Health and Medical Research Institute, National Institute of Advanced Industrial Science and Technology, Tsukuba, Ibaraki, Japan.
Graduate School of Informatics and Engineering, The University of Electro-Communications, Chofu, Tokyo, Japan.
Int J Comput Assist Radiol Surg. 2024 Mar;19(3):449-457. doi: 10.1007/s11548-023-03019-5. Epub 2023 Oct 3.
Scanning path planning is an essential technology for fully automated ultrasound (US) robotics. During biliary scanning, the subcostal boundary is critical body surface landmarks for scanning path planning but are often invisible, depending on the individual. This study developed a method of estimating the rib region for scanning path planning toward fully automated robotic US systems.
We proposed a method for determining the rib region using RGB-D images and respiratory variation. We hypothesized that detecting the rib region would be possible based on changes in body surface position due to breathing. We generated a depth difference image by finding the difference between the depth image taken at the resting inspiratory position and the depth image taken at the maximum inspiratory position, which clearly shows the rib region. The boundary position of the subcostal was then determined by applying training using the YOLOv5 object detection model to this depth difference image.
In the experiments with healthy subjects, the proposed method of rib detection using the depth difference image marked an intersection over union (IoU) of 0.951 and average confidence of 0.77. The average error between the ground truth and predicted positions was 16.5 mm in 3D space. The results were superior to rib detection using only the RGB image.
The proposed depth difference imaging method, which measures respiratory variation, was able to accurately estimate the rib region without contact and physician intervention. It will be useful for planning the scan path during the biliary imaging.
扫描路径规划是全自动超声（US）机器人的关键技术。在胆道扫描中，肋缘边界是扫描路径规划的关键体表标志，但根据个体的不同，肋缘边界可能不可见。本研究开发了一种方法，用于估计肋骨区域，以便为全自动机器人 US 系统规划扫描路径。
我们提出了一种使用 RGB-D 图像和呼吸变化来确定肋骨区域的方法。我们假设，由于呼吸导致的体表位置变化，检测肋骨区域是可行的。我们通过找到在休息吸气位置拍摄的深度图像与在最大吸气位置拍摄的深度图像之间的深度差，生成了一个深度差图像，该图像清晰地显示了肋骨区域。然后，通过使用 YOLOv5 对象检测模型对这个深度差图像进行训练，确定肋缘下界的边界位置。
在对健康受试者的实验中，使用深度差图像的肋骨检测方法的交并比（IoU）为 0.951，平均置信度为 0.77。在 3D 空间中，地面真实值和预测位置之间的平均误差为 16.5 毫米。该方法的结果优于仅使用 RGB 图像的肋骨检测。
该研究提出的深度差成像方法，通过测量呼吸变化，无需接触和医生干预，就能准确估计肋骨区域。它将有助于规划胆道成像过程中的扫描路径。
