Capturing COPD heterogeneity: anomaly detection and parametric response mapping comparison for phenotyping on chest...

Area of Science:

• Radiology
• Pulmonary Medicine
• Artificial Intelligence in Medicine

Background:

• Chronic Obstructive Pulmonary Disease (COPD) presents a significant global health challenge, necessitating advanced diagnostic tools for early identification and precise phenotyping.
• Current diagnostic methods require enhancement to capture the full spectrum of COPD heterogeneity and its dynamic progression.

Purpose of the Study:

• To compare a novel self-supervised anomaly detection approach with traditional Parametric Response Mapping (PRM) for characterizing COPD on chest CT.
• To explore the spatial and quantitative relationships between these methods and their correlation with pulmonary function tests (PFTs).
• To gain additional insights into the transitional stages of COPD.

Main Methods:

• Retrospective analysis of 1,310 non-contrast inspiratory and expiratory CT scans from the COPDGene dataset, including never-smokers, GOLD 0, and COPD patients (GOLD 1-4).
• Application of a self-supervised anomaly detection approach to quantify regional lung abnormalities as deviations.
• Comparison of anomaly scores with PRM volumes (emphysema: PRM^Emph, functional small-airway disease: PRM^fSAD), PCA, and clustering, alongside evaluation of PFT correlations.

Main Results:

• Self-supervised latent space visualization revealed distinct spatial patterns separating emphysema and air trapping.
• Anomaly scores and specific clusters showed significant trends with increasing GOLD stage, unlike PRM^Emph and PRM^fSAD.
• Anomaly scores demonstrated moderate correlations with PFTs and strong correlations with PRM^Emph and PRM^fSAD, significantly improving multivariate model fitting for clinical parameters.

Conclusions:

• The anomaly detection approach and PRM synergistically capture COPD heterogeneity, offering complementary strengths.
• Integrating these methods provides a promising pathway for a deeper understanding of COPD pathophysiology.
• This integrated approach may lead to more tailored diagnostic and intervention strategies, ultimately improving patient outcomes.