Machine Learning Based Multi-Parameter Modeling for Prediction of Post-Inflammatory Lung Changes.

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence in Healthcare
• Data Science

Background:

• Predicting lung function deficits after pulmonary infections like COVID-19 is often inaccurate.
• Lung function testing (LFT) and chest imaging are crucial for assessing post-infection lung changes.
• Existing methods may not fully capture the complexity of lung recovery.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting lung function deficits post-COVID-19.
• To identify key predictors of impaired lung function using a multi-parameter approach.
• To compare ML model performance against traditional assessments.

Main Methods:

• Prospective data collection from 140 COVID-19 survivors (CovILD study).
• Utilized lung function tests, AI-driven chest CT analysis (density, severity scoring), demographics, and symptoms.
• Developed and evaluated four ML algorithms (Random Forest, GBM, NN, SVM) for LFT prediction.

Main Results:

• Models accurately predicted reduced diffusion capacity for carbon monoxide (DLCO) with 82-85% accuracy and AUC 0.87-0.9.
• CT-derived features like opacity and severity score were significant predictors of DLCO impairment.
• No reliable models were established for FEV1 or FVC prediction.

Conclusions:

• Multi-parameter ML models integrating AI-enhanced CT data reliably predict LFT deficits.
• This AI-driven approach outperforms single markers and human radiologist assessments.
• The models show potential for improved diagnostics and personalized treatment strategies for post-COVID-19 lung complications.