Automated three-dimensional diaphragm-related surface morphometry from paired chest CT in chronic obstructive pulmonary disease.

BACKGROUND

Chest CT supports quantitative COPD phenotyping, but diaphragm assessment remains limited by poor muscle conspicuity and labour-intensive analysis. We developed a fully automated, annotation-light framework extracting three-dimensional diaphragm-related inferior lung-interface surfaces from paired inspiratory-expiratory CT as surrogates of diaphragmatic dome geometry.

METHODS

In this retrospective multicentre analysis of prospectively enrolled real-world cohort participants, paired CT examinations underwent automated lung isolation, mesh-based inferior-interface extraction, surface completion, and morphometry. After phase-ordering quality control, the internal cohort included 516 paired examinations. External analyses included 35 of 49 COPD participants with successful extraction for cross-scanner feasibility and 60 individuals for COPD-non-COPD replication.

RESULTS

Automated dome-level localisation agreed strongly with independent readers (intraclass correlation coefficients, 0.92-0.96). COPD examinations showed larger expiratory surface area, longer unadjusted expiratory apico-diaphragmatic lung dimension, altered inspiratory convexity, and lower excursion. Bilateral surface-area differences persisted after adjustment for achieved expiratory CT lung volume, whereas cranio-caudal extent findings were attenuated. In exploratory COPD-non-COPD discrimination, adding diaphragm-related descriptors to expiratory LAA% increased the AUC from 0.780 (95% CI 0.738-0.820) to 0.824 (95% CI 0.794-0.867). Principal morphometric group differences were replicated externally; the ROC model was not externally validated.

CONCLUSIONS

Automated paired-CT morphometry of the diaphragm-related inferior lung interface was feasible and reproducible, captured COPD-related thoraco-diaphragmatic geometry, and provided information complementary to parenchymal density measures.