基于双向门控循环神经网络的在线超前呼吸运动预测模型

Online Advance Respiratory Motion Prediction Model Based on Bidirectional Gated Recurrent Unit

超声引导经皮穿刺手术机器人系统中,患者的呼吸运动会引起胸腹部组织变形和位置改变,同时控制系统对命令的响应存在延迟,导致机器人引导穿刺过程中超声探头与皮肤的稳定接触难以被保证,进而降低超声图像质量和穿刺精度。为了对呼吸运动进行提前预测以实现稳定接触和精准穿刺,本文提出了一种基于双向门控循环神经网络(Bi-GRU)的在线超前呼吸运动预测模型。本文基于17例受试者的体表呼吸运动数据,分析了影响在线超前呼吸运动预测模型准确性的主要因素。研究结果表明,本文提出的模型具有更高的实时性和准确性。

In ultrasound-guided percutaneous puncture surgery robot system,the patient's respiratory motion would cause soft tissue deformation and position change among the thorax and abdomen,at the same time the delay of control system response to commands exists.Thus,it is difficult to guarantee the stability of the contact between the ultrasonic probe and the skin of patient in the process of puncture,which reduces the quality and accuracy in ultrasound images collection.This paper proposed an Online advance respiratory motion prediction model based on bidirectional Gated Recurrent Unit(Bi-GRU).Based on the surface respiratory motion data of 17 subjects,this paper studied the main factors affecting the accuracy of the online advance respiratory motion prediction model.The results show that the proposed model has higher real-time performance and accuracy.