COVID-19 diagnosis utilizing wavelet-based contrastive learning with chest CT images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Chest CT scans are crucial for COVID-19 diagnosis, but annotation costs limit large-scale AI model training.
• Existing computer-aided diagnosis (CAD) systems require extensive labeled data, posing a challenge due to data scarcity.

Purpose of the Study:

• To develop an automated and accurate COVID-19 diagnosis system utilizing limited labeled chest CT images.
• To address the challenge of insufficient annotated medical data for training robust CAD systems.
• To enhance the performance of COVID-19 detection through innovative self-supervised learning techniques.

Main Methods:

• Implemented a framework based on self-supervised contrastive learning (SSCL).
• Integrated two-dimensional discrete wavelet transform with contrastive learning to maximize feature utilization.
• Utilized a redesigned COVID-Net encoder for task specificity and learning efficiency.
• Employed a novel pretraining strategy and an auxiliary classification task to improve generalization and performance.

Main Results:

• The system achieved high performance metrics: 93.55% accuracy, 91.59% recall, 96.92% precision, and 94.18% F1-score.
• Demonstrated superior performance compared to existing schemes in COVID-19 diagnosis using CT images.
• Validated the effectiveness of SSCL and integrated enhancements in handling limited labeled medical data.

Conclusions:

• The proposed automated system effectively diagnoses COVID-19 from limited CT images, outperforming current methods.
• Self-supervised contrastive learning offers a viable solution for developing accurate CAD systems with scarce annotated medical data.
• The integrated approach of wavelet transform, COVID-Net, and auxiliary tasks significantly boosts diagnostic accuracy and generalization ability.