A multidimensional feature-based pulmonary function prediction model derived from paired inspiratory-expiratory CT...

Area of Science:

• Medical Imaging
• Pulmonary Medicine
• Computational Biology

Background:

• Pulmonary function tests (PFTs) are crucial for assessing respiratory disorders like COPD but are effort-dependent and have limited accessibility.
• Developing noninvasive methods for pulmonary function assessment is essential for broader clinical application and repeated monitoring.

Purpose of the Study:

• To develop a multidimensional feature-based model using paired inspiratory-expiratory CT images for noninvasive prediction of key pulmonary function indices.
• To quantitatively predict the ratio of forced expiratory volume in one second to forced vital capacity (FEV1/FVC) and percent predicted forced expiratory volume in one second (FEV1%pred).

Main Methods:

• Constructed a multidimensional feature set integrating clinical variables, parametric response mapping (PRM) lung density features, ventilation heterogeneity features, and local biomechanical response features.
• Employed an elastic-net regression model to predict FEV1/FVC and FEV1%pred.
• Utilized SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The multidimensional model outperformed models using only clinical or PRM features.
• Achieved R2 of 0.572 (MAE 8.38) for FEV1/FVC and R2 of 0.472 (MAE 13.65) for FEV1%pred in the test set.
• Identified specific ventilation (Vol0-50) and biomechanical (fSAD_J_Skew, fSAD_J_Kurt, Emph_J_Skew) features as critical predictors.

Conclusions:

• The developed CT-based model offers a noninvasive, quantitative approach to predict pulmonary function indices.
• This method can complement PFTs for opportunistic functional assessment and longitudinal follow-up in patients undergoing dual-phase CT.
• Highlights the potential of integrating advanced imaging features for improved respiratory assessment.