Deep learning for predicting COVID-19 malignant progression.

Area of Science:

• Medical imaging and artificial intelligence
• Infectious disease epidemiology
• Clinical data analytics

Background:

• The COVID-19 pandemic overwhelmed healthcare systems, necessitating efficient patient triage.
• Initial symptom severity assessment may not predict rapid clinical deterioration in some COVID-19 patients.
• Early identification of patients at risk for severe disease is critical for resource allocation and treatment optimization.

Purpose of the Study:

• To develop and validate a deep learning-based early-warning system for predicting malignant progression of COVID-19.
• To improve the generalizability of the predictive model across different clinical settings.
• To identify key clinical indicators associated with COVID-19 deterioration.

Main Methods:

• Utilized deep learning techniques integrating computed tomography (CT) scans and clinical data from outpatients.
• Developed a domain adaptation approach to enhance model performance in multicenter studies.
• Evaluated model performance using the Area Under the Curve (AUC) metric.

Main Results:

• Achieved an AUC of 0.920 in a single-center study for predicting COVID-19 malignant progression.
• Demonstrated an average AUC of 0.874 in multicenter studies after applying domain adaptation.
• Identified Troponin, Brain natriuretic peptide, White cell count, Aspartate aminotransferase, Creatinine, and Hypersensitive C-reactive protein as crucial predictive indicators.

Conclusions:

• The developed deep learning system effectively predicts severe COVID-19 progression using CT and clinical data.
• Domain adaptation significantly improves the model's generalizability for multicenter applications.
• The identified biomarkers provide valuable insights into the pathophysiology of severe COVID-19 and can guide clinical decision-making.