Accurately Differentiating Between Patients With COVID-19, Patients With Other Viral Infections, and Healthy...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Infectious Disease Diagnostics

Background:

• Accurate identification of COVID-19 patients using non-PCR methods is crucial for clinical outcomes.
• A comprehensive understanding of biomedical features and analytical approaches for early COVID-19 detection is lacking.

Purpose of the Study:

• To combine clinical, lab, and computed tomography (CT) imaging data to differentiate between healthy individuals, COVID-19 patients, and non-COVID viral pneumonia.
• To achieve accurate differentiation, especially during the early stages of infection.

Main Methods:

• Recruited 214 nonsevere COVID-19, 148 severe COVID-19, 198 healthy, and 129 non-COVID viral pneumonia patients.
• Utilized 3 input modalities: clinical information (23 features), lab testing results (10 features), and CT scans.
• Developed a deep learning model for CT feature extraction and integrated it with machine learning models (k-NN, random forest, SVM) using 43 combined features.

Main Results:

• Multimodal features significantly improved performance compared to single modalities.
• Machine learning models achieved high overall prediction accuracy (95.4%-97.7%).
• Class-specific prediction accuracy ranged from 90.6% to 99.9%.

Conclusions:

• The hybrid deep learning-machine learning framework offers a novel and effective approach for clinical applications, outperforming single-modality benchmarks.
• Findings from this large-scale study and analytical workflow can support clinical decision-making for COVID-19 diagnosis.
• The methodology can be applied to other clinical scenarios involving high-dimensional multimodal biomedical data.