An externally validated fully automated deep learning algorithm to classify COVID-19 and other pneumonias on chest...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Distinguishing between COVID-19, influenza, and other pneumonias using CT scans is crucial for effective patient management.
• Accurate and rapid diagnostic tools are needed to address the challenges posed by respiratory infections.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) framework for classifying computed tomography (CT) scans.
• The AI framework aims to differentiate between COVID-19, influenza/community-acquired pneumonia (CAP), and no infection.

Main Methods:

• A three-dimensional convolutional neural network (CNN) based on inflated 3D Inception architecture was utilized.
• The model was trained and validated on CT images from a large cohort of adult patients, with subsequent evaluation on internal and external test sets.

Main Results:

• The AI model demonstrated excellent performance on an external validation set, achieving an area under the curve (AUC) of 0.90 for COVID-19, 0.92 for influenza/CAP, and 0.92 for no infection.
• Automatic lung segmentation and abnormality detection reduced analysis time to 56 seconds per scan, decreasing computational load.

Conclusions:

• The developed AI framework offers a rapid and accurate diagnostic solution for patients presenting with symptoms suggestive of COVID-19 or influenza.
• This AI-powered tool has the potential to significantly improve the diagnostic workflow in clinical settings.