Surgical Tool Localization in 3D Ultrasound Images

Marián Uherčík1,2, Jan Kybic1, Hervé Liebgott2, Christian Cachard2,
Martin Barva1,2, Jean-Martial Mari2, Václav Hlaváč1

1 Center for Machine Perception, Department of Cybernetics, Faculty of Electrical Engineering, Czech Technical University, Prague, Czech Republic
2 CREATIS-LRMN, CNRS UMR 5515, INSERM U630, Université Lyon 1, INSA Lyon, Lyon, France
 

Motivation

During surgical interventions, small instruments are often introduced into the body with ultrasound guidance:

  • Needle aspiration biopsy (Figure 1).
  • Neuronal cortical recording (Figure 2).
  • Prostate brachytherapy.
  • Laparoscopic surgery.

Aim si to develop an automatic localization method in 3D ultrasound images which shows the location of the tool in surrounding tissue.

Figure 1 Figure 2
 

Proposed Methods

Assumptions:

  • High-intensity appearance of the tool.
  • Curvilinear shape of the tool (diameter around 1 mm).

Localization method consists of two steps:

  1. Tool axis localization:
     
    • Method I -- Parallel Integral Projection (PIP) transform [1].
      Finds such orientation that tool voxels are on a single projection line. It works only for straight tools and it is too slow from real-time. We proposed a speed-up by multi-resolution [3].
       
    • Method II -- Model fitting using RANSAC procedure [2].
      Describes the tool axis as a polynomial curve and by using an appearance tool model. It is robust to irregular tool appearance, tool deformation and other high intensity structures. It provides a fast and reliable estimation, refined by local optimization.
       
  2. Tool tip localization.
    Assuming the tool axis to be known, the tip is determined as significant drop of intensities on the axis.
Figure 3 - 3D ultrasound image of PVA cryogel phantom submerged into water. Inside the phantom there was a tungsten electrode. Moving the mouse cursor over the image shows the result of localization marked by red line.

 

Results

Testing datasets can be obtained on this webpage. There are simulated data and real ultrasound data.

Comparison of localization algorithms on the PVA phantom data (Figure 3):

Method Time [sec] Axis ac. [mm] Tip ac. [mm]
MR PIP 62.5 0.44 ± 0.21 0.51 ± 0.17
RANSAC 0.66 0.47 ± 0.31 0.46 ± 0.26

Comparison of localization algorithms on real data of the breast biopsy (Figure 4):

Method Time [sec] Axis ac. [mm] Tip ac. [mm]
MR PIP 61.0 0.108 0.569
RANSAC 1.66 0.25 ± 0.1 0.97 ± 0.25

MR PIP = Multi-resolution PIP [3]

Figure 4 - 3D view of data from the breast biopsy with a localized needle marked by green color line.
 

Demonstration Applications

Real-time tool localization in 3D US implemented on Ultrasonix scanner [4]

GUI application for testing of various tool localization methods in 3D US

 

Selected Publications

  1. M. Barva, M. Uherčík, J.-M. Mari, J. Kybic, J.-R. Duhamel, H. Liebgott, V. Hlaváč, C. Cachard: Parallel Integral Projection Transform for Straight Electrode Localization in 3D Ultrasound Images, IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control (UFFC), pp. 1559-1569, July 2008

  2. M. Uherčík, J. Kybic, H. Liebgott, C. Cachard: Model Fitting using RANSAC for Surgical Tool Localization in 3D Ultrasound Images, IEEE Transactions on Biomedical Engineering (BME), pp. 1907-1916, Aug. 2010. 

  3. M. Uherčík, J. Kybic, H. Liebgott, C. Cachard: Multi-resolution Parallel Integral Projection for Fast Localization of a Straight Electrode in 3D Ultrasound Images, IEEE International Symposium on Biomedical Imaging (ISBI), pp. 33-36, May 2008.

  4. F. Gaufillet, H. Liebgott, M. Uherčík, F. Cervenansky, J. Kybic, and C. Cachard: 3D ultrasound real-time monitoring of surgical tools. In IEEE International Ultrasonics Symposium (IUS), October 2010.

Last update: 8th December 2010