Periphery-aware COVID-19 diagnosis with contrastive representation enhancement

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computer-aided diagnosis (CAD) is crucial for rapid COVID-19 screening.
• Distinguishing COVID-19 from other pneumonias (e.g., H1N1, CAP) using CT remains challenging.
• Improving diagnostic performance in complex multi-type pneumonia classification is essential.

Purpose of the Study:

• To propose a novel periphery-aware COVID-19 diagnosis approach with contrastive representation enhancement.
• To accurately identify COVID-19 from influenza-A (H1N1) viral pneumonia, community-acquired pneumonia (CAP), and healthy subjects using chest CT images.

Main Methods:

• Developed an unsupervised Periphery-aware Spatial Prediction (PSP) task to integrate spatial patterns into deep networks.
• Implemented an adaptive Contrastive Representation Enhancement (CRE) mechanism to capture intra-class similarity and inter-class differences.
• Integrated PSP and CRE for highly discriminative representations in COVID-19 screening.

Main Results:

• The proposed approach demonstrated effectiveness in COVID-19 diagnosis using both volume-level and slice-level CT images.
• Comprehensive evaluations were conducted on a large-scale constructed dataset and two public datasets.
• The integrated PSP and CRE methods yielded highly discriminative representations for accurate screening.

Conclusions:

• The novel periphery-aware approach with contrastive representation enhancement significantly improves COVID-19 diagnosis accuracy.
• The PSP task and CRE mechanism are effective in enhancing deep network performance for pneumonia classification.
• This method offers a promising tool for computer-aided diagnosis of COVID-19 in complex clinical scenarios.