DBP2:Queens:Introduction

From NAMIC Wiki
Jump to: navigation, search
Home < DBP2:Queens:Introduction

Back to JHU DBP 2

Segmentation and Registration Tools for Robotic Prostate Interventions

Team and Institutes

  • PI: Gabor Fichtinger, Queen’s University (gabor at cs.queensu.ca)
  • Co-I: Purang Abolmaesumi, Queen’s University (purang at cs.queensu.ca)
  • Software Engineer Lead: David Gobbi, Queen’s University (dgobbi at cs.queensu.ca
  • Software Engineer Support: Siddharth Vikal, Queen’s University (siddharthvikal at yahoo.com)
  • JHU Software Engineer Support: Csaba Csoma, Johns Hopkins University, csoma at jhu.edu
  • NA-MIC Engineering Contact: Katie Hayes, MSc, Brigham and Women's Hospital, hayes at bwh.harvard.edu
  • NA-MIC Algorithms Contact: Allen Tannenbaum, PhD, GeorgiaTech, tannenba at ece.gatech.edu
  • Host Institutes: Queen's University & Johns Hopkins University

Research Goals

The Queen’s & Hopkins teams are developing novel systems and procedures for prostate cancer interventions, such as biopsy and needle-based local therapies.

Prostate cancer is the most common subcutaneous cancer in American men. In 2007 will be an estimated 220,000 new cases of prostate cancer and 28,000 deaths caused by prostate cancer in the United States alone. [1]

The current standard of care for verifying the existence of prostate cancer is transrectal ultrasound (TRUS) guided biopsy. TRUS provides limited diagnostic accuracy and image resolution. In a study [2] the authors conclude that TRUS is not accurate for tumor localization and therefore the precise identification and sampling of individual cancerous tumor sites is limited. As a result, the sensitivity of TRUS biopsy is only between 60% and 85%. [3, 4]

Targeted biopsies of suspicious areas identified by MRI could potentially increase the sensitivity of prostate biopsies. To address this problem the investigators have several active research projects in prostate biopsy and therapies under direct MRI guidance inside the bore. We have developed and clinically tried a semi-robotic device and system for planning and execution of prostate biopsy under MRI guidance [5]. We have conducted several clinical trials [5] and more are to follow. The generic workflow is as follows:

  1. Pre-Op: segment the prostate, identify suspicious areas, plan targets;
  2. Intra-Op: import plan, update plan, execute the biopsy/therapy
  3. Post-Op: compare post-op data with plan, evaluate technical variables

Currently, these functions are achieved by fragmented in-house code, some based on VTK/ ITK.

The objective of this DBP will be professional-grade clinical software engineering of existing and upcoming functionality in Slicer. This will allow the team to focus on project specific tasks, and benefit from the Slicer's advances in IGT capability.

Application

Multi-parametric MRI Guided Transrectal Prostate Biopsy

Objective

We would like to create an end-to-end application within the NA-MIC Kit to enable an existing transrectal prostate biopsy device to perform multi-parametric MRI guided prostate biopsy in closed-bore high-field MRI magnets.

This page describes the technology roadmap for robotic prostate biopsy in the NA-MIC Kit. The basic components necessary for this application are:

  • Tissue segmentation: Should be multi-modality, correcting for intensity inhomogeneity and work for both supine and prone patients, all imaged with an endorectal coil (ERC).
  • Registration: co-registration of MRI datasets taken at different times, in different body positions, and under different imaging parameters
  • Prostate Measurement: Measure volume of all segmented structures
  • Biopsy Device Parameters: Geometry, kinematics, and calibration/registration of the robot system must be available in some form. This capability is not currently part of the NA-MIC kit. The application will be modular, to enable use of different devices.
  • Tutorial: Documentation will be written for a tutorial and sample data sets will be provided to perform simulated biopsies.


Roadmap

We will obtain a biopsy plan from multi-parametric endorectal image volumes, executable with an existing prostate biopsy device. The system will be will be implemented under Slicer3 as an interactive application.


Current Status of System Components

  • Segmentation: Semi-automatic segmentation has been implemented by Fichtinger et al. Statistical atlas based segmentation has been prototyped by Tannenbaum et al. Pose-independent segmentation (workable in both supine/prone) is being implemented.
  • Registration: contour based registration has been prototyped by Fichtinger et al., needs to be re-implemented with native NA-MIC tools.
  • Prostate Measurement: Prototyped by Fichtinger et al., needs to be re-implemented with native NA-MIC tools.
  • Device Modeling: Prototyped by Fichtinger et al, needs to be re-implemented with native NA-MIC tools.
  • Biopsy Planning: Clinically functional, needs to be implemented with native NA-MIC tools
  • Tutorial: Not yet started. It will be derived from the actual clinically-functional system, with demo data.


Schedule

Data Collection: Done. Initial data available, hand segmented for ground truth.

