A blockchain-empowered federated learning-based framework for data privacy in lung disease detection system

Abstract

Lung diseases are one of the prime reasons for mortality globally, having an estimated five million per year fatal cases worldwide. This is a growing global concern so early detection using a Computed Tomography (CT) scan is crucial to prevent loss that grabs the attention of cutting-edge technologies to bring the concept called “Smart Healthcare”. However, the paucity and heterogeneity of medical data across the globe make it challenging to develop a global classification framework, while the other concerns that arise from legal and privacy leakage become an obstacle for data sharing as single source data is hardly enough to represent universal. Federated Learning has issued a solution to licensing research and data heterogeneity concerns allowing collaborative and on-device learning without sharing raw data. FL faces security issues such as Denial-of-service, Reverse engineering attacks, etc, where it is impossible to track the data and store it securely. The study proposes an innovative framework that combines Blockchain technology and Federated Learning (FL) to enable collaborative model training while preserving data privacy. Through this approach, patient data is authenticated using blockchain, and FL facilitates on-device learning without sharing raw data. The framework utilizes the DenseNet-201 model for lung disease classification, with model parameter aggregation using the FedAvg algorithm and storage on the blockchain via IPFS. Finally, we have conducted a substantial investigation with Python and its widely used libraries, like TensorFlow and Scikit-Learn to demonstrate that the algorithm accurately detects lung diseases and attained an accuracy, precision, recall, and F1-score of 90%.