MHA-CoroCapsule: Multi-Head Attention Routing-Based Capsule Network for COVID-19 Chest X-Ray Image Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• COVID-19 poses a global health threat, straining healthcare systems worldwide.
• Efficient and accurate detection methods for COVID-19 are crucial but challenging.
• AI-assisted deep learning shows promise for automated COVID-19 diagnosis from chest X-rays, yet requires large annotated datasets.

Purpose of the Study:

• To develop a novel deep learning model for fast and accurate COVID-19 detection using chest X-ray images.
• To address the challenge of limited annotated data for AI model training in medical imaging.
• To propose an efficient AI diagnostic tool that assists healthcare professionals.

Main Methods:

• A capsule network model named MHA-CoroCapsule was developed, incorporating convolutional layers and two capsule layers.
• A non-iterative, parameterized multi-head attention routing algorithm was employed to quantify relationships between capsule layers.
• Experiments were conducted on a combined dataset of normal, non-COVID pneumonia, and COVID-19 chest X-ray images.

Main Results:

• The MHA-CoroCapsule model achieved high performance metrics: 97.28% accuracy, 97.36% recall, and 97.38% precision.
• The model demonstrated effectiveness even with a limited number of training samples.
• Compared to transfer learning and deep feature extraction, MHA-CoroCapsule showed encouraging results with fewer trainable parameters and no need for pretraining.

Conclusions:

• MHA-CoroCapsule offers a promising solution for rapid and accurate COVID-19 diagnostics from chest X-rays.
• The model overcomes limitations of data scarcity and computational burden associated with traditional deep learning approaches.
• This AI-driven method can potentially alleviate pressure on healthcare systems by improving diagnostic efficiency and accuracy.