Implementation of a Non-Linear Autoregressive Model with Modified Gauss-Newton Parameter Identification to Determine Pulmonary Mechanics of Respiratory Patients that are Intermittently Resisting Ventilator Flow Patterns

Abstract

Modelling the respiratory system of intensive care patients can enable individualized mechanical ventilation therapy and reduce ventilator induced lung injuries. However, spontaneous breathing (SB) efforts result in asynchronous pressure waveforms that mask underlying respiratory mechanics. In this study, a nonlinear auto-regressive (NARX) model was identified using a modified Gauss-Newton (GN) approach, and demonstrated on data from one SB patient. The NARX model uses three pressure dependent basis functions to capture respiratory system elastance, and contains a single resistance coefficient and positive end expiratory pressure (PEEP) coefficient. The modified GN method exponentially reduces the contribution of large residuals on the step in the coefficients at each GN iteration. This approach allows the model to effectively ignore the anomaly in the pressure waveform due to SB efforts, while successfully describing the shape of normal breathing cycles. This method has the potential to be used in the ICU to more robustly capture patient-specific behaviour, and thus enable clinicians to select optimal ventilator settings and improve patient care.

Estimating the incidence of spontaneous breathing effort of mechanically ventilated patients using a non-linear auto regressive (NARX) model

“…The original data and airway pressure constructed by the NARX model are to be compared and the mean residual error is to be calculated with the intend to quantify the incidence and the magnitude of asynchronous in larger errors, where smaller errors due to noise and model identification. Ergo, when the value of the mean residual error inclines, the higher the magnitude of spontaneous breathing, and more frequently it is, the greater the incidence of SB effort [ 20 ]. This study considered each breathing cycles as independent and can be analysed separately.…”

“…This is because the NARX model which uses the 70:30 ratio has more features to predict major concrete outliers in the testing data. However, by adding more data to the training may cause a swarm of overfitting since additional features may either be irrelevant or redundant, especially when it involves noise in the pressure waveform [ 20 , 22 ]. Hence, this will disrupt the reconstruction of the reduced pressure.…”

“…Equally, esophageal pressure measurement is too invasive and burdensome for regular clinical use, along with the correction method that can be too computationally intensive [ [16] , [17] , [18] , [19] ]. This study aims to utilize a non-linear autoregressive (NARX) model to estimate and quantify SB efforts by reconstructing the unaltered passive mechanics airway pressure using the NARX model as suggested in [ 20 ]. The autoregressive model is able to define the input-output relationships in the system, and is capable of constructing predictions based on previous inputs [ 21 ].…”

Optimising mechanical ventilation through model-based methods and automation

“…Spontaneously breathing patients apply their own inspiratory efforts on top of a ventilator supported breathing cycle ( Langdon, Docherty, Chiew, Damanhuri, & Chase, 2015 ). A time-varying elastance model was developed to describe the mechanics of spontaneously breathing patients on partial assist mechanical ventilation ( Chiew, Pretty, Docherty, et al., 2015 ).…”

“…Work has been done to predict lung mechanics at a high PEEP using information provided at a lower PEEP ( Langdon, Docherty, Chiew, & Chase, 2016 ). A non-linear autoregressive (NARX) resistance and basis function elastance model was developed from a viscoelastic form of the single compartment model ( Langdon et al., 2016a , Langdon et al., 2015 ). Basis functions were developed from overlapping B-spline functions of different orders ( Langdon et al., 2016a , Langdon et al., 2015 , Langdon et al., 2016b ).…”