COVID-19 severity detection using machine learning techniques from CT-images

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pulmonology

Background:

• COVID-19, a global pandemic, significantly impacts lung health, causing severe respiratory distress.
• Accurate differentiation of COVID-19 from other lung conditions and precise severity assessment are critical for patient management.
• Current diagnostic challenges necessitate advanced tools for timely and effective intervention.

Purpose of the Study:

• To develop and validate a two-step computational approach for detecting COVID-19 infection from lung CT images.
• To determine the severity of COVID-19 illness using integrated image and clinical data.
• To improve patient care by enabling focused attention on high-risk individuals based on severity classification.

Main Methods:

• Feature extraction from lung CT images using pre-trained models: AlexNet, DenseNet-201, and ResNet-50.
• COVID-19 detection using an Artificial Neural Network (ANN) model.
• Severity classification (High, Moderate, Low) via Cubic Support Vector Machine (SVM) integrating image features and clinical data.

Main Results:

• Achieved 92.0% accuracy, 96.0% sensitivity, and 91.44% F1-Score for COVID-19 detection.
• Attained 90.0% overall accuracy for three-class COVID-19 severity detection.
• Demonstrated the efficacy of the integrated approach in identifying infection and grading severity.

Conclusions:

• The proposed two-step method effectively detects COVID-19 infection and classifies its severity from lung CT scans.
• The integration of deep learning models and clinical data offers a robust solution for managing COVID-19 patients.
• This AI-driven approach can aid clinicians in prioritizing care for patients with severe conditions.