Lightweight deep learning models for detecting COVID-19 from chest X-ray images

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Deep learning has shown significant success in medical imaging analysis.
• COVID-19 detection from medical images is a critical area of research.
• Limited data availability poses a challenge for training robust deep learning models.

Purpose of the Study:

• To develop an effective deep learning approach for COVID-19 detection using chest X-ray images.
• To address the challenge of limited data by generating synthetic images.
• To evaluate the performance and reliability of proposed models against existing methods.

Main Methods:

• Utilized a conditional generative adversarial network (cGAN) for synthetic medical image generation to augment limited datasets.
• Developed two lightweight deep learning models tailored for COVID-19 detection.
• Conducted experiments for binary classification (COVID-19 vs. Normal) and multi-classification (COVID-19 vs. Normal vs. Bacterial Pneumonia).

Main Results:

• The best binary classification model achieved 98.7% accuracy, 100% sensitivity, and 98.3% specificity.
• The best three-class model achieved 98.3% accuracy, 99.3% sensitivity, and 98.1% specificity.
• Proposed models demonstrated superior robustness and reliability compared to a baseline ResNet8 model in COVID-19 detection.

Conclusions:

• The proposed deep learning approach, enhanced by cGAN-generated data, is effective for COVID-19 detection from chest X-rays.
• Lightweight models offer competitive performance and are suitable for limited data scenarios.
• The developed models show promise as reliable tools for radiologists and physicians in diagnosing COVID-19.