Download PDFOpen PDF in browser

Hierarchical 3-D Registration of Computed Tomography to Ultrasound Using Reinforcement Learning

6 pagesPublished: September 25, 2020

Abstract

In ultrasound (US)-based computer-assisted orthopedic surgery (CAOS), accurate and robust intra-operative registration in real-time is vital in securing the reliable outcomes for surgical image guidance. For this purpose, we focus on developing a hierarchical registration method, using reinforcement learning (RL), for 3-D registration of pre-operative computed tomography (CT) data to intra-operative US. In the RL-based registration procedure, we proposed a supervised Q-learning framework for learning the sequence of motion action to achieve the optimal alignment. Within the approach, the agent was modeled using PointNet++ with the mis-aligned point set from US and CT as the input, and the next optimal action as the output. Evaluation studies achieved average target registration error (TRE) of 3.82 mm with success rate of 92.7% and an average time of 8.36 seconds. We achieve 57.1% improvement in success rate over state of the art.

Keyphrases: Computed Tomography, deep learning, registration, Ultrasound

In: Ferdinando Rodriguez Y Baena and Fabio Tatti (editors). CAOS 2020. The 20th Annual Meeting of the International Society for Computer Assisted Orthopaedic Surgery, vol 4, pages 306--311

Links:
BibTeX entry
@inproceedings{CAOS2020:Hierarchical_3_D_Registration_of,
  author    = {Xuxin Zeng and Michael Vives and Ilker Hacihaliloglu},
  title     = {Hierarchical 3-D Registration of Computed Tomography to Ultrasound Using Reinforcement Learning},
  booktitle = {CAOS 2020. The 20th Annual Meeting of the International Society for Computer Assisted Orthopaedic Surgery},
  editor    = {Ferdinando Rodriguez Y Baena and Fabio Tatti},
  series    = {EPiC Series in Health Sciences},
  volume    = {4},
  pages     = {306--311},
  year      = {2020},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-5305},
  url       = {https://easychair.org/publications/paper/rncg},
  doi       = {10.29007/12lv}}
Download PDFOpen PDF in browser