Combining Neural Network and Wavelet Transformn for Trigger Asynchrony Detection

Trigger asynchrony is a phenomenon when the ventilator is out of synchronization with breathes of respiratory patients. The occurrence of trigger asynchrony would cause discomfort and harms to the patients. Thus understanding the trigger asynchrony situation for better setting the ventilator parameters to lower the possibility of occurrence of trigger asynchrony is critical to respiratory patient care. This paper proposes the combination of neural networks and wavelet feature extraction for trigger asynchrony detection. The performances using various training situations are also compared. A breath cycle is composed of inspiratory phase and expiratory phase. In this paper we also explore the performance differences between the situation when the neural network detection is applied with the same trained neural network for inspiratory and expiratory phases and the situation when the detection and is applied with different neural network for inspiratory and expiratory phases. It was found that although separating detection with different neural networks for the inspiratory and expiratory phases requires slightly more time, it achieves higher performance than that the detection is applied with the same neural network for both phases. The results are also compared with the results by physicians´ observations for accuracy evaluation.