FLPneXAINet: Federated deep learning and explainable AI for improved pneumonia prediction utilizing GAN-augmented...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Pneumonia diagnosis is challenging due to symptom overlap and privacy concerns with clinical data sharing.
• Accurate and timely pneumonia detection from chest X-rays (CXRs) is crucial for effective treatment.
• Existing methods struggle to balance diagnostic accuracy with patient data privacy.

Purpose of the Study:

• To develop and evaluate FLPneXAINet, a novel framework integrating federated learning (FL), deep learning (DL), and explainable AI (XAI).
• To enable secure and accurate pneumonia prediction using CXR images while preserving patient privacy.
• To enhance diagnostic capabilities for healthcare professionals through explainable AI insights.

Main Methods:

• Utilized a Kaggle CXR dataset (8,402 images) preprocessed and augmented with CycleGAN.
• Employed pre-trained DL models (VGG16, NASNetMobile, MobileNet) and ensemble DL (EDL) for feature extraction.
• Performed feature optimization (RFE, ANOVA, RF) and prediction using ML models (KNN, NB, SVM, RF) within a federated learning environment.

Main Results:

• The ensemble deep learning (EDL) model within the FLPneXAINet framework achieved superior performance.
• Achieved high accuracy (97.61%), F1 score (98.36%), recall (98.13%), and precision (98.59%) in pneumonia prediction.
• Explainable AI techniques (LIME, Grad-CAM) validated the framework's predictions, enhancing interpretability.

Conclusions:

• FLPneXAINet provides a robust, privacy-preserving solution for accurate pneumonia diagnosis from CXR images.
• The integration of FL, DL, and XAI significantly improves diagnostic accuracy and clinical utility.
• This framework supports healthcare professionals in making timely and informed treatment decisions for pneumonia.