Efficient federated learning for pediatric pneumonia on chest X-ray classification.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Data privacy in machine learning

Background:

• Pneumonia is a leading cause of death in children under five globally.
• Traditional machine learning for chest X-ray diagnosis faces data privacy and security challenges due to data centralization.
• Federated learning offers a solution by training models without sharing sensitive patient data.

Purpose of the Study:

• To develop a federated learning approach for accurate and secure diagnosis of pediatric pneumonia from chest X-rays.
• To address the challenge of data heterogeneity in federated learning that impacts classification performance.
• To improve upon existing federated learning and diffusion model-based methods for pediatric pneumonia detection.

Main Methods:

• Utilized federated learning (FL) with modifications to the FedAvg algorithm.
• Modified the client-side loss function by incorporating regular or penalty terms.
• Implemented server-side momentum after average aggregation to handle data heterogeneity.
• Employed a two-end control variable method to mitigate classification performance degradation.

Main Results:

• The proposed federated learning method effectively protects data security, preventing privacy leakage.
• Achieved an average accuracy of 98% and individual accuracy up to 99% for pediatric pneumonia classification.
• Demonstrated an average improvement of 2% compared to previous federated learning algorithms and diffusion models.
• Successfully addressed data heterogeneity issues impacting classification accuracy.

Conclusions:

• Federated learning provides a secure alternative to traditional machine learning for medical image analysis.
• The proposed two-end control variable method significantly improves classification accuracy in heterogeneous federated learning settings.
• This approach offers a valuable tool for clinicians worldwide to enhance the detection of pediatric pneumonia.