Blockchain-Federated-Learning and Deep Learning Models for COVID-19 Detection Using CT Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Blockchain Technology

Background:

• The rapid spread of COVID-19 necessitates effective diagnostic tools, but current testing kit shortages and data privacy concerns hinder global collaboration.
• Training deep learning models for COVID-19 detection is challenging due to data heterogeneity from various CT scanners and the need for patient privacy.

Purpose of the Study:

• To propose a secure and privacy-preserving framework for global deep learning model training using federated learning and blockchain.
• To develop an effective method for detecting COVID-19 patients from Computed Tomography (CT) images.

Main Methods:

• A data normalization technique was developed to address heterogeneity from diverse CT scanners.
• Capsule Network-based segmentation and classification were employed for COVID-19 detection.
• A collaborative global model was trained using blockchain technology with federated learning to preserve organizational privacy.

Main Results:

• The proposed framework successfully trained a global deep learning model using federated learning and blockchain.
• The system demonstrated improved recognition of COVID-19 from CT images by utilizing up-to-date, normalized data.
• Comprehensive experiments validated the framework's effectiveness, showing superior performance in detecting COVID-19 patients.

Conclusions:

• The developed framework offers an effective solution for COVID-19 diagnosis through secure, privacy-preserving, and collaborative deep learning.
• This approach addresses the challenges of data sharing and heterogeneity in medical imaging AI.
• The study provides a valuable resource for the research community with newly collected COVID-19 patient data.