Data augmentation based semi-supervised method to improve COVID-19 CT classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Computed tomography (CT) imaging is crucial for COVID-19 detection.
• Current deep learning methods for COVID-19 CT classification often require extensive labeled data, increasing costs and time.

Purpose of the Study:

• To develop a cost-effective and accurate semi-supervised learning framework for COVID-19 CT image classification.
• To address limitations in existing semi-supervised methods by focusing on improved pseudo-label generation.
• To enhance the performance of COVID-19 detection models using data augmentation.

Main Methods:

• A semi-supervised classification framework was designed, integrating data augmentation techniques.
• The classic teacher-student deep learning framework was modified.
• The Mixup data augmentation method was incorporated to improve pseudo-label accuracy and model performance.

Main Results:

• The proposed method demonstrated significant improvements in precision, F1 score, accuracy, and specificity across multiple datasets (COVID-CT, SARS-COV-2, Harvard Dataverse).
• Specific performance gains ranged from 7.59% to 38.29% compared to average values of other methods.
• The framework effectively enhances the classification of COVID-19 CT images with limited labeled data.

Conclusions:

• The developed semi-supervised framework with Mixup data augmentation offers a promising approach for accurate and efficient COVID-19 CT image classification.
• This method reduces reliance on large labeled datasets, making it more cost-effective.
• The study provides a valuable tool for improving early detection and diagnosis of COVID-19.