Machine Learning for COVID-19 Detection Using Biomarkers and X-Ray Imaging
Author Information
Author(s): Tur Kagan
Primary Institution: Ahi Evran University, Kirsehir, Turkey
Hypothesis
Can a multi-modal machine learning model combining clinical biomarkers and chest X-ray images enhance COVID-19 diagnostic accuracy?
Conclusion
The multi-modal machine learning approach significantly improves diagnostic accuracy for COVID-19 compared to single-modality models.
Supporting Evidence
- The Gradient Boosting + VGG model achieved an AUC-ROC of 0.94, indicating high diagnostic performance.
- SHAP and LIME analyses identified key biomarkers and imaging features that contributed to model predictions.
- Stratified k-fold cross-validation was used to ensure robust evaluation and prevent data leakage.
Takeaway
This study shows that using both blood tests and X-ray images together helps doctors find COVID-19 more accurately.
Methodology
The study used a dataset of 250 patients, applying Random Forest and Gradient Boosting for biomarkers, and ResNet and VGG CNN architectures for imaging data, with a late-fusion strategy for integration.
Potential Biases
Potential bias due to the dataset being sourced from specific medical hospitals.
Limitations
The dataset size is modest, and findings may not generalize beyond the specific population studied.
Participant Demographics
Mean age of participants was 55.4 years, with 52% male and 48% female, including diverse cases from Middle and Greater Anatolia.
Statistical Information
P-Value
0.0313
Confidence Interval
95% CI: 0.91–0.96
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website