Early Tuberculosis Detection via Privacy-Preserving, Adaptive-Weighted Deep Models.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Public Health

Background:

• Tuberculosis (TB) poses a global health challenge, especially in areas with limited radiological expertise.
• Accurate and early detection of TB from chest X-rays is crucial for effective treatment and control.
• Centralized training of deep learning models raises privacy concerns and may not generalize well across diverse datasets.

Purpose of the Study:

• To develop a scalable, privacy-preserving deep learning system for early tuberculosis identification using chest X-ray images.
• To implement federated learning with a Genetic Algorithm (GA)-optimized adaptive-weighted ensemble to overcome limitations of centralized training and single-model approaches.
• To enhance model robustness and reduce bias through controlled augmentation and ensemble learning.

Main Methods:

• An ensemble learning approach combining multiple locally trained models was developed.
• A Genetic Algorithm (GA) optimized an adaptive-weighted ensemble for optimal model contribution.
• Federated learning enabled collaborative training across institutions while preserving patient data privacy by transmitting only model parameters.

Main Results:

• The federated, GA-optimized ensemble significantly outperformed individual models and fixed-weight ensembles.
• The system achieved 98% accuracy, 97% F1 score, and 0.999 AUC, demonstrating high diagnostic performance.
• Federated learning ensured model robustness across heterogeneous data sources while maintaining complete patient privacy.

Conclusions:

• The proposed federated, GA-optimized ensemble provides highly accurate and robust early tuberculosis detection.
• The system effectively preserves patient privacy across distributed clinical sites.
• This scalable framework shows significant potential for AI-assisted TB screening in resource-limited healthcare settings.