Key Laboratory of Brain-Machine Intelligence Technology, Ministry of Education, Nanjing University of Aeronautics and Astronautics, Nanjing, China; College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China.
Key Laboratory of Brain-Machine Intelligence Technology, Ministry of Education, Nanjing University of Aeronautics and Astronautics, Nanjing, China; College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China.
Comput Methods Programs Biomed. 2023 Oct;240:107605. doi: 10.1016/j.cmpb.2023.107605. Epub 2023 May 18.
A capsule robot can be controlled inside gastrointestinal (GI) tract by an external permanent magnet outside of human body for finishing non-invasive diagnosis and treatment. Locomotion control of capsule robot relies on the precise angle feedback that can be achieved by ultrasound imaging. However, ultrasound-based angle estimation of capsule robot is interfered by gastric wall tissue and the mixture of air, water, and digestive matter existing in the stomach.
To tackle these issues, we introduce a heatmap guided two-stage network to detect the position and estimate the angle of the capsule robot in ultrasound images. Specifically, this network proposes the probability distribution module and skeleton extraction-based angle calculation to obtain accurate capsule robot position and angle estimation.
Extensive experiments were finished on the ultrasound image dataset of capsule robot within porcine stomach. Empirical results showed that our method obtained small position center error of 0.48 mm and high angle estimation accuracy of 96.32%.
Our method can provide precise angle feedback for locomotion control of capsule robot.
胶囊机器人可以通过人体外部的永久磁铁在胃肠道内进行控制，从而完成非侵入性诊断和治疗。胶囊机器人的运动控制依赖于可以通过超声成像实现的精确角度反馈。然而，基于超声的胶囊机器人角度估计受到胃壁组织以及存在于胃中的空气、水和消化物混合物的干扰。
为了解决这些问题，我们引入了一个热图引导的两阶段网络来检测超声图像中胶囊机器人的位置并估计其角度。具体来说，该网络提出了概率分布模块和基于骨架提取的角度计算，以获得胶囊机器人位置和角度的准确估计。
在猪胃内的胶囊机器人超声图像数据集上完成了广泛的实验。实验结果表明，我们的方法获得了较小的位置中心误差 0.48mm 和较高的角度估计精度 96.32%。
我们的方法可以为胶囊机器人的运动控制提供精确的角度反馈。
