Pneumonia detection with QCSA network on chest X-ray.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Pneumonia is a leading cause of infant mortality globally.
• Diagnosing pneumonia via chest X-rays is complex and can lead to inter-radiologist disagreement.
• Early diagnosis is crucial for mitigating pneumonia's impact, and computer-aided diagnostics show promise.

Purpose of the Study:

• To propose a novel Quaternion Channel-Spatial Attention (QCSA) network for accurate pneumonia detection in chest X-ray images.
• To leverage the enhanced classification capabilities of Quaternion neural networks combined with attention mechanisms.
• To improve upon existing diagnostic methods for early and reliable pneumonia identification.

Main Methods:

• Development of the Quaternion Channel-Spatial Attention (QCSA) network, integrating spatial and channel attention with a Quaternion residual network.
• Utilizing a Kaggle X-ray dataset for training and evaluation of the proposed QCSA network.
• Comparative analysis to demonstrate performance improvements with the integration of attention mechanisms in Quaternion Convolutional Neural Networks (QCNNs).

Main Results:

• The proposed QCSA network achieved a diagnostic accuracy of 94.53% for pneumonia detection.
• An Area Under the Curve (AUC) of 0.89 was obtained, indicating strong classification performance.
• Integration of the attention mechanism within the QCNN architecture demonstrably improved performance.

Conclusions:

• The QCSA network presents a promising and effective approach for the computer-aided diagnosis of pneumonia from chest X-ray images.
• The study highlights the benefits of combining Quaternion neural networks with attention mechanisms for medical image analysis.
• The developed method offers a potential solution to improve the accuracy and consistency of pneumonia diagnosis, particularly in infants.