Deep Transfer Learning for COVID-19 Prediction: Case Study for Limited Data Problems.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• The COVID-19 pandemic highlighted the need for rapid diagnostic tools.
• Chest X-rays are a common imaging modality for respiratory diseases.
• Automated analysis of medical images can improve diagnostic efficiency.

Purpose of the Study:

• To develop and evaluate deep learning models for automatic COVID-19 detection from chest X-ray images.
• To compare the accuracy of different convolutional neural network (CNN) models against expert diagnosis.
• To assess the clinical potential of AI in early COVID-19 detection.

Main Methods:

• Utilized deep transfer learning with pre-trained CNN models (InceptionNetV3, Inception ResNetV2, NASNetlarge).
• Trained models on chest X-ray images from COVID-19 patients and healthy individuals.
• Evaluated model performance with and without data augmentation techniques.

Main Results:

• InceptionNetV3 demonstrated the highest accuracy, reaching 98.90% without data augmentation and 98.63% with augmentation.
• All tested models showed a tendency to overfit with limited datasets when data augmentation was not used.
• The study suggests Inception ResNetV2 and NASNetlarge may perform better with larger datasets.

Conclusions:

• Deep transfer learning offers a promising approach for automated COVID-19 detection from chest X-rays.
• The developed model can support clinical decision-making by providing accurate and effective diagnostic assistance.
• This research provides insights into applying transfer learning for infectious disease detection.