DPDH-CapNet: A Novel Lightweight Capsule Network with Non-routing for COVID-19 Diagnosis Using X-ray Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• COVID-19 diagnosis remains critical, with deep learning methods often needing extensive labeled data.
• Existing capsule networks for COVID-19 detection face challenges with computational cost and complex routing mechanisms.

Purpose of the Study:

• To develop a lightweight capsule network, DPDH-CapNet, for enhanced automated diagnosis of COVID-19 from chest X-ray images.
• To overcome the limitations of existing deep learning models, particularly their reliance on large datasets and computational expense.

Main Methods:

• A novel feature extractor using depthwise convolution (D), point convolution (P), and dilated convolution (D) to capture local and global dependencies.
• A classification layer employing homogeneous (H) vector capsules with an adaptive, non-iterative, and non-routing mechanism.
• Experimental validation on combined datasets of normal, pneumonia, and COVID-19 chest X-ray images.

Main Results:

• DPDH-CapNet achieved significant parameter reduction (9x) compared to state-of-the-art capsule networks.
• The model demonstrated faster convergence, improved generalization, and high performance metrics: 97.99% accuracy, 98.05% precision, 98.02% recall, and 98.03% F-measure.
• The proposed model effectively functions without pre-training or large training datasets, unlike transfer learning methods.

Conclusions:

• DPDH-CapNet offers a computationally efficient and effective solution for COVID-19 detection using chest X-rays.
• The model's ability to perform well with limited data and without pre-training makes it highly suitable for clinical applications.
• This advancement contributes to the urgent need for improved diagnostic technologies for infectious diseases like COVID-19.