Automated Detection of COVID-19 Using Machine Learning Analysis of White Blood Cell Flow Cytometry Images.

Area of Science:

• Medical Diagnostics
• Machine Learning in Healthcare
• Flow Cytometry

Background:

• Traditional COVID-19 diagnosis via real-time polymerase chain reaction (PCR) of nasopharyngeal swabs can lead to delays.
• Machine learning (ML) models may offer faster identification of SARS-CoV-2 infections by analyzing standard laboratory tests.
• Early diagnosis is crucial for timely patient management and protecting healthcare personnel.

Purpose of the Study:

• To develop and evaluate a machine learning model for rapid COVID-19 detection.
• To assess the model's performance using only white blood cell flow cytometry images, excluding clinical data.
• To determine if ML analysis of cytometry images can provide a viable alternative to current diagnostic methods.

Main Methods:

• A machine learning model was trained on flow cytometry images from 106 SARS-CoV-2 positive patients and 211 controls.
• Texture-feature analysis with 17 methods and three classifiers was employed.
• The Particle Swarm Optimization algorithm was used to combine optimal models, with performance evaluated using AUROC, sensitivity, specificity, precision, and F1-Score.

Main Results:

• The final ML model achieved an 88.96% diagnostic accuracy using five-fold cross-validation.
• The model demonstrated a sensitivity of 78.30%, specificity of 94.31%, precision of 87.83%, and an F1-score of 0.83.
• The area under the receiver operating characteristic curve (AUROC) was 0.86, indicating good discriminative ability.

Conclusions:

• The developed algorithm exhibits strong diagnostic performance for COVID-19.
• Cytometry image analysis shows potential for significantly advancing early COVID-19 diagnosis.
• This ML-based approach offers a promising, rapid diagnostic tool compared to models incorporating extensive clinical data.