WACPN: A Neural Network for Pneumonia Diagnosis

Area of Science:

• Artificial Intelligence in Medicine
• Computational Diagnostics
• Biomedical Signal Processing

Background:

• Community-acquired pneumonia (CAP) diagnosis outside clinical settings presents challenges.
• Existing diagnostic methods require enhancement for improved efficiency and accuracy.
• Novel computational models are needed to address diagnostic limitations in CAP.

Purpose of the Study:

• To develop and evaluate a novel neural network model for efficient community-acquired pneumonia (CAP) diagnosis.
• To introduce and integrate a 2-dimensional wavelet entropy (2d-WE) layer and an adaptive chaotic particle swarm optimization (ACP) algorithm.
• To assess the diagnostic performance of the proposed WACPN model against existing state-of-the-art methods.

Main Methods:

• Development of a feed-forward neural network trained using an adaptive chaotic particle swarm optimization (ACP) algorithm.
• Integration of a 2-dimensional wavelet entropy (2d-WE) layer into the neural network architecture.
• The ACP algorithm incorporates adaptive inertia weight factor (AIWF) and Rossler attractor (RA) for enhanced optimization.

Main Results:

• The WACPN model achieved high diagnostic performance metrics: sensitivity (91.87%), specificity (90.70%), accuracy (91.29%), and AUC (0.9577).
• Experiments confirmed the effectiveness of AIWF and RA in improving the standard particle swarm optimization.
• The WACPN model demonstrated superior performance compared to six other state-of-the-art diagnostic models.

Conclusions:

• The proposed WACPN model is highly effective and efficient for diagnosing community-acquired pneumonia (CAP).
• The integration of 2d-WE and ACP significantly enhances diagnostic capabilities.
• The model's availability in a cloud computing environment will facilitate wider accessibility and application.