Automated quantification of COVID-19 severity and progression using chest CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• COVID-19 pneumonia presents a significant challenge in diagnosis and monitoring.
• Accurate quantification and progression assessment of pneumonia are crucial for patient management.

Purpose of the Study:

• To develop and validate computer software for detecting, quantifying, and monitoring COVID-19 pneumonia progression using chest CT scans.
• To leverage deep learning and computer vision for automated analysis of lung infiltrates.

Main Methods:

• Deep learning algorithms were trained on chest CT scans to segment lung regions and vessels.
• Algorithms were developed to identify, quantify, and track the progression of pneumonic infiltrates in serial scans.
• Quantitative (Dice coefficient) and qualitative (radiologist Likert scale) assessments were performed.

Main Results:

• The software demonstrated strong agreement (Dice coefficient 81%) with manual delineations of pneumonic regions.
• High sensitivity (95%) and specificity (84%) were achieved in detecting large pneumonia regions.
• Radiologists found 95% of generated heatmaps acceptable for representing disease progression.

Conclusions:

• Computer software utilizing deep learning shows feasibility for detecting and quantifying COVID-19 pneumonia on CT scans.
• The developed software can generate heatmaps to visualize and assess disease progression.
• This technology holds potential for assisting in disease detection, progression monitoring, and treatment efficacy assessment.