PCXRNet: Pneumonia Diagnosis From Chest X-Ray Images Using Condense Attention Block and Multiconvolution Attention...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• The COVID-19 pandemic necessitates accurate and rapid diagnostic tools.
• Convolutional Neural Networks (CNNs) show promise for analyzing chest X-ray images.
• Existing CNNs often fail to fully leverage feature map inter- and intra-relationships.

Purpose of the Study:

• To propose an attention-based CNN, PCXRNet, for improved pneumonia diagnosis using chest X-ray images.
• To address limitations in feature map analysis within current CNN models for medical imaging.

Main Methods:

• Developed PCXRNet, incorporating a novel condense attention module (CDSE) for channel information utilization.
• Implemented a multi-convolution spatial attention module (MCSA) to focus on informative spatial regions.
• CDSE and MCSA were integrated in series to mitigate feature map redundancy and enhance information extraction.

Main Results:

• PCXRNet achieved high performance on a COVID-19 dataset.
• Achieved accuracy of 94.619%, recall of 94.753%, precision of 95.286%, and F1-score of 94.996% for COVID-19 diagnosis.

Conclusions:

• The proposed PCXRNet effectively diagnoses pneumonia from chest X-ray images.
• The attention-based modules (CDSE and MCSA) significantly improve the model's ability to interpret complex feature maps.
• PCXRNet demonstrates strong potential as a diagnostic tool for COVID-19.