COVIDC

COVIDC classifies chest computed tomography (CT) scans to detect COVID-19 infection and predict disease severity for diagnostic and triage purposes.

Key Features:

• Deep feature extraction: Deep feature maps are extracted from chest CT scans to support detection and severity prediction.
• Two-phase classification: A two-phase strategy first distinguishes COVID-19 from non-COVID conditions and then assesses severity of confirmed cases.
• Machine learning algorithms: Support Vector Machines (SVMs) and Random Forest classifiers are employed as core classification methods.
• Data scarcity management: The approach explicitly addresses data scarcity through the chosen machine learning strategies.
• Evaluation protocols: Performance was evaluated using 10-fold cross-validation, external validation datasets, and real-world assessment under an experienced radiologist's supervision.
• Reported performance: The method achieved an F1 score of 0.94 on the validation dataset and correctly classified 9 out of 10 CT scans in a practical setting.
• Comparative performance: Reported results indicate superior performance compared to existing state-of-the-art methods for COVID-19 detection in the evaluations described.

Scientific Applications:

• COVID-19 diagnosis from CT: Automated detection of COVID-19 versus non-COVID conditions using chest CT imaging.
• Severity prediction and triage: Prediction of disease severity for confirmed COVID-19 cases to inform clinical assessment and prioritization.
• Non-invasive screening: Use of chest CT as a rapid, non-invasive screening modality for suspected COVID-19 cases.

Methodology:

Deep feature maps are extracted from chest CT scans and processed in a two-phase classification workflow using Support Vector Machines (SVMs) and Random Forest classifiers; performance was assessed with 10-fold cross-validation, external validation datasets, and radiologist-supervised real-world evaluation.