An Ensemble Learning Model for COVID-19 Detection from Blood Test Samples.

Area of Science:

• Artificial Intelligence in Medicine
• Machine Learning for Diagnostics
• Computational Biology

Background:

• Current artificial intelligence (AI) methods show promise for COVID-19 diagnosis but face limitations.
• Challenges include limited datasets, class imbalance, and high misclassification rates with existing models.
• Real-time detection using reverse transcription polymerase chain reaction (RT-PCR) data needs improvement.

Purpose of the Study:

• To investigate ensemble learning for COVID-19 detection using routine laboratory blood test results.
• To develop and apply prediction models for effective COVID-19 diagnosis.
• To create an AI-driven system to assist clinicians in diagnosing COVID-19.

Main Methods:

• An ensemble machine learning system was developed using a two-stage classification approach.
• First-stage classification utilized custom convolutional neural network (CNN) models.
• Second-stage classification involved 15 supervised machine learning algorithms, including ExtraTrees and AdaBoost.

Main Results:

• An ensemble model combining deep neural networks (DNN) and ExtraTrees achieved 99.28% mean accuracy and 99.4% AUC.
• The AdaBoost model demonstrated 99.28% mean accuracy and 98.8% AUC on the San Raffaele Hospital dataset.
• The proposed method outperformed other state-of-the-art COVID-19 diagnostic approaches on the same dataset.

Conclusions:

• Ensemble learning approaches, particularly DNN with ExtraTrees, show high efficacy for COVID-19 detection from blood tests.
• The developed AI system effectively aids clinicians in diagnosing COVID-19.
• This method offers a superior alternative to existing COVID-19 diagnostic techniques.