Scholarlink

<p>To improve the quality of gastrointestinal (GI) tract screening, a wireless capsule endoscope (WCE) has been utilized. However, wide space localization of a WCE is challenging due to the complex geometry of GI and the limited sensors to observe the state of the WCE. In this paper, we propose a wide workspace localization of a WCE by integrating a local localization via magnetic fields and a global localization via a robotic arm. To overcome the optimization difficulty of solving a magnetic dipole model equation, we implemented a data-driven method with equation-based constraints to estimate the position and orientation using a physics-guided neural network (PGNN). To validate the performance of the proposed algorithm, we tested the system to track the position of the magnetic capsule in a series of motions while estimating the orientation of the capsule simultaneously. The results show that the proposed algorithm can track the capsule in real-time with estimation time of 20&#xa0;ms and a position error of 1.29 ± 0.68&#xa0;mm with orientation error of 2.51 ± 1.43° on circle motion, and position error of 1.05 ± 0.44&#xa0;mm with orientation error of 2.65 ± 0.88° on the 3D small intestine model. In addition, the algorithm could perform accurate estimation for patient movement. The proposed approach in this study can presume the possible applicability to a real environment in the future.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-Driven Magnetic Capsule Posture Estimation in a Wide Space using PGNN: Feasibility Study

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Stevanus Darwin,
  • Yeongoh Ko,
  • Ivan Adi Kuncara,
  • Chang-Sei Kim

摘要

To improve the quality of gastrointestinal (GI) tract screening, a wireless capsule endoscope (WCE) has been utilized. However, wide space localization of a WCE is challenging due to the complex geometry of GI and the limited sensors to observe the state of the WCE. In this paper, we propose a wide workspace localization of a WCE by integrating a local localization via magnetic fields and a global localization via a robotic arm. To overcome the optimization difficulty of solving a magnetic dipole model equation, we implemented a data-driven method with equation-based constraints to estimate the position and orientation using a physics-guided neural network (PGNN). To validate the performance of the proposed algorithm, we tested the system to track the position of the magnetic capsule in a series of motions while estimating the orientation of the capsule simultaneously. The results show that the proposed algorithm can track the capsule in real-time with estimation time of 20 ms and a position error of 1.29 ± 0.68 mm with orientation error of 2.51 ± 1.43° on circle motion, and position error of 1.05 ± 0.44 mm with orientation error of 2.65 ± 0.88° on the 3D small intestine model. In addition, the algorithm could perform accurate estimation for patient movement. The proposed approach in this study can presume the possible applicability to a real environment in the future.