Blockchained Federated Learning for Privacy and Security Preservation: Practical Example of Diagnosing Cerebellar...

Area of Science:

• Neuroscience
• Computer Science
• Medical Informatics

Background:

• Cerebellar ataxia (CA) causes movement incoordination due to cerebellar dysfunction.
• Conventional machine learning (ML) for CA diagnosis requires centralized data, raising privacy concerns.
• Federated learning (FL) offers a privacy-preserving alternative by sharing model weights, not raw data.

Purpose of the Study:

• To propose and evaluate a novel blockchained federated learning (FL) approach for diagnosing cerebellar ataxia (CA).
• To enhance the security and privacy of ML models used in CA diagnosis.
• To assess the feasibility of distributed FL with individual validators across multiple clinics.

Main Methods:

• Development of a blockchained federated learning (FL) framework incorporating individual validators.
• Simulation of the proposed approach using a real-world dataset from kinematic sensors of CA patients.
• Data collection from four geographically separated clinics to ensure distributed and diverse data.

Main Results:

• The blockchained FL model achieved a competitive diagnostic accuracy of 89.30% for cerebellar ataxia.
• The proposed method effectively preserved patient privacy by avoiding raw data centralization.
• Security was enhanced through the blockchain integration and individual validators, mitigating network attack risks.

Conclusions:

• Blockchained FL with individual validators presents a viable and secure solution for diagnosing cerebellar ataxia.
• This approach successfully balances high diagnostic accuracy with robust data privacy and security.
• The findings support the adoption of decentralized and secure ML techniques in clinical diagnostics.