High Precision Drill/Needle Placement with UR5


Pedicle screw placement procedures require a high degree of precision and accuracy in screw positioning in order to yield the most successful outcomes. As the current standard of care often involves a clinician manually placing a pedicle screw in a patient based upon knowledge and experience, there exists a range of placement error that could be minimized with some sort of assistance. Dr. J. Siewerdsen of the JHMI I-STAR lab has acquired a UR5 robotic arm for potential use in image-guided surgery. Though the UR5's usage is typically envisioned in a more industrial environment, the 6-DOF robot arm’s application will be expanded to non-invasively assist with pedicle screw placement, thereby promoting ease, efficiency, and accuracy of such procedures. Ultimately, we aim to universally improve the quality of pedicle screw placement for patients.

  • Students: Vignesh Ramchandran & Thomas Yi
  • Mentor(s): Dr. Jeff Siewerdsen & Ali Uneri


Background, Specific Aims, and Significance

Pedicles constitute small structures in vertebral segments that are often chosen as a gateway to anchoring pedicle screws that may be embedded for a variety of reasons (spinal stability, correction, etc.). In pedicle screw placement procedures, a “successful” procedure may be defined as one in which a physician has secured the screw within a surgical “acceptance window” in a vertebrae as shown in the figure below. In terms of methodology, a physician will often attempt to manually place a pedicle screw in a patient based upon cumulative experiences. This “free-hand” technique presents an array of errors that could otherwise be minimized with some sort of surgical assistance/guidance. Given that complications include spinal cord breach or dislodgement (which could lead to paralysis or infection), it is imperative that a pedicle screw is properly positioned/secured.


Given that the UR5 is a 6-DOF robot arm capable of fluid, forward kinematics, it would be beneficial to pedicle screw patients if we could adapt the arm for use in high-precision, image-guided drill guide placement. To elaborate, suppose a patient is to have pedicle screws placed into his or her spine. One could consider that the patient is likely to have some spinal, pre-operative CT data. Thus, if the patient’s physical body volume was registered to this CT volume using two, orthogonal intraoperative radiographs via 3D-2D registration, one would effectively have a digital version of the patient to devise path planning with. Following, if one were to register the UR5 to this CT volume and do some initial axis planning, the UR5 could position a drill-guide along said axis to provide a physician with a simple and safe method of placing a pedicle screw into a vertebrae. In other words, we ultimately aim to be able to position a drill-guide at a prescribed pose relative to the patient so as to assist a physician with threading the screw into a patient


  • Minimum: (Expected by 3/1/2016)
    1. Enable tracker based guidance for UR5 robot (i.e. register robot to tracking system)
    2. Experimental minimization/validation of calibration error
  • Expected: (Expected by 4/20/2016)
    1. Perform 2D-3D registration between radiographs and CT volume
    2. Integrate image-based guidance for UR5
    3. Experimental optimization of axis planning and error reduction
  • Maximum: (Expected by 5/1/2016)
    1. Devise path planning for robot motion in cadaver studies
    2. Additional details to come as appropriate

Technical Approach


The first course of action is to collect AX=XB calibration data by collecting optical tracker and UR5 positions at varying poses in a given 3 space. This data can then be used to mathematically compute a transformation from the robot tip to gripping tool tip. Following this calibration, experiments to test error in cardinal directions as well as rotational angles will be conducted to verify that our calibration method and AX=XB solver meets our goal of minimizing all translation errors to be < 1.5 mm and rotational errors to be < 1 radian relative to the intended pose. Once we have verified that we are able to acquire a “good” calibration for our robot, the next step is to repeat the process with a CT volume and accompanying phantom. This process will first entail a similar calibration between the given CT volume and the UR5 robot. Then, we would register the phantom to the 3D-2D registration using two orthogonal radiographs of the phantom. Once this calibration and registration is complete, we would do preliminary axis planning to optimize UR5 path planning. Then, we plan to conduct experiments with the physical phantom itself in order to identify sources of calibration error and attempt to minimize the ones we find. Finally, once we validate and refine our calibration process – ie registering the UR5 to a CT volume that is registered to a 2D radiograph – we can pursue further testing in cadavers to test how well and efficiently the robot can place a drill guide along the pedicle axis and have the orientation and position be within the acceptance window. Beyond this, we hope to pursue the integration of the UR5 with live patient procedures.


