Related Experiment Video
Updated: Nov 4, 2025
08:20
Author Spotlight: AI-Driven Trypanosome Species Detection from Microscopic Images
Published on: October 27, 2023
2.0K
COVID-19 diagnosis by routine blood tests using machine learning
Matjaž Kukar1,2, Gregor Gunčar1,3, Tomaž Vovko4
1Smart Blood Analytics Swiss SA, Höschgasse 25, 8008, Zurich, Switzerland.
Scientific Reports
|May 25, 2021
Summary
A machine learning model using routine blood tests aids COVID-19 diagnosis. This tool, with 81.9% sensitivity and 97.9% specificity, assists physicians in identifying potential cases for further RT-PCR confirmation.
Area of Science:
- Medical diagnostics
- Machine learning in healthcare
- Hematology
Background:
- Routine blood parameters change in COVID-19 patients, complicating diagnosis.
- Existing diagnostic methods like RT-PCR and chest CT have limitations.
- Need for accessible and rapid diagnostic tools for COVID-19.
Purpose of the Study:
- To develop and validate a machine learning model for COVID-19 diagnosis using routine blood tests.
- To identify key blood parameters indicative of COVID-19.
- To assess the diagnostic performance of the model.
Main Methods:
- Constructed a machine learning model trained on blood tests from 5333 patients with various infections and 160 COVID-19 patients.
- Utilized XGBoost algorithm for feature importance scoring.
- Employed t-SNE visualization to analyze blood parameter patterns.
Main Results:
- Achieved a cross-validated AUC of 0.97 with 81.9% sensitivity and 97.9% specificity.
- Identified MCHC, eosinophil count, albumin, INR, and prothrombin activity as key diagnostic parameters.
- Observed that severe COVID-19 blood profiles resemble bacterial infections more than viral ones.
Conclusions:
- The machine learning model demonstrates high accuracy for COVID-19 diagnosis, comparable to RT-PCR and CT scans.
- Routine blood tests can serve as a valuable tool for initial COVID-19 screening.
- This approach can complement existing diagnostic methods, improving patient management.
