A Hybrid Approach for Predicting Acute Hypotensive Episodes

This paper describes a hybrid approach for long term prediction of mean arterial blood pressure signal in order to detect an acute hypotensive episode (AHE) during one hour forecasting window. An acute hypotensive episode is defined as any period of 30 minute or more during which at least 90% of the non-overlapping one minute averages of the arterial blood pressure waveform is under 60 mmHg. The proposed method is based on wavelet transform and time-delay embedding neural networks.The wavelet transform is implemented to decompose mean arterial blood pressure (MAP) time series into set of wavelet components. A recurrent neural network architecture with embedded memory is then applied to forecast the wavelet approximation coefficients which represent the trend of the time series. Wavelet detail components were predicted using local radial basis models with time varying parameters. To obtain the predicted MAP time series, the neural network outputs were recombined using the same wavelet technique.The effectiveness of this strategy is validated using 40 records of arterial blood pressure signals from Multi-parameter Intelligent Monitoring in Intensive Care II (MIMIC II) database . Simulation results revealed the reasonable forecasting accuracy in prediction of AHE in one hour forecasting window

Keywords:

Acute Hypotensive Episode(AHE) , Mean Arterial Blood Pressure(MAP) , Neural Network , Prediction , Wavelet Transform