Machine learning models using dual-phase CT radiomics for early detection of PRISm.

Area of Science:

• Pulmonary Medicine
• Radiology
• Artificial Intelligence in Healthcare

Background:

• Preserved Ratio Impaired Spirometry (PRISm) is an early stage of chronic obstructive pulmonary disease (COPD).
• Early identification of PRISm is critical for improving patient prognosis and potentially preventing disease progression.
• Current diagnostic methods may not fully capture the nuances of early lung function impairment.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for the early identification of PRISm.
• To compare the performance of models using inspiratory, expiratory, and dual-phase CT images.
• To assess the added value of radiomics features combined with clinical data for PRISm detection.

Main Methods:

• Development of multiple ML models, including logistic regression (LR), using clinical and radiomics features from CT scans.
• Prospective enrollment of 270 subjects for data collection.
• Validation of models using training, internal, and external datasets, analyzing inspiratory, expiratory, and dual-phase CT images.

Main Results:

• Combined clinical-radiomics models outperformed clinical models alone across all CT phases.
• LR-based combined models using expiratory or dual-phase CT achieved the best performance (AUCs up to 0.901 in training).
• Adding inspiratory CT data did not significantly improve model performance compared to single-phase expiratory CT.

Conclusions:

• Single-phase expiratory CT scans, when combined with clinical features and analyzed using LR models, are recommended for efficient PRISm identification.
• This approach supports early diagnosis and timely intervention for patients at risk of COPD.
• The findings highlight the potential of ML and radiomics in early lung disease detection using readily available CT data.