Deep forest model for diagnosing COVID-19 from routine blood tests.

Area of Science:

• Medical Informatics
• Machine Learning
• Computational Biology

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Limited availability of traditional swab tests poses a challenge for widespread screening.
• Clinical and laboratory data offer a potential alternative for disease detection.

Purpose of the Study:

• To develop and evaluate a machine learning model for COVID-19 diagnosis using clinical and laboratory data.
• To leverage an ensemble-based deep forest (DF) method for enhanced prediction accuracy.
• To provide a rapid screening tool for COVID-19, especially in resource-limited settings.

Main Methods:

• Implementation of a deep forest (DF) ensemble model.
• Utilizing a cascade structure with Extra Trees, XGBoost, and LightGBM classifiers.
• Training and validation on two public datasets comprising clinical and laboratory parameters.

Main Results:

• The DF model achieved high diagnostic performance: 99.5% accuracy, 95.28% sensitivity, and 99.96% specificity.
• Performance metrics are comparable to existing machine learning techniques.
• The model demonstrates effectiveness in identifying COVID-19 cases from routine data.

Conclusions:

• The proposed deep forest model is a highly accurate and efficient tool for COVID-19 screening.
• This approach can supplement traditional testing methods, particularly where access is restricted.
• Machine learning models utilizing routine data show significant promise for infectious disease diagnostics.