Clinical Applicable AI System Based on Deep Learning Algorithm for Differentiation of Pulmonary Infectious Disease

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Pneumonia classification from computed tomography (CT) scans is crucial for effective patient management.
• Distinguishing between different types of pneumonia (bacterial, fungal, viral, COVID-19) presents a diagnostic challenge for radiologists.
• Current diagnostic methods may lack the speed and accuracy required for timely intervention.

Purpose of the Study:

• To evaluate a novel deep learning (DL) based artificial intelligence (AI) system for classifying pneumonia types using CT scans.
• To develop and validate a clinically relevant machine learning (ML) system integrating imaging and clinical features for improved pneumonia diagnosis and triage.
• To identify key CT imaging and clinical features associated with specific pneumonia types.

Main Methods:

• A multi-center retrospective study analyzed 3,463 pneumonia CT images across four categories: bacterial (n=507), fungal (n=126), common viral (n=777), and COVID-19 (n=2,053).
• Deep learning (DL) models were employed for image-based classification of pulmonary infections.
• A machine learning (ML) model was developed using DL-derived imaging features and clinical data for risk interpretation.
• Algorithm performance was assessed using areas under the receiver operating characteristic curves (AUCs).

Main Results:

• DL models achieved high median AUCs for differentiating pulmonary infections: 99.5% (COVID-19), 98.6% (viral), 98.4% (bacterial), and 99.1% (fungal).
• The integrated ML model demonstrated excellent performance with AUCs of 99.7% (SARS-CoV-2), 99.4% (common virus), 98.9% (bacteria), and 99.6% (fungus).
• Key discriminative features identified included ground-glass opacity for COVID-19, larger lesions for viral pneumonia, older age for bacterial pneumonia, and consolidation for fungal pneumonia.

Conclusions:

• The developed AI system shows significant promise in classifying common pneumonia types and identifying influential factors for triage.
• The integration of DL image analysis and ML-based clinical feature interpretation enhances diagnostic accuracy.
• This AI system has the potential to assist clinicians in rapid and accurate pneumonia diagnosis, facilitating early therapeutic interventions.