WMR-DepthwiseNet: A Wavelet Multi-Resolution Depthwise Separable Convolutional Neural Network for COVID-19 Diagnosis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Accurate and timely COVID-19 detection is crucial for effective treatment and containment.
• Conventional COVID-19 testing methods often lack sensitivity.
• Medical imaging, including chest X-ray and computed tomography (CT), are explored for COVID-19 diagnosis.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN) model, WMR-DepthwiseNet, for automated COVID-19 identification.
• To address the challenge of limited radiograph datasets in COVID-19 detection.
• To enhance the accuracy and efficiency of COVID-19 screening using medical imaging.

Main Methods:

• Developed a depthwise separable convolution network integrated with a wavelet multiresolution analysis module (WMR-DepthwiseNet).
• The model learns features from both spatial and channel dimensions to improve diagnostic accuracy.
• Employed a depthwise separable connectivity framework for efficient feature map reusability, crucial for small datasets.

Main Results:

• Achieved 98.63% accuracy, 98.46% sensitivity, and 97.99% specificity on a chest X-ray dataset.
• Attained 96.83% accuracy, 97.78% sensitivity, and 96.22% specificity on a computed tomography (CT) dataset.
• Demonstrated high performance with limited training data, indicating its utility for rapid COVID-19 screening.

Conclusions:

• WMR-DepthwiseNet offers a robust and accurate method for COVID-19 detection from medical images.
• The model's ability to perform well with limited data is critical for managing rapidly evolving outbreaks.
• This AI-driven approach shows significant potential for improving COVID-19 screening and contributing to the fight against infectious diseases.