Comparative assessment of extravascular lung water index in ARDS patients on veno-venous ECMO: transpulmonary thermodilution versus AI-driven chest CT segmentation

In severe acute respiratory distress syndrome (ARDS) patients on veno-venous extracorporeal membrane oxygenation (VV ECMO), restrictive fluid management reduces pulmonary edema, quantified by extravascular lung water index (EVLWI), and improves outcomes. Bedside EVLWI is commonly assessed using single-indicator transpulmonary thermodilution (TPTD), which is increasingly employed to guide fluid therapy. However, TPTD-derived EVLWI values can be significantly affected by extracorporeal blood flow (ECBF), leading to inaccurate estimates during ECMO. In contrast, artificial intelligence (AI)-based automated lung segmentation of chest computed tomography (CT) enables reliable ECBF-independent visualization and quantification of pulmonary edema (EVLWICT). The primary objective was to compare EVLWI measured by TPTD (EVLWITPTD) with EVLWICT in ARDS patients on VV ECMO. The secondary objective was to derive and validate an ECBF-dependent correction factor to improve the accuracy of EVLWITPTD. In this retrospective, single-center observational cohort study of 64 patients with severe ARDS on VV ECMO, routine chest CT images and concurrent TPTD measurements, performed immediately before or after the CT scan, were analyzed. EVLWI was quantified using AI-based automated CT segmentation (EVLWICT) and TPTD (EVLWITPTD), the latter performed immediately before or after the CT scan. A multiple regression model using EVLWICT as the reference was applied to derive an ECBF-dependent correction factor for EVLWITPTD (EVLWITPTDcorr). EVLWITPTD significantly overestimated pulmonary edema compared to EVLWICT (24.3 15.8–30.0 vs. 10.2 5.8–14.8 ml/kg; p < 0.001), with a mean bias of − 12.3 ml/kg and limits of agreement from − 30.9 to 6.2 ml/kg. A correction factor of 3.439 ml/kg per liter of ECBF was identified. After adjustment, EVLWITPTDcorr did not differ significantly from EVLWICT (8.7 4.2–15.8 vs. 10.2 5.8–14.8 ml/kg; p = 0.393), with a reduced bias of 0.96 ml/kg and narrower limits of agreement (–16.4 to 18.4 ml/kg). In severe ARDS patients on VV ECMO, EVLWI obtained via TPTD was significantly overestimated due to ECBF interference. TPTD systematically overestimates EVLWI due to ECBF interference. Incorporating an ECBF-adjusted correction factor markedly improves agreement with AI-based CT quantification, enhancing the reliability and clinical applicability of TPTD-derived EVLWI for fluid management in this high-risk population. German Clinical Trials Register (DRKS00026246). Registered 14/09/2021. https://drks.de/search/en/trial/DRKS00026246.