Quantum-Inspired Privacy-Preserving Federated Learning Framework for Secure Dementia Classification

2025-03-05

AI Summary

To address privacy leakage and advanced threats—including gradient inversion and eavesdropping—in cross-institutional dementia classification using MRI data, this paper proposes the first privacy-enhancing federated learning framework integrated with quantum key distribution (QKD). The method deeply embeds QKD into the federated weight aggregation process, enabling end-to-end secure key distribution and gradient encryption, thereby fundamentally mitigating model inversion and communication eavesdropping. Evaluated on the OASIS dataset using a CNN architecture, the framework achieves classification accuracy comparable to centralized baselines while reducing training loss by 0.97%, significantly improving communication robustness and model update security. Its lightweight design ensures practical deployment in resource-constrained clinical settings. The core contribution lies in the first native integration of QKD into the federated aggregation protocol—uniquely bridging information-theoretic security guarantees with real-world medical applicability.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Search and Optimization: Distributed Search
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Security and Privacy: Blockchains and distributed ledgers
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Dementia, a neurological disorder impacting millions globally, presents significant challenges in diagnosis and patient care. With the rise of privacy concerns and security threats in healthcare, federated learning (FL) has emerged as a promising approach to enable collaborative model training across decentralized datasets without exposing sensitive patient information. However, FL remains vulnerable to advanced security breaches such as gradient inversion and eavesdropping attacks. This paper introduces a novel framework that integrates federated learning with quantum-inspired encryption techniques for dementia classification, emphasizing privacy preservation and security. Leveraging quantum key distribution (QKD), the framework ensures secure transmission of model weights, protecting against unauthorized access and interception during training. The methodology utilizes a convolutional neural network (CNN) for dementia classification, with federated training conducted across distributed healthcare nodes, incorporating QKD-encrypted weight sharing to secure the aggregation process. Experimental evaluations conducted on MRI data from the OASIS dataset demonstrate that the proposed framework achieves identical accuracy levels to a baseline model while enhancing data security and reducing loss by almost 1% compared to the classical baseline model. The framework offers significant implications for democratizing access to AI-driven dementia diagnostics in low- and middle-income countries, addressing critical resource and privacy constraints. This work contributes a robust, scalable, and secure federated learning solution for healthcare applications, paving the way for broader adoption of quantum-inspired techniques in AI-driven medical research.

Problem

• Enhances privacy in federated learning for dementia classification.
• Integrates quantum-inspired encryption to secure model weight transmission.
• Improves data security and reduces loss in MRI-based diagnostics.

Innovation

• Quantum-inspired encryption for secure federated learning
• CNN-based dementia classification with QKD protection
• Enhanced privacy and security in healthcare AI