Patient-specific pulmonary digital twins from multi-volume CT: model construction and forward simulation of personalized protective mechanical ventilation in ARDS

Key information

Acute respiratory distress syndrome (ARDS) affects more than three million patients each year, with 30–50 % mortality. Mechanical ventilation is life-saving but aggravates lung injury through excessive strain (ventilator-induced lung injury, VILI). Because lung mechanics vary widely between patients, optimal ventilator settings remain largely empirical, and ventilation should be personalized.

This thesis is the methodological foundation of the ANR OPTIVENT project. Its core objective is to build functional pulmonary digital twins — patient-specific models reproducing the mechanical response of an individual lung — and to use them for forward simulation of mechanical ventilation. It builds on a framework demonstrated by our group, in which regional properties derived from CT images at two lung volumes are embedded in a model coupling an anatomically realistic bronchial tree, regional compliance and recruitment/derecruitment dynamics. Lung segmentation and deformable registration are already established and serve as input tools: the thesis concentrates on the model and on what it predicts.

Three challenges are addressed: model construction — a reduced-order (0D) regional lung model whose parameters are identified for each patient by solving an inverse problem constrained by the measured regional deformation and respiratory mechanics, then verified and validated; forward simulation — exposing each twin to a broad range of ventilation strategies to map, patient by patient, the relationship between settings and recruitment, overdistension and regional strain; and scale-up — running construction and simulation reproducibly on >200 ARDS patients of the multicentre CT4ARDS-2 study (NCT06113276) and on public multi-volume CT collections. The thesis delivers the validated twin cohort and the virtual clinical trial corpus on which the project’s explainable predictive model depends.

All details in the file attached to this offer !