Deconvolution of Motion Blurred Neuro-PET Images

John Votaw and Nivedata Ragunath

Aim: This project involves the implementation of tracking and deconvolution methods for correcting for head motion during FDG PET acquisition.

Tracking: The Polaris Vicra (Northern digital Inc.) is used as the optical tracking system. The system uses a position sensor (or tracker) to detect infrared-emitting or retro-reflective markers affixed to a tool or object.

Figure

Figure 1. (Left) The Polaris Vicra tracker mounted on the rear of the scanner gantry (Center) Reference tool mounted inside the scanner, (Right) reference tool

Figure 2. Tool that is attached to the head

The position of the tracking tool is returned by the Vicra as quaternions (rotations) and translations. This is converted to transformation matrices that, after calibration are used in the deconvolution process.

Graphs

Figure 3. Example graphs of the quaternions and translations (mm) over the number of samples.

Deconvolution: A phantom study is described here. The Hoffman phantom filled with radioactivity was imaged in 5 different positions (simultaneously recorded by the Vicra) to create a blurred image. The phantom was also imaged without any motion to create the reference or motionless image.

The blurred image was deconvolved based on the iterative maximum likelihood method.

Tij is determined from independent (Vicra) measurements of the motion of the patient during the scan. Storing Tij is memory intensive as well as a challenging task. To make deconvolution of large images feasible, we resorted to using ordered subsets.

Deconvolution of a 131x153x11 image using 125 subsets and 6 iterations is shown in Video 1

Figure 5. Deconvolution of a blurred phantom image.

Details of the deconvolution:

  • Image size 169x160x111
  • Number of transformations: 5
  • Number of subsets: 8
  • Number of iterations: 2
  • Time taken (2.8 GHz, 1.5 GB RAM)
  • Creating Tijs: 0.895 min
  • Deconvolving: 1.738 min

Conclusion: Visually the corrected image looks far superior to the blurred image, and is closer to the motionless image in appearance.