基于半监督卷积神经网络进行人机不同步的识别

目的 基于半监督卷积神经网络(semi-supervised convolutional neural network, semi-CNN)构建人机不同步现象(patient-ventilator asynchrony, PVA)识别模型,评价其在压力支持通气(pressure support ventilation,PSV)模式下的诊断效能。方法 分析85例接受PSV通气脑损伤患者的机械通气数据,结合食道压监测数据进行人工标识。使用Transformer时间序列预测模型对已标识的正常或发生PVA的呼吸进行转化,转化后的数据输入semi-CNN模型判断是否发生PVA。在测试集中验证模型的准确性、灵敏度、特异度以及与专家标识结果的一致性。结果 初始训练集包含正常呼吸513次,异常呼吸69次,经过500次迭代后模型收敛。测试集包含正常呼吸48次,异常呼吸24次。在测试集中,Transformer联合semi-CNN模型识别PVA的准确率为0.92(0.83~0.97),灵敏度为0.79(0.58~0.93),特异度为0.98(0.89~1.00),Kappa值为0.80 (0.65~0.95),测试结果与专家人工标识结果具有高度一致性。结论 本研究提供了一种基于semi-CNN算法的PVA识别模型,其识别PVA的准确率和特异度高,识别结果与专家人工标识结果的一致性好,可用于临床实时PVA监测。

Detection of patient-ventilator asynchrony based on semi-supervised convolutional neural network

Objective To construct a patient-ventilator asynchrony (PVA) detection model based on semi-supervised convolutional neural networks (semi-CNN)and evaluate its diagnostic efficiency in pressure support ventilation (PSV) mode.Methods The mechanical ventilation data of 85 brain injury patients with PSV mode were analyzed retrospectively. Tracing of flow, and airway pressure combined with esophageal pressure data were used to identify manually the presence of PVA visually. Patient's future breathing observations were predicted by the Transformer-based time series prediction module. The output of the Transformer module was used as the input parameter of the semi-CNN to detect the occurrence of PVA. The accuracy, sensitivity, and specificity of the semi-CNN model and agreement between the semi-CNN model and human experts were validated in the test set. Results The initial training set included 513 normal breaths and 69 PVA breaths. After 500 epochs, the model converges.The test set included 48 normal breaths and 24 PVA breaths. In the test set, the accuracy,the sensitivity, and the specificity of the Transformer + semi-CNN model in identifying PVA are 0.92(0.83－0.97), 0.79(0.58 - 0.93), and 0.98(0.89 －1.00), respectively.There is substantial agreement between the semi-CNN and the human experts in identifying PVA (Kappa=0.80, 95%CI:0.65-0.95, P<0.01). Conclusion This study provided a PVA detecting method based on a semi-CNN algorithm with high accuracy and specificity. The proposed method showed a substantial agreement with human experts in identifying PVA and could be applied in real-time monitoring of PVA clinically.