Automatic COVID-19 and Common-Acquired Pneumonia Diagnosis Using Chest CT Scans.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Despite high COVID-19 vaccination rates, the disease remains a significant threat, necessitating advanced diagnostic tools.
• Accurate and rapid identification of COVID-19 and pneumonia is critical, especially for Intensive Care Unit (ICU) patient management.
• Monitoring disease progression or regression is vital in combating the ongoing epidemic.

Purpose of the Study:

• To develop and validate a secure Computer-Aided Diagnostic (CADx) system for COVID-19 detection and severity assessment using CT scans.
• To enhance diagnostic capabilities in identifying COVID-19 and differentiating it from Common-Acquired Pneumonia (CAP).
• To provide an efficient tool for monitoring disease status in critical care settings.

Main Methods:

• Merged public CT scan datasets to train lung and lesion segmentation models using diverse data distributions.
• Employed eight Convolutional Neural Network (CNN) models for classifying COVID-19 and CAP.
• Utilized Resnetxt101 Unet++ and Mobilenet Unet for segmentation, and Densenet201 for lesion classification and severity assessment.

Main Results:

• Achieved high accuracy (98.05%) and F1-score (98.70%) in lung and lesion segmentation using validated models.
• Demonstrated high performance in classifying COVID-19 and CAP, with Densenet201 achieving 90.47% accuracy and 100% recall for lesion detection.
• The system processed full CT scans in 19.70 seconds, showing efficiency in clinical application.

Conclusions:

• The developed CADx pipeline accurately detects and segments lesions associated with COVID-19 and CAP in CT scans.
• The system effectively differentiates between COVID-19, CAP, and normal lung exams.
• The findings confirm the system's efficiency and effectiveness in disease identification and severity assessment, supporting clinical decision-making.