Paris Flood

Paris Flood
Mentor: Dr. Scott Banks
College of Engineering
 
"I enjoy computational mathematics/scientific computing and was given an opportunity to use these passions to make a contribution to the field of orthopedic biomechanics. It's  rewarding to see other researchers use my ideas/software."

Major

Mathematics, Industrial & Systems Engineering

Minor

N/A

Research Interests

  • Computational Science/Mathematics
  • Medical Image Registration
  • Visualization

Academic Awards

  • College of Engineering Scholarship - John W. & Mittie Collins Award
  • National Merit
  • Anderson Scholar

Organizations

  • Gary J. Miller Orthopedic Biomechanics Laboratory
  • Entezari Research Group

Volunteer

  • Jordan Klausner Foundation

Hobbies and Interests

  •  Piano
  • Architecture
  • Traveling North Florida
  •  Dunking

Research Description

A Robust, Automated Method for Registration of Three-Dimensional Bone and Artificial Implant Models to Fluoroscopy Images

Total Knee Arthroplasty (TKA) is a common surgical procedure in which the weight-bearing surface of the knee joint is replaced with an artificial implant. The implant consists of a polyethylene insert and two metallic components that are impacted onto the bone or set using polymethylmethacrylate (PMMA) cement. Although many evaluations have shown TKA significantly reduces pain and increases mobility, high shear stress on the polyethylene insert can lead to accelerated wear and premature failure. In vivo implant kinematic data may allow for the design of implants with better longevity and performance. X-ray fluoroscopy is an important tool to study in vivo implants because it provides a perspective projection of the knee with high contrast between soft-tissues and metallic implants. Although fluoroscopic films only provide information about the silhouette of the implant, pose information (translation coordinates and Euler angles) can be recovered by registering a three-dimensional CAD model to the X-ray. Current registration procedures use a combination of edge-based and intensity-based metrics to measure the similarity between a fluoroscopic image and the projection of an implant. A global optimizer (often simulated annealing) is then used to find the true pose of the implant within an accepted degree of accuracy. However, image noise, occlusions, and low object-to-background contrast limit the search range of the optimizer to a few millimeters/degrees. The small size of the search space severely increases the time it takes to extract the kinematics because every frame requires an observer to determine a highly accurate pose estimate. This can also lead to misregistration of the model if the true pose does not lie within the narrow tolerance of the optimizer. Fluoroscopic imaging captures X-ray stills at a real-time rate of 25-30 frames a second. As the patient is being measured during activities such as gait, stair step, and chair rise, the maximum translation and rotation of an implant from frame to frame is approximately 40 millimeters/degrees. My research investigates a robust method that expands the convergence range of the registration routine to this maximum - essentially automating the process. I intend to achieve an increase in range by implementing a modified similarity metric and a novel Lipschitzian optimization algorithm.