基于深度学习的自动静脉穿刺机器人研制

Development of automatic venepuncture robot based on deep learning

目的:基于传统静脉穿刺方法面临的困难,研制基于深度学习的自动静脉穿刺机器人,提高静脉穿刺的一次穿刺成功率。方法:该机器人主要由输入单元、处理器、控制器及执行机构组成。通过Faster-RCNN算法对超声图像血管进行粗定位,采用自动阈值分割算法提取血管轮廓进行测量获得血管位置和内径。通过微控制器STM32F103对执行机构中的电动机进行闭环控制,进而提高控制精度。采用超声穿刺训练体模进行穿刺试验验证该机器人的性能。结果:体模穿刺试验表明该机器人穿刺重复性和准确度较高,基本能满足应用要求。结论:该机器人既可避免穿刺时误入动脉、刺穿血管或误伤周围组织等弊端,又提高了一次穿刺的成功率,减少了穿刺次数及置管并发症的发生,可较好地应用于医院、重大灾害或军事救援中。

Objective To develop an automatic venepuncture robot based on deep learning to improve the one-time venepuncture success rate.Methods The robot was mainly composed of an input unit,a processor,a controller and an actuator.Preliminary positioning of the vessel in the ultrasound image was executed with deep learning Faster-RCNN algorithm,then the vessel's position and inradium were measured by using automatic threshold segmentation algorithm to determine its contour.Through the microcontroller STM32F103,the motor in the actuator was closed-loop controlled to improve the control accuracy.The performance of the robot was verified by venepuncture test with the training phantom of ultrasound venepuncture.Results The robot developed proved to gain high repeatability and accuracy when used for venepuncture,so that it could meet the requirement for practical application.Conclusion The robot avoids mis-puncturing the artery,surrounding tissue and etc,enhances one-time venepuncture rate and decreases venepuncture times and cathetering complications,which can be applied in the hospital,disaster relief and military rescue.[Chinese Medical Equipment Journal,2020,41(6):52-56]