Covid-19 rapid test by combining a Random Forest-based web system and blood tests

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Infectious Disease Diagnostics

Background:

• COVID-19, caused by SARS-CoV-2, presents significant global public health challenges.
• Over 2 million deaths underscore the urgent need for effective diagnostic and management tools.
• Limited testing availability can hinder timely diagnosis and patient care.

Purpose of the Study:

• To develop and evaluate Heg.IA, a web-based artificial intelligence solution for optimizing COVID-19 diagnosis.
• To assist healthcare professionals in decision-making regarding COVID-19 diagnosis and patient hospitalization levels (ward, semi-ICU, ICU).

Main Methods:

• Implementation of a Random Forest machine learning model with 90 decision trees.
• Input of 41 hematological and arterial gasometry parameters from routine blood tests.
• Development of a web solution for automated diagnostic reporting and hospitalization recommendations.

Main Results:

• Achieved high accuracy (92.89%) for COVID-19 diagnosis, with a kappa index of 0.858 and AUC of 0.984.
• Demonstrated excellent performance for hospitalization indication, with accuracies exceeding 99%.
• The computationally efficient AI model provides reliable diagnostic support.

Conclusions:

• The Heg.IA system effectively supports COVID-19 diagnosis and hospitalization decisions.
• This AI-driven approach offers a valuable alternative to overcome testing limitations.
• Heg.IA has the potential to enhance healthcare system efficiency during infectious disease outbreaks.