updating the parameters of DNNs and incorporates a temporally weighted aggregation technique to enhance the accuracy and convergence of the central model. To evaluate the effectiveness of the proposed AFLCP method, two datasets with various DNN architectures are tested, and the results demonstrate that the AFLCP approach outperforms the baseline method in terms of both communication cost and model accuracy.
医疗保健专业人员认为预测心脏病是一项重要任务，而深度学习已被证明是实现这一目标的一种有前景的方法。本研究论文介绍了一种名为用于心脏预测的异步联邦深度学习方法（AFLCP）的新方法，该方法将心脏病数据集和深度神经网络（DNN）与异步学习技术相结合。所提出的方法采用了一种异步更新DNN参数的方法，并结合了时间加权聚合技术来提高中心模型的准确性和收敛性。为了评估所提出的AFLCP方法的有效性，对两个具有不同DNN架构的数据集进行了测试，结果表明AFLCP方法在通信成本和模型准确性方面均优于基线方法。
