Keywords
ARDS
Volumetry
Segmentation
Introduction
Acute Respiratory Distress Syndrome (ARDS) is characterised by diffuse, heterogeneous lung injury. Gattinoni established that only a small fraction of parenchyma remains normally aerated, the 'baby lung'; while the remainder collapses, floods, or hyperinflates. This reframes ARDS as a 'small' rather than 'stiff' lung, with critical implications for lung-protective ventilation targeting overdistension avoidance, recruitment, and injury prevention.
Objectives
To develop an open-source CT-based computational pipeline for automated 3D aeration volumetry in ARDS, generating clinically actionable compartment volumes to guide bedside ventilator titration.
Methods
Automated lung segmentation (TotalSegmentator/nnU-Net) was performed on isotropic 1 mm-resampled CT volumes. A rule-based algorithm classified voxels into four compartments by HU threshold: hyperinflated (?1000 to ?900), normally aerated (?900 to ?500), poorly aerated (?500 to ?100), and non-aerated (?100 to +100). K-means voxel clustering identified patient-level aeration phenotypes.
Results
In 23 ARDS cases, mean total lung volume was 3328 mL: normally aerated 2540 mL, poorly aerated 486 non-aerated 170 mL, and hyperinflated 113 mL. Voxel clustering (k = 3) revealed three phenotypes: normally aerated-dominant (n = 12), hyperinflation-enriched (n = 7), and recruitable-dominant (n = 4).
Discussion
This pipeline operationalises Gattinoni's baby lung framework computationally, delivering regional aeration and recruitability information equivalent to Electrical Impedance Tomography at zero incremental cost. Built entirely on open-source platforms (Python, SimpleITK, TotalSegmentator, scikit-learn, PyVista), it runs on standard laptop hardware using routine diagnostic CT scans already acquired for clinical care.
Conclusion
This pipeline operationalises Gattinoni's baby lung framework computationally, delivering regional aeration and recruitability information equivalent to Electrical Impedance Tomography. Built entirely on open-source platforms, it runs on standard laptop hardware using routine diagnostic CT scans already acquired for clinical care.
References
1. Gattinoni L, Pesenti A. The concept of 'baby lung.' Intensive Care Med. 2005;31(6):776–784.