Segmentation: We plan to use shape-based segmentation methods for the MRI prostate data. Several parts of the procedure have already been implemented with NA-MIC tools such as the conformal flattening procedure. Spherical wavelets for shape analysis are already available in ITK. Despeckling techniques will be used to enhance ultrasound imagery as a pre-processing step for segmentation of the prostate data.

System Implementation: Apart from the one research element (segmentation), the rest of the project is a massive software engineering effort, and will follow these major milestones and schedule:


Application Workflow Development:

10/15/2007 Define the workflow for the application (David, Gabor, Purang)

10/22/2007 Create GUI templates for the workflow steps (David)


Device Modeling:

11/15/2007 Conversion of engineering data into VTK-viewable objects (David, Gabor)

11/15/2007 Create GUI controls for positioning the virtual device (David)


Data Display:

12/01/2007 Provide display logic for seeds and prostate outlines (David)


Measurement Tools:

01/15/2008 Semi-automatic identification of fiducials via region growing & centroids (new tool) (Siddharth, David)

02/15/2008 Logic for robot registration with fiducials (Siddharth, David)

03/01/2008 Prostate measurement tools (use existing NA-MIC tools) (David, Siddharth)


Biopsy Planning:

03/15/2008 Implement planning tools (display, logic) in Slicer3 (Siddharth, David)

04/01/2008 Verification against pre-existing software and data (Siddharth)


Integration of contour-based prostate registration into Slicer3:

06/01/2008 Create ITK implementation (Siddharth, David, Purang, Gabor)

06/15/2008 Create Slicer3 module, incorporate into workflow (Siddharth)


Integration of prostate segmentation into Slicer3 Siddharth, David)

08/01/2008 Integrate MRI segmentation tools (Siddharth, David, Allen)

08/15/2008 Integrate US segmentation tools (Siddharth, David, Allen)


Verification and testing of integrated system

09/01/2008 Workflow validation (Siddharth, David, Gabor, Purang)

11/15/2008 Complete testing, performance characterization (Siddharth, David)


Tutorial and Documentation

Continuous unit documentation (Siddharth, David)

12/15/2008 Complete tutorial with example data

Experimental Data

The system for MRI guided transrectal prostate interventions was tested in patients [5] and a new embodiment has been tested recently in phantom experiments at NIH (Bethesda, MD) on a 3T Philips Intera MRI scanner (Philips Medical Systems, Best, NL) using standard MR compatible biopsy needles and non artifact producing glass needles [7]. The system has been tried on humans at NIH. Replication of the system for multiple collaborating clinical sites (Princess Margaret Hospital in Toronto, BWH in Boston, Johns Hopkins in Baltimore) is in progress.


Patient Data:

Typically, 3D axial MRI prostate datasets (patient positions mixed between prone and supine) acquired using different endorectal coils (coil diameter = 13 mm for two datasets, coil diameter = 26 mm for third) were used for algorithm evaluation. The scans were performed on Philips Intera 3T MRI system; T2-weighted images acquired using Spin Echo (SE) sequence with following parameters: SENSE protocol with acceleration factor of 1; TE/TR = 180 / 7155 ms for some datasets, TE/TR = 120 / 7155 ms for some others; matrix 256 x 256; field of view 140 x 140 mm; voxel size 0.55 x 0.55 mm; slice thickness 3 mm.

Phantom Data:

Biopsy Needle Accuracies: The manipulator was placed in a prostate phantom and its initial position was registered. Twelve targets were selected within all areas of the prostate on T2 weighted axial TSE images. Targets one through four were selected in the base of the prostate, targets five through eight in the mid gland, and targets nine through twelve in the apex of the prostate. For each target, the targeting program calculated the necessary targeting parameters for the needle placement.

The phantom was pulled out of the MRI scanner on the scanner table, the physician rotated the manipulator, adjusted the needle angle and inserted the biopsy needle according to the displayed parameters.

The phantom was rolled back into the scanner to confirm the location of the needle on axial TSE proton density images which show the void created by the biopsy needle tip close to the target point. The in-plane error for each of the twelve biopsies, defined as the distance of the target to the biopsy needle line was subsequently calculated to assess the accuracy of the system.

The needle line was defined by finding the first and the last slice of the acquired confirmation volume, where the needle void is clearly visible. The center of the needle void on the first slice and the center of the void on the last slice define the needle line. The out of plane error is not critical in biopsy procedures due to the length of the biopsy core and was not calculated. Hence, from the purpose of accuracy, there is no need for a more precise motorized needle insertion. The average in-plane error for the biopsy needles was 2.1 mm with a maximum error of 2.9 mm.

Glass Needle Accuracies: The void created by the biopsy needle is mostly due to susceptibility artifact caused by the metallic needle. The void is not concentric around the biopsy needle and depends on the orientation of the needle to the direction of the main magnetic field in the scanner (B0), and the direction of the spatially encoding magnetic field gradients [6]. Consequently, center of needle voids do not necessarily correspond to actual needle centers.

