首页|一种通道自适应与局部增强的Transformer术中血压预测方法

一种通道自适应与局部增强的Transformer术中血压预测方法

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
准确预测术中患者的血压状态来预防术中低血压，对提高手术安全性和降低术后并发症有积极作用，以往的低血压预测方法主要视为二分类任务，忽略了患者血压变化的过程，从而限制了干预策略的制定。因此提前预测血压的变化趋势，具有更重要的临床研究和应用价值。针对以上问题，对通过监测的术中生理序列实时预测未来5min、10min、15min血压的连续值展开研究，并提出了一种通道自适应与局部增强Transformer模型，该模型采用卷积注意力机制捕捉血压序列的局部相似性，同时提出一种通道自适应模块嵌入模型来建模生理序列潜在交互关系。结果表明，该模型相比于基准模型在5min、10min、15min预测精度分别提升4。88%、8。2%和8。42%，预测的平均动脉压的MAE分别为2。997、3。393、3。743，且显著优于其余对比模型，为术中血压预测提供新的解决方案。
A Transformer Intraoperative Blood Pressure Prediction Method Based on Channel Adaptation and Local Enhancement
Accurately predicting the intraoperative blood pressure status of patients to prevent intraoperative hypotension has a positive effect on improving surgical safety and reducing postoperative complications.Previous hypotension prediction methods were mainly regarded as binary classification tasks,ignoring the process of patient blood pressure changes,thus limiting the formu-lation of intervention strategies.Therefore,predicting the change trend of blood pressure in advance has more important clinical re-search and application value.In this study,it focuses on the real-time prediction of future blood pressure values at 5min,10min and 15min using monitored intraoperative physiological sequences.A Channel-Adaptive and Locally-Enhanced Transformer model is proposed,which captures the local similarity of blood pressure sequences using convolutional attention mechanisms and incorpo-rates a Channel-Adaptive module to model the underlying interactions in physiological sequences.Experimental results show that the proposed model achieves significant improvements in prediction accuracy at 5min,10min and 15min,with respective increases of 4.88%,8.2%and 8.42%compared to the baseline model.The Mean Absolute Error(MAE)for predicted mean arterial pressure is 2.997,3.393 and 3.743,respectively,outperforming other comparative models significantly.The findings provide a new solution for intraoperative blood pressure prediction.

intraoperative blood pressure predictionTransformerphysiological sequenceattention mechanismchannel adaptation

王尘、蔡晶晶、郝学超、张伟义、舒红平、王亚强、陈果

展开 >

成都信息工程大学软件工程学院 成都 610225

成都信息工程大学数据科学与工程研究所 成都 610225

成都信息工程大学软件自动生成与智能服务实验室 成都 610225

四川大学华西医院麻醉手术中心 成都 610044

展开 >

术中血压预测 Transformer 生理序列 注意力机制 通道自适应

四川大学华西医院"学科卓越发展1·3·5工程"交叉学科创新项目四川大学华西医院1·3·5项目国家重点研发计划

2023H022ZYJC210082018YFC2001800

2024

10.3969/j.issn.1672-9722.2024.01.007

计算机与数字工程
中国船舶重工集团公司第七0九研究所

计算机与数字工程

CSTPCD
影响因子：0.355
ISSN：1672-9722
年,卷(期)：2024.52(1)
  • 19