A convolutional neural network-based COVID-19 detection method using chest CT images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• High-throughput screening for COVID-19 is crucial for disease control.
• Convolutional Neural Networks (CNNs) show potential for medical diagnosis using CT images.
• Chest CT scans can reveal pulmonary abnormalities indicative of COVID-19.

Purpose of the Study:

• To compare the performance of GoogLeNet and ResNet CNN architectures for COVID-19 detection using CT images.
• To evaluate the effectiveness of transfer learning (TL) in mitigating overfitting with limited CT data.
• To assess the diagnostic accuracy, recall rate, and F1 score for COVID-19 identification.

Main Methods:

• CT image preprocessing using fuzzy c-means (FCM) to isolate the pulmonary parenchyma.
• Multiscale transformation and RGB space construction of preprocessed images.
• Implementation and comparison of GoogLeNet and ResNet models, incorporating transfer learning (TL).

Main Results:

• The ResNet-50 model combined with transfer learning (ResNet-50-TL) achieved the highest diagnostic accuracy at 82.7%.
• ResNet-50-TL demonstrated a recall rate of 79.1% for COVID-19 detection.
• Transfer learning significantly improved model accuracy with limited sample sizes.

Conclusions:

• Deep learning applied to chest CT images presents a promising method for COVID-19 screening.
• The study highlights the potential for developing automated diagnostic systems for rapid COVID-19 screening.
• Transfer learning is effective in enhancing the accuracy of deep learning models for COVID-19 detection with limited data.