A new lung cancer detection method based on the chest CT images using Federated Learning and blockchain systems

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Cybersecurity in Medicine

Background:

• Lung cancer is a leading cause of global mortality, necessitating improved diagnostic accuracy.
• Computed Tomography (CT) scans are crucial for lung disease diagnosis, but human interpretation faces limitations.
• Sharing medical data globally for AI model training is hindered by privacy concerns and data variability.

Purpose of the Study:

• To develop an accurate method for detecting malignant lung nodules in CT scans.
• To categorize lung cancer severity using advanced computational techniques.
• To establish a privacy-preserving framework for collaborative global deep learning model training.

Main Methods:

• Utilized Deep Learning (DL) algorithms, specifically CapsNets, for nodule detection and local classification.
• Implemented a data normalization technique to standardize CT scan data from diverse sources.
• Developed a blockchain-based Federated Learning (FL) approach for secure, collaborative global model training.

Main Results:

• Achieved high accuracy in detecting lung cancer patients, demonstrating the effectiveness of the proposed method.
• The integrated blockchain and FL system successfully trained a global DL model while maintaining institutional anonymity.
• The technique yielded an accuracy of 99.69% with minimal categorization error on multiple datasets.

Conclusions:

• The proposed blockchain-based FL method offers a robust solution for privacy-preserving collaborative lung cancer detection.
• This approach addresses critical challenges in medical data sharing and global AI model development.
• The findings indicate a significant advancement in the automated diagnosis and classification of lung cancer from CT imaging.