  • Transportation to working site - readily available in the form of the JHMI shuttle
  • Fully operational UR5 that can be modified by program - provided by I-STAR Lab
  • Fully operational optical tracker along with OT markers - provided by I-STAR Lab
  • Optical tracking tools (calibrated) - provided by I-STAR Lab
  • Work bench for UR5 mounting - provided by I-STAR Lab (to be modified as necessary)
  • Computer for UR5 programmatic control and loaded with visualization software for optical tracking - provided by I-STAR Lab
  • 3D-2D registration software (in TREK) - obtained from I-STAR Lab
  • Imaging device to acquire intra-operative radiographs - provided by I-STAR LAb
  • CT data accompanied by corresponding phantom - obtained from I-STAR Lab
  • Machine shop access to modify drill guide design (along with safety equipment) - readily available in Carnegie
  • Mentors - Dr. Siewerdsen and Ali generously devote time to us to ensure best possible progress
  • Additional dependencies will likely arise as we dive deeper into the project

Milestones and Status

  1. Milestone name: Initial setup of UR5 w/ optical tracking system
    • Planned Date: 2/1/2016
    • Expected Date: 2/1/2016
    • Status: Complete
  2. Milestone name: Develop proficiency with UR5 SDK (URX)
    • Planned Date: 2/15/2016
    • Expected Date: 2/15/2016
    • Status: Complete
  3. Milestone name: Register UR5 to tracker via AX=XB calibration
    • Planned Date: 3/1/2016
    • Expected Date: 3/1/2016
    • Status: Complete
  4. Milestone name: Validate UR5 to OT registration with error analysis experiments
    • Planned Date: 3/8/2016
    • Expected Date: 3/8/2016
    • Status: Complete
  5. Milestone name: Develop proficiency 3D-2D registration techniques/modules
    • Planned Date: 3/20/2016
    • Expected Date: 3/20/2016
    • Status: Complete
  6. Milestone name: Acquire phantom with corresponding CT data
    • Planned Date: 3/20/2016
    • Expected Date: 3/20/2016
    • Status: Complete
  7. Milestone name: Register UR5 to phantom CT image
    • Planned Date: 3/30/2016
    • Expected Date: 3/30/2016
    • Status: Complete
  8. Milestone name: Validate UR5 to phantom CT image registration
    • Planned Date: 4/10/2016
    • Expected Date: 4/10/2016
    • Status: Complete
  9. Milestone name: Develop drill guide for UR5
    • Planned Date: 4/15/2016
    • Expected Date: 4/15/2016
    • Status: Complete
  10. Milestone name: Drive UR5 w/ drill guide to prescribed pose along phantom surface
    • Planned Date: 4/25/2016
    • Expected Date: 4/25/2016
    • Status: Complete
  11. Milestone name: Minimized drill guide deviations for optimal assistance of pedicle screw placement
    • Planned Date: 4/30/2016
    • Expected Date: 5/5/2016
    • Status: Complete-to be refined

Reports and presentations

Project Bibliography

Gramkow, Claus. “On Averaging Rotations”. International Journal of Computer Vision 42.1/2 (2001): 7-16. Web. 10 Feb. 2016.

Puvanesarajah, Varun. “Techniques And Accuracy Of Thoracolumbar Pedicle Screw Placement”. WJO 5.2 (2014): 112. Web. 10 Feb. 2016.

Markelj, P. et al. “A Review Of 3D/2D Registration Methods For Image-Guided Interventions”. Medical Image Analysis 16.3 (2012): 642-661. Web. 4 Feb. 2016.

Otake, Y et al. “Automatic Localization Of Vertebral Levels In X-Ray Fluoroscopy Using 3D-2D Registration: A Tool To Reduce Wrong-Site Surgery”. Physics in Medicine and Biology 57.17 (2012): 5485-5508. Web. 25 Mar. 2016.

Penney, G.P. et al. “A Comparison Of Similarity Measures For Use In 2-D-3-D Medical Image Registration”. IEEE Transactions on Medical Imaging 17.4 (1998): 586-595. Web. 25 Mar. 2016.

Shah, Mili, Roger D. Eastman, and Tsai Hong. “An Overview Of Robot-Sensor Calibration Methods For Evaluation Of Perception Systems”. Proceedings of the Workshop on Performance Metrics for Intelligent Systems - PerMIS '12 (2012): n. pag. Web. 4 Feb. 2016.

Uneri, A et al. “Known-Component 3D–2D Registration For Quality Assurance Of Spine Surgery Pedicle Screw Placement”. Physics in Medicine and Biology 60.20 (2015): 8007-8024. Web. 25 Mar. 2016.

Uneri, A et al. “3D–2D Registration For Surgical Guidance: Effect Of Projection View Angles On Registration Accuracy”. Physics in Medicine and Biology 59.2 (2013): 271-287. Web. 25 Mar. 2016.

Other Resources and Project Files

Code repository controlled via Git–additional permissions required for access.

courses/446/2016/446-2016-01/project_01_main_page.txt · Last modified: 2019/08/07 12:01 (external edit)