A Precise Method to Detect Post-COVID-19 Pulmonary Fibrosis Through Extreme Gradient Boosting

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Radiology

Background:

• COVID-19 is linked to pulmonary fibrosis, a significant cause of mortality.
• Early detection of post-COVID-19 pulmonary fibrosis is critical for patient outcomes.
• Automatic disease detection systems assist clinicians in achieving rapid and accurate diagnoses.

Purpose of the Study:

• To propose an ensemble machine learning architecture for detecting COVID-19-associated pulmonary fibrosis.
• To optimize Extreme Gradient Boosting (XGBoost) for predicting severe COVID-19 patients who develop pulmonary fibrosis post-discharge.
• To leverage Electronic Health Records (EHR) and High-resolution computed tomography (HRCT) data for risk analysis.

Main Methods:

• Utilized an ensemble machine learning approach, specifically focusing on Extreme Gradient Boosting (XGBoost).
• Tuned hyperparameters of the XGBoost model to enhance predictive performance.
• Analyzed a dataset of 1175 COVID-19 patients, including EHR and HRCT scans, with 725 pulmonary fibrosis cases and 450 normal cases.

Main Results:

• Achieved high performance metrics: 98% accuracy, 99% precision, and 99% sensitivity.
• Demonstrated the model's effectiveness in predicting pulmonary fibrosis risk in COVID-19 patients.
• The model successfully differentiated between patients with and without pulmonary fibrosis.

Conclusions:

• The proposed XGBoost-based model is the first to integrate EHR and HRCT data for detecting COVID-19-associated pulmonary fibrosis.
• This approach assists clinicians in identifying severe COVID-19 cases at risk of developing pulmonary fibrosis.
• Early identification facilitates timely intervention, potentially reducing life-threatening conditions associated with post-COVID-19 lung damage.