Hierarchical Parameter Identification in Models of Respiratory Mechanics

Abstract

Potential harmful effects of ventilation therapy could be reduced by model-based predictions of the effects of ventilator settings to the patient. To obtain optimal predictions, the model has to be individualized based on patients' data. Given a nonlinear model, the result of parameter identification using iterative numerical methods depends on initial estimates. In this work, a feasible hierarchical identification process is proposed and compared to the commonly implemented direct approach with randomized initial values. The hierarchical approach is exemplarily illustrated by identifying the viscoelastic model (VEM) of respiratory mechanics, whose a priori identifiability was proven. To demonstrate its advantages over the direct approach, two different data sources were employed. First, correctness of the approach was shown with simulation data providing controllable conditions. Second, the clinical potential was evaluated under realistic conditions using clinical data from 13 acute respiratory distress syndrome (ARDS) patients. Simulation data revealed that the success rate of the direct approach exponentially decreases with increasing deviation of the initial estimates while the hierarchical approach always obtained the correct solution. The average computing time using clinical data for the direct approach equals 4.77 s (SD = 1.32) and 2.41 s (SD = 0.01) for the hierarchical approach. These investigations demonstrate that a hierarchical approach may be beneficial with respect to robustness and efficiency using simulated and clinical data.

Traversing the Fuzzy Valley: Problems Caused by Reliance on Default Simulation and Parameter Identification Programs for Discontinuous Models

“…While numerical integration coupled with gradient based identification methods are frequently used successfully (Sundaresan et al 2009;Schranz et al 2011), it is important to know the limitations of such methods. In particular, this investigation has shown that numerical integration methods which handle discontinuities poorly, such as the proprietary MATLABTM methods, are likely to cause parameter identification failure when poorly applied.…”

A polynomial model of patient-specific breathing effort during controlled mechanical ventilation

“…The initial values for t s and t f are determined by fitting the conventional single compartment lung model using a hierarchical approach [10]. The fitted model will have regions where the modelled pressure is lower than the pressure data.…”

Identifiability Analysis of a Pressure-Depending Alveolar Recruitment Model

“…Accurate initial parameter values can significantly reduce the incidence of finding local minima. Thus, a hierarchical parameter identification process is applied (Schranz et al, 2011).…”

“…The hierarchical method provides patient-specific initial values by identifying simpler models with fewer variable parameters first (Schranz et al, 2011). These first results provide appropriate initial values for the identification of the next, more complex model.…”