Autonomous image-based ultrasound probe positioning via deep learning

EasyChair Preprint 119

Abstract

Although ultrasound (US) is a widely used, non-invasive, and radiation-free imaging modality, manual adjustment of the US probe can be cumbersome and time consuming. An autonomous US scanning device could not only reduce dependence on sonographer’s skills and experience but also improve workflow efficiency especially during interventional procedures. Robot-assisted ultrasound imaging has the potential to improve patient care in rural and underserved areas. There are many previous efforts in this direction but none is fully automatic or accurate enough. In this work, as an initial small step towards independent US imaging workflow solution, we developed and evaluated a robot-assisted fully autonomous ultrasound (RAFAUS) probe positioning system. Desired motion of the probe toward the target view is directly derived from anatomical features implicitly extracted via deep neural network; thus, making this technique (a) invariant to anatomical differences (patient size or organ displacement), (b) decoupled from the robotic system, (c) registration-free, and (d) independent from any external tracking technologies.

Keyphrases: Autonomous, Ultrasound, deep learning, image-based, robot-assisted

@booklet{EasyChair:119,
  author    = {Grzegorz Toporek and Haibo Wang and Marcin Balicki and Hua Xie},
  title     = {Autonomous image-based ultrasound probe positioning via deep learning},
  doi       = {10.29007/dj33},
  howpublished = {EasyChair Preprint 119},
  year      = {EasyChair, 2018}}