Attention-based convolutional long short-term memory neural network for detection of patient-ventilator asynchrony from mechanical ventilation. (September 2022)

• Record Type: Journal Article
• Authors: Chen, Dingfu; Lin, Kangwei; Deng, Ziheng; Li, Dayu; Deng, Qingxu
• Is Part Of: Biomedical signal processing and control. Volume 78(2022)
• Publication Date: 2022-09
• Subjects: Mechanical ventilation -- Patient-ventilator asynchrony -- Convolutional neural network -- LSTM -- Attention mechanism; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103923
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Combining CNN with LSTM into a fusion network for PVA detection. Local features and long-term dependencies are extracted and fused for classification. Attention mechanism focuses on salient features and make the result interpretable. Ablation and 5-fold cross-validation prove the algorithm gets better performance. The F1 score and MCC of the proposed algorithm are 0.992 and 0.927, respectively. Abstract: During mechanical ventilation, the mismatch between the patient's needs and the ventilator settings will lead to the occurrence of patient-ventilator asynchrony (PVA), which adversely affects the patient's recovery. Therefore, it is essential to develop an algorithm that can detect PVA accurately and automatically. However, common methods including deep learning methods have low recognition efficiency and lack of interpretability. In this study, we proposed an attention-based convolutional long short-term memory network for recognizing two common types of PVA. Combining the CNN network with the LSTM network could capture the local features of the input while ensuring the long-term dependencies of the sequence data. Furthermore, an attention mechanism was introduced to improve the accuracy and efficiency of recognition as well as the interpretability of the prediction. In the test dataset, the mean accuracy, F1 score, and Matthews correlation coefficient (MCC) for identifying IE and DT were 0.989, 0.992, and 0.927, respectively. Moreover, the attention mechanism enabled a more intuitive view of the information that the model focuses on for different labels. The experimental results suggest that the algorithm proposed in this paper can detect PVA more accurately than existing algorithms, help doctors to detect and correct PVA in time, which is conducive to the recovery of patients.