COVID-19 detection method based on SVRNet and SVDNet in lung x-rays

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate and rapid diagnosis of coronavirus disease 2019 (COVID-19) is crucial for patient management and public health.
• Lung X-rays are a common imaging modality for assessing respiratory conditions, including COVID-19.

Purpose of the Study:

• To develop and evaluate novel deep learning models, separable VGG-ResNet (SVRNet) and separable VGG-DenseNet (SVDNet), for improved COVID-19 detection from lung X-rays.
• To design a detection system that enhances diagnostic speed and accuracy.

Main Methods:

• Utilized a dataset of 1560 lung X-ray images from the COVID-19 Radiography Database.
• Fine-tuned and trained established image classification models (VGG16, ResNet50, InceptionV3, Xception) using deep learning and transfer learning.
• Developed and evaluated two new models, SVRNet and SVDNet, on a test set of 312 images.

Main Results:

• SVRNet and SVDNet achieved high classification accuracy, sensitivity, and specificity (e.g., 99.13% accuracy for SVRNet, 99.37% for SVDNet).
• These models demonstrated significant improvements in performance metrics compared to the VGG16 network.
• SVRNet and SVDNet substantially reduced the number of parameters (by 61.56% and 55.31% respectively), indicating greater efficiency.

Conclusions:

• The proposed SVRNet and SVDNet models offer a significant advancement in the automated detection of COVID-19 from lung X-rays.
• These models provide a faster and more accurate diagnostic tool with reduced computational requirements.
• The enhanced efficiency and accuracy make SVRNet and SVDNet promising for clinical application in COVID-19 screening.