摘要
目的 探讨深度学习相关技术在医疗器械不良事件风险类型判别中的应用价值.方法 选取自2023年4月至9月国家医疗器械不良事件监测系统数据库中的12 350条数据为研究对象,每条数据具有25个维度的信息,其中包括企业、注册证号、产品批号、不良事件描述等.采用fastText技术将文本特征转化为向量表示,采用k均值聚类方法将具有相似主题或语义内容的文本数据分组到同一类别中,搭建BP神经网络将不良事件的风险类型分为轻微危害、严重危害、死亡.结果 该研究选取的深度学习方法能够很好地处理文本数据且搭建的BP神经网络模型的准确度(accuracy)为92.86％,精密度(precision)为 93.65％,召回率(recall)为 93.08％,F1-score 为 92.31％,曲线下面积(AUC)为 0.98,具有良好的准确性和泛化能力.结论 基于深度学习的医疗器械不良事件风险类型的研究可有效地为医疗器械不良事件监测工作提供帮助.
Abstract
[Objective]To explore the value of deep learning related techniques in the risk type discrimination of medical device adverse events.[Methods]The study population was selected from April to September 2023,12,350 data in the Medical Device Adverse Event Surveillance System database,each with 25 dimensions of information,including company,registration certificate number,product batch number,and description of adverse events.This study uses fastText technology to convert text features into vector representations,uses k-means clustering method to classify text data with similar themes or semantic content into the same category,and uses BP neural network algorithm to classify the severity of adverse events into mild injury,severe injury,and death.[Results]This method can handle text data well,and the accuracy of the BP neural network model is 92.86％,accuracy is 93.65％,recall is 93.08％,F1 score is 92.31％,AUC is 0.98,with good accuracy and generalization ability.[Conclusion]The research on the risk types of adverse events in medical devices based on deep learning can effectively provide assistance for the monitoring of adverse events in medical devices.
维普
万方数据