Software Development For Correction of Gradient-Nonlinearity Distortions in MR Images

1
Software Development For Correction of
Gradient-Nonlinearity Distortions in MR Images

• T.S. Lee, K.E. Schubert
• Computer Science
• CSUSB

R.W. Schulte Radiation Medicine LLUMC
2
Functional Proton Radiosurgery

• Functional Neurosurgery
• Trigeminal Neuralgia
• Parkinsons Disease
• Brain regions (lt 1 cm)
• Proton Radiosurgery
• Accurate to less than 1 mm
• MRI for target localization
• Distinguish tissue types
• 512?512 images
• 262,144 pixels/study
• Gradient nonlinearity distortions (2mm)

3
Example MR Phantom Images
Axial Plane
Coronal Plane
Sagittal Plane
4
Bad Slides
Partial Phantom
Partial Phantom
Off Center Phantom
No Phantom
5
Finding Edges
6
Disconnects
7
Remove Extraneous Features
8
Can We Fix it?
Air Bubble
Leaky Slice
9
Example Edge Images
Axial Plane Coronal Plane Sagittal Plane
10
Midplanes
Stack Midpoints
Fit Midplane
Calculate Midpoints
Ideal shape, size, and orientation of phantoms
faces
11
Ideal Planes

• Shift ½ the phantom dimension
• Perpendicular to face

12
Distortion Modeling

• Magnetic Field of Cylinder
• Sum of spherical harmonics

13
Distortion Modeling
Measured
Corrected
14
Applying Distortion CorrectionAxial Plane
15
Applying Distortion CorrectionCoronal Plane
16
Applying Distortion CorrectionSagittal Plane
17
Results

• Theoretical undistorted points vs. corrected
  points
• Standard deviations of correction
• X face Standard Deviation 0.23503 mm
• -X face Standard Deviation 0.25168 mm
• Y face Standard Deviation 0.15322 mm
• -Y face Standard Deviation 0.15823 mm
• Z face Standard Deviation 0.13548 mm
• -Z face Standard Deviation 0.14173 mm

18
Conclusions

• 3s range 0.4 0.8 mm
• 1-2 pixels on each image
• Originally 2mm (5-6 pixels)
• Accurate localization of anatomical targets

19
Future Work

• Further verification and testing
• Clinical trials
• FDA approval
• Treatment on humans