Basnin Nanziba, Mahmud Tanjim, Islam Raihan Ul, Andersson Karl
Cybersecurity Laboratory, Luleå University of Technology, 97187 Luleå, Sweden.
Department of Computer Science and Engineering, Rangamati Science and Technology University, Rangamati 4500, Bangladesh.
Diagnostics (Basel). 2025 Jan 1;15(1):80. doi: 10.3390/diagnostics15010080.
Alzheimer's disease (AD) leads to severe cognitive impairment and functional decline in patients, and its exact cause remains unknown. Early diagnosis of AD is imperative to enable timely interventions that can slow the progression of the disease. This research tackles the complexity and uncertainty of AD by employing a multimodal approach that integrates medical imaging and demographic data. To scale this system to larger environments, such as hospital settings, and to ensure the sustainability, security, and privacy of sensitive data, this research employs both deep learning and federated learning frameworks. MRI images are pre-processed and fed into a convolutional neural network (CNN), which generates a prediction file. This prediction file is then combined with demographic data and distributed among clients for local training. Training is conducted both locally and globally using a belief rule base (BRB), which effectively integrates various data sources into a comprehensive diagnostic model. The aggregated data values from local training are collected on a central server. Various aggregation methods are evaluated to assess the performance of the federated learning model, with results indicating that FedAvg outperforms other methods, achieving a global accuracy of 99.9%. The BRB effectively manages the uncertainty associated with AD data, providing a robust framework for integrating and analyzing diverse information. This research not only advances AD diagnostics by integrating multimodal data but also underscores the potential of federated learning for scalable, privacy-preserving healthcare solutions.
阿尔茨海默病（AD）会导致患者出现严重的认知障碍和功能衰退，其确切病因尚不清楚。AD的早期诊断对于及时采取干预措施以减缓疾病进展至关重要。本研究通过采用整合医学影像和人口统计学数据的多模态方法来应对AD的复杂性和不确定性。为了将该系统扩展到更大的环境，如医院环境，并确保敏感数据的可持续性、安全性和隐私性，本研究采用了深度学习和联邦学习框架。磁共振成像（MRI）图像经过预处理后输入卷积神经网络（CNN），该网络生成一个预测文件。然后将此预测文件与人口统计学数据相结合，并分发给客户端进行本地训练。使用信念规则库（BRB）在本地和全局进行训练，该规则库可有效地将各种数据源整合到一个综合诊断模型中。来自本地训练的聚合数据值在中央服务器上收集。评估了各种聚合方法以评估联邦学习模型的性能，结果表明联邦平均算法（FedAvg）优于其他方法，实现了99.9%的全局准确率。BRB有效地管理了与AD数据相关的不确定性，为整合和分析各种信息提供了一个强大的框架。本研究不仅通过整合多模态数据推进了AD诊断，还强调了联邦学习在可扩展、保护隐私的医疗保健解决方案中的潜力。
