基于自编码器的医疗器械设备异常状态预警方法  

A medical device abnormal state warning method based on autoencoder

作　　者：毛媛 秦琴 MAO Yuan;QIN Qin(Disinfection and Supply Center,363 Hospital,Chengdu 610041)

机构地区：[1]三六三医院消毒供应中心,成都610041

出　　处：《北京生物医学工程》2024年第4期391-397,共7页Beijing Biomedical Engineering

基　　金：2022年成都市医学科研课题项目(2022190);通用医疗2022年度科研基金项目(TYYLKYJJ-2022-050)资助。

摘　　要：目的 为了实时监测医疗器械设备异常状态,以便及时预警和应对设备异常情况,确保医疗器械的安全运行,提出基于自编码器的医疗器械设备异常状态预警方法。方法 以某医院2022年8月起因故障停用的10台心电图监护仪为研究对象,使用数据恢复设备连接到心电图监护仪的内部存储器,恢复和提取器械因故停用点的前1 min的全部特征数据,具体包括:信号质量和设备状态。利用多元状态估计技术构建医疗器械设备健康状态评估模型,通过对比各参数的贡献率,确定引发设备异常的关键参数。基于这些关键参数构建自编码器预警模型,将异常状态参数输入模型进行归一化处理,并计算参数误差平均值与基准值的偏差。当误差平均值偏离基准值超过60%时,触发预警机制。结果 经过实验,该方法可以在3 s内准确检测出198个异常样本,证实了该预警方法能够在短时间内准确检测出异常样本。同时,为进一步分析该医疗器械设备的具体故障原因,运用3种方法分别对设备各个故障原因进行识别与定位,以贡献率为衡量标准,并与实际贡献率进行对比分析,该方法贡献率与实际贡献率数值较为接近,可以全面预测引起医疗器械设备异常情况。结论 基于自编码器的医疗器械设备异常状态的预警方法具有高度的精确性和先导性,为实时监测和预警医疗器械设备异常状态提供了有效的技术手段,有助于提升医疗器械的安全性和可靠性。Objective In order to monitor the abnormal state of medical devices in real time,so as to timely warn and respond to the abnormal state of medical devices,and to ensure the safe operation of medical devices,an automatic encoder based medical device abnormal state early warning method was proposed.Methods Ten electrocardiogram monitors in a hospital were used as research objects due to failure in August 2022.We used the data recovery device connected to the internal memory of the ECG monitor to recover and extract all the characteristic data of the last minute of the device's untimely outage point,including signal quality and device status.Multivariate state estimation technology was used to build the health state assessment model of medical devices,and the key parameters causing equipment abnormalities were determined by comparing the contribution rates of the parameters.Based on these key parameters,an autoencoder early warning model was constructed,abnormal state parameters were input into the model for normalization processing,and the deviation between the average parameter error and the reference value was calculated.When the average error deviates from the benchmark value by more than 60%,the early warning mechanism was triggered.Results The experiment showed that the method could accurately detect 198 abnormal samples within 3s,which confirmed that the method could accurately detect abnormal samples in a short time.At the same time,in order to further analyze the specific failure causes of the medical equipment,three methods were used to identify and locate the causes of each failure of the equipment,taking the contribution rate as the measurement standard,and comparative analysis with the actual contribution rate,the contribution rate of this method was relatively close to the actual contribution rate,and could comprehensively predict the abnormal situation caused by the medical equipment.Conclusions The autoencoder-based medical device abnormal state early warning method has a high degree of accuracy and

关 键 词：医疗器械设备 多元状态估计 贡献率分析 自编码器 异常状态预警 

分 类 号：R318[医药卫生—生物医学工程]