A Deep Learning Model Based on Capsule Networks for COVID Diagnostics through X-ray Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• X-ray diagnostics are crucial for detecting diseases but require expert availability.
• Current deep learning models like Convolutional Neural Networks (CNNs) have limitations in medical image analysis.
• Capsule Networks (CapsNet) offer a promising alternative to address these limitations.

Purpose of the Study:

• To develop an improved deep learning model for detecting diseases in chest X-ray images.
• To evaluate the effectiveness of Capsule Networks (CapsNet) in diagnosing conditions like pneumonia and COVID-19.
• To propose a novel model, DRCaps, that enhances CapsNet performance and efficiency.

Main Methods:

• Utilized Capsule Networks (CapsNet) for chest X-ray analysis.
• Developed an improved model named DRCaps, incorporating a dilation rate (dr) parameter.
• Managed 226 × 226 resolution images and employed a reconstruction stage to prevent overfitting.
• Avoided max-pooling operations, using stride and dilation rate for downsampling.

Main Results:

• Achieved a high accuracy of 90% on a dataset of 16,669 chest X-ray images.
• The DRCaps model has a size of 11M, demonstrating efficiency.
• DRCaps outperformed comparable models in accuracy, parameter count, and handling image sizes.
• Demonstrated the effectiveness of the reconstruction stage in preventing overfitting.

Conclusions:

• DRCaps offers a superior deep learning solution for chest X-ray disease detection compared to existing models.
• The model is effective in diagnosing conditions such as pneumonia and COVID-19.
• DRCaps provides a simplified approach, avoiding complex preprocessing and data augmentation.