Predicting global and regional respiratory response in ARDS patients through individualized physics-based computational lung models: A prospective clinical pilot study

Technical Summary: Predicting Global and Regional Respiratory Response in ARDS via Individualized Physics-Based Computational Lung Models

Problem Statement
The management of Acute Respiratory Distress Syndrome (ARDS) relies on lung-protective mechanical ventilation to prevent ventilator-associated lung injury (VALI), such as atelectrauma and volutrauma. However, current strategies are limited by the significant heterogeneity of lung pathology among patients and a lack of individualized, predictive tools. While imaging techniques like Computed Tomography (CT) and Electrical Impedance Tomography (EIT) provide insights into lung physiology, they are either static (CT) or lack spatial resolution and predictive capability (EIT). There is a critical need for patient-specific models that can predict global and regional respiratory responses to ventilator adjustments (e.g., changes in Positive End-Expiratory Pressure, or PEEP) to optimize treatment and minimize harm.

Methodology
This prospective, monocentric clinical pilot study involved 10 mechanically ventilated ARDS patients at University Hospital Schleswig-Holstein. The study protocol included systematic ventilatory maneuvers with variations in tidal volume, PEEP, and driving pressure under deep sedation.

The core methodology involved generating a personalized, regionally resolved, physics-based computational lung model for each patient:

1. Model Generation: Tracheal trees and lung lobes were segmented from thoracic CT scans. A morphology-based tree-growing algorithm created an anatomically accurate conducting airway structure.
2. Calibration: The model's non-linear pressure-volume characteristics were calibrated using patient-specific respiratory mechanics data (pressure-controlled ventilation at initial PEEP) and CT-derived pathology. The model incorporated edematous fluid, pressure- and time-dependent recruitment/derecruitment dynamics, and a ventral-to-dorsal pressure gradient.
3. Virtual EIT: The mechanical model was coupled with an EIT model based on a 3D thorax segmentation and finite element method simulations to generate virtual EIT datasets.
4. Validation: The model was tested against clinical data across four PEEP levels: initial (PEEPinit), maximum (PEEPmax), minimum (PEEPmin), and an EIT-derived crossover point (PEEPEIT).
   • Global Metrics: Simulated tidal volume ($V_T$), driving pressure ($\Delta P_{aw}$), end-expiratory transpulmonary pressure ($P_{tp,ee}$), and end-expiratory lung volume (EELV) were compared to clinical measurements using Pearson correlation ($r$), Root Mean Square Error (RMSE), and Bland-Altman analysis.
   • Regional Metrics: Simulated regional ventilation distributions were compared to clinical EIT images via pixel-wise correlation and established indices (regional ventilation fractions $F_{dorsal}$, $F_{left}$ and centers of ventilation $CoV_{rl}$, $CoV_{vd}$).

Key Results

• Global Mechanical Parameters: The model demonstrated strong agreement with clinical data across all PEEP levels.
  • Tidal Volume ($V_T$): $r = 0.925$ ($p < 0.001$), RMSE = 0.053 L.
  • Driving Pressure ($\Delta P_{aw}$): $r = 0.940$ ($p < 0.001$), RMSE = 1.328 mbar.
  • Transpulmonary Pressure ($P_{tp,ee}$): $r = 0.971$ ($p < 0.001$), RMSE = 1.898 mbar.
  • Performance was strongest near the calibration point (PEEPinit) and at PEEPEIT, with moderate deviations at extreme PEEP levels (min/max).
• Regional Ventilation (EIT):
  • At the initial PEEP level, high pixel-wise correlations ($r > 0.8$) were observed in 5 out of 7 patients with available EIT data.
  • Left-Right Distribution: Strong correlations were found for left-right metrics ($F_{left}$: $r=0.915$; $CoV_{rl}$: $r=0.843$).
  • Dorsal-Ventral Distribution: Correlations were weaker for dorsal-ventral metrics ($F_{dorsal}$: $r=0.214$; $CoV_{vd}$: $r=0.274$). The model tended to overestimate dorsal ventilation, a discrepancy attributed to the model's current lack of abdominal pressure influence and specific fluid dynamics.
• End-Expiratory Lung Volume (EELV): The model-derived EELV showed internal consistency with CT-derived EELV but exhibited discrepancies when compared to the methodologically independent nitrogen dilution technique ($EELV_{breathgas}$) in several patients, likely due to timing differences and the limitations of the reference method in heterogeneous ARDS.

Significance and Claims
The authors claim that this study demonstrates the capability of individualized, physics-based computational lung models to reproduce patient-specific respiratory mechanics and key features of regional ventilation patterns in ARDS.

• Predictive Validity: The strong agreement between simulated and measured global parameters supports the potential of this approach to estimate physiological responses to ventilator adjustments.
• Clinical Utility: The model offers a non-invasive method to estimate parameters like transpulmonary pressure, which are otherwise difficult to obtain without invasive monitoring. It serves as a complementary tool to support, rather than replace, clinical decision-making.
• Limitations and Future Work: The authors modestly acknowledge that the current precision is best suited for supportive guidance. They highlight specific limitations, including the small sample size, the need for improved modeling of abdominal pressure effects on dorsal ventilation, and the necessity for automated calibration pipelines to enable real-time ICU application. The study lays the groundwork for future research integrating these models into clinical workflows to test whether model-guided PEEP titration translates into measurable clinical benefits.