Prediction of Mechanical Lung Parameters Using Gaussian Process Models

Abstract. Mechanical ventilation can cause severe lung damage by inadequate adjustment of the ventilator. We introduce a Machine Learning approach to predict the pressure-dependent, non-linear lung compliance, a crucial parameter to estimate lung protective ventilation settings. Features were extracted by fitting a generally accepted lumped parameter model to time series data obtained from ARDS (adult respiratory distress syndrome) patients. Numerical prediction was performed by use of Gaussian processes, a probabilistic, non-parametric modeling approach for non-linear functions. 1 Medical Background and Clinical Purpose Under the condition of mechanical ventilation a high volume distensibility – or compliance C – of the lung is assumed to reduce the mechanical stress to the lung tissue and hence irreversible damage to the respiratory system. A common technique to determine the maximal compliance Cmax inflates the lung with almost zero flow (so-called ’static’ conditions) ov...

• Type: Conference
• Year: 2009
• Where: AIME
• Authors: Steven Ganzert, Stefan Kramer, Knut Möller, Daniel Steinmann, Josef Guttmann
• Keywords: AIME 2009, Lung, Lung Protective Ventilation, Mechanical Ventilation