And since the same imaging sequence and similar orientation of the needle is used for all targets in a procedure, a systematic shift between needle void and actual needle might occur, which introduces a bias in the accuracy calculations. To explore this theory, every biopsy needle placement in the prostate phantom was followed by a placement of a glass needle to the same depth. The void created by the glass needle is purely caused by a lack of protons in the glass compared to the surrounding tissue, and is thus artifact free and concentric to the needle. The location of the glass needle was again confirmed by acquiring axial TSE proton density images. The average in-plane error for the glass needles was 1.3 mm with a maximum error of 1.7 mm.

The procedure time for six needle biopsies not including the glass needle insertion was measured at 45 minutes.

Current Image Software

The targeting program runs on a laptop computer located in the control room. The only data transfer between laptop and scanner computer are DICOM image transfers. The fiber optic encoders from the robot interface via a USB counter (USDigital, Vancouver, Washington) to the laptop computer.

The targeting software displays the acquired MR images, provides the automatic segmentation for the initial registration of the manipulator, allows the physician to select targets for needle placements, it provides targeting parameters for the placement of the needle, and tracks rotation and needle angle change provided by the encoders, while the manipulator is moved on target.

After targeting the software overlays the target and projected needle path with the confirmation volume scan. This allows the physician to quickly asses the success of the intervention.

Image Processing Needs

Although the current software covers the intervention needs for the first project, additional functions are necessary to allow easy and quick access to the data before, during and after procedure, and to accommodate the needs of the other two projects.

Segmentation and deformable registration functions, 3D visualization instead of the current 2D view, and the extensive data analysis technology of Slicer are all on the projects software specification list.

A basic requirement is the good memory management and stable software. Since the program is running on a laptop, there are only very limited resources available (CPU and memory).

The automatic segmentation algorithms are not always accurate, so it should provide interactive correction capabilities, like moving the slider to change the threshold followed by re-segmentation. Other interactive part is modification of the proposed needle path within the robot constraints.

LPS coordinate system: During the procedure targets are selected using the 2D projection image obtained from the scanner and the target coordinates are in the DICOM image coordinate system. This is also used to display parameters for the manual prescription and for the real time tracking.

As each landmarking defines an independent coordinate system, and this Frame of Reference (FoR) correspondence between volumes is essential for the patient safety. The operating personnel should not be able to use registration data from one volume and target other volume if there’s no transformation between the two coordinate systems.

With OpenTracker capable MRI scanners we would like to use real time needle tracking.

Summary

Manifold benefits exist for both NA-MIC and the Brigham-Hopkins joint program in MRI-guided prostate interventions, owing to existing loops of collaborations, cross-compatibility of research (MR guided prostate interventions), and shared Slicer/VTK/ITK based software platforms.

The project's clinical partners are based in the intramural research program of the National Cancer Institute. Thus the proposed NA-MIC DBP will tie a significant segment of extramural cancer research into a prominent intramural effort, thereby leading to a better understanding, coherency, and active collaboration between these otherwise disjoint efforts. For NA-MIC the benefits are also tangible: the functions will be developed in a controlled and professional environment in the CISST ERC that has been in close collaboration with NA-MIC/Brigham. The development environment used in both groups are similar, in that we both base our image processing tools on VTK, ITK and Slicer and uses many of the same development tools, including CVS, CMake, Doxygen and Dart. In short, the proposed work will be conducted on a shared platform (VTK, ITK, and Slicer) with a compatible development process, and thus the results will be directly absorbable by NA-MIC.

Projects

References

  1. Jemal, A., Siegel, R., Ward, E., Murray, T., Xu, J., Thun, M.J.:
    Cancer statistics, 2007
    CA Cancer J Clin 57(1) (2007) 43–66
  2. Yu, K.K., Hricak, H.:
    Imaging prostate cancer
    Radiol Clin North Am 38(1) (2000) 59–85, viii
  3. Norberg, M., Egevad, L., Holmberg, L., Sparn, P., Norln, B.J., Busch, C.:
    The sextant protocol for ultrasound-guided core biopsies of the prostate underestimates the presence of cancer
    Urology 50(4) (1997) 562–566
  4. Terris, M.K.:
    Sensitivity and specificity of sextant biopsies in the detection of prostate cancer: preliminary report
    Urology 54(3) (1999) 486–489
  5. Krieger A, Csoma C, Guion P, Iordachita I, Metzger G, Qian D, Singh A, Whitcomb L, Fichtinger G:
    Design and Preliminary Accuracy Studies of an MRI-Guided Transrectal Prostate Intervention System
    MICCAI 2007
  6. DiMaio, S.P., Kacher, D.F., Ellis, R.E., Fichtinger, G., Hata, N., Zientara, G.P., Panych, L.P., Kikinis, R., Jolesz, F.A.:
    Needle artifact localization in 3T MR images
    Stud Health Technol Inform 119 (2006) 120–125
  7. Krieger A, Csoma C, Iordachita I, Guion P, Fichtinger G, Whitcomb LL,
    Design and Preliminary Accuracy Studies of an MRI-Guided Transrectal Prostate Intervention System
    MICCAI 2007 (accepted)