Differentiating novel coronavirus pneumonia from general pneumonia based on machine learning

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Infectious Disease Diagnostics

Background:

• Chest CT scans are vital for diagnosing COVID-19 due to high sensitivity.
• Machine learning excels at analyzing complex medical images for accurate diagnoses.

Purpose of the Study:

• To develop and validate a machine learning framework for differentiating COVID-19 from general pneumonia using chest CT images.
• To leverage AI for improved accuracy in diagnosing COVID-19, aiding clinicians.

Main Methods:

• An integrated machine learning framework was developed using chest CT images from 73 COVID-19 and 27 general pneumonia cases.
• Region of interest delineation based on ground-glass opacities (GGOs) was performed, followed by extraction of 34 statistical texture features.
• The ReliefF algorithm was used for feature selection, and an ensemble of bagged tree (EBT) classifier was employed for differentiation.

Main Results:

• The proposed method achieved a classification accuracy of 94.16%, sensitivity of 88.62%, and specificity of 100.00%.
• The area under the receiver operating characteristic curve (AUC) reached 0.99, indicating excellent diagnostic performance.
• The EBT algorithm demonstrated high transferability, efficiency, and accuracy in differentiating COVID-19 from general pneumonia.

Conclusions:

• Machine learning, particularly the EBT algorithm with statistical textural features from GGOs, accurately differentiates COVID-19 from general pneumonia.
• This AI-driven approach offers high specificity and sensitivity, proving beneficial for inexperienced doctors in diagnosing COVID-19.
• The method is essential for controlling the spread of COVID-19 by enabling earlier and more accurate diagnoses.