COVID-19 lesion detection and segmentation-A deep learning method

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate and timely detection of COVID-19 cases is crucial for effective disease surveillance.
• Medical imaging, particularly chest CT, plays a significant role in diagnosing COVID-19.
• Deep learning offers potential for automating and enhancing the analysis of medical images.

Purpose of the Study:

• To develop and evaluate a deep learning model for screening positive COVID-19 cases in chest CT scans.
• To provide rapid and precise assistance for disease surveillance using medical imaging analysis.
• To improve the accuracy and efficiency of COVID-19 detection in radiological assessments.

Main Methods:

• Utilized deep learning, combining semantic segmentation and object detection techniques.
• Developed a novel end-to-end model leveraging Spatio-temporal features for COVID-19 lesion analysis.
• Incorporated a fully connected Conditional Random Field (CRF) for enhanced Region of Interest (ROI) segmentation.

Main Results:

• The proposed deep learning model demonstrated superior performance across various metrics compared to existing models.
• The method exhibited strong robustness when tested on processed and augmented imaging samples.
• The model effectively identified lesions associated with COVID-19 in chest CT scans.

Conclusions:

• Fusion of Spatio-temporal correlations enhances feature extraction for precise localization of lesions, including tiny ones.
• Temporal feature extraction, even in discrete form, significantly improves the precision of COVID-19 detection.
• The developed deep learning approach offers a valuable tool for robust and accurate COVID-19 screening in medical imaging.