MTSS-AAE: Multi-task semi-supervised adversarial autoencoding for COVID-19 detection based on chest X-ray images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Efficient diagnosis of COVID-19 is crucial for disease control.
• Chest X-rays (CXRs) offer an economical diagnostic approach but require significant human expertise.
• Deep learning-based computer-aided diagnosis (CAD) holds potential for automated, cost-effective COVID-19 detection in CXRs.

Purpose of the Study:

• To address the challenge of limited labeled data for COVID-19 detection in CXRs.
• To develop a highly accurate and robust deep learning model for COVID-19 detection.
• To improve the cost-effectiveness of CXR-based COVID-19 diagnosis through automation.

Main Methods:

• Proposed a multi-task semi-supervised learning (MTSSL) framework.
• Utilized auxiliary tasks (Pneumonia, Lung Opacity, Pleural Effusion) with publicly available data (ChesXpert dataset).
• Integrated an adversarial autoencoder (AAE) within the MTSSL framework for enhanced feature learning and semi-supervised learning.

Main Results:

• The MTSSL framework significantly improved COVID-19 detection performance, especially with limited labeled data.
• The integration of AAE enhanced feature discriminability and model generalization through semi-supervised learning.
• The proposed model achieved state-of-the-art performance on the largest publicly available COVID-19 dataset.

Conclusions:

• The developed MTSSL framework effectively leverages auxiliary data to enhance COVID-19 detection from CXRs.
• Semi-supervised learning, combined with adversarial autoencoders, improves model robustness and accuracy.
• This approach offers a promising, automated, and cost-effective solution for COVID-19 diagnosis using medical imaging.