IRCM-Caps: An X-ray image detection method for COVID-19.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Deep Learning for Disease Detection

Background:

• Traditional COVID-19 testing (RT-PCR) is slow and prone to false negatives.
• Lung imaging offers a faster screening alternative, but existing deep learning models have limitations.
• Convolutional Neural Networks (CNNs) require large datasets, and basic Capsule Networks (CapsNets) have limited multi-classification accuracy.

Purpose of the Study:

• To propose a novel deep learning model integrating CNN and CapsNet for improved COVID-19 detection from X-ray images.
• To enhance model performance using attention and multi-branch lightweight modules.
• To address the limitations of existing models in terms of dataset size and classification accuracy.

Main Methods:

• A hybrid model combining CNN and CapsNet architectures was developed.
• Contrast Adaptive Histogram Equalization (CLAHE) was used for image preprocessing to improve contrast.
• Attention and multi-branch lightweight modules were incorporated to boost performance, with ReLU as the activation function.

Main Results:

• The proposed model demonstrated superior performance compared to basic CapsNet on a dataset of 1200 X-ray images (COVID-19, viral pneumonia, normal).
• Significant improvements were observed in accuracy, AUC, recall, and F1 scores across multiple CNN backbones integrated with CapsNet.
• Binary classification experiments also showed notable increases in accuracy, AUC, recall, and F1 scores.

Conclusions:

• The novel CNN-CapsNet model effectively combines the strengths of both architectures.
• The proposed model achieves good classification performance, particularly on smaller COVID-19 X-ray image datasets.
• This approach offers a promising tool for efficient and accurate COVID-19 screening using medical imaging.