MRI image validation using MRI simulation Emily Koch

MRI image validation using MRI simulation Emily Koch CIS II April 10, 2001

The Problem • Validation of MRI based images can be difficult. • Without landmarks there is no guarantee that the image is correct. • Need to evaluate the effectiveness of a postimaging algorithm. • Without “base standard” there is no guarantee that the post-imaging processing was accurate

• Flexibility of MRI makes it extremely difficult to set a known standard to compare against – Differences in image contrast – Differences in image quality

Goal • Want to create realistic image of known object. • The more accurate the image of the object, the more accurate the image of the unknown object • Want to create the maximally accurate image of known objects

References • R. K. -S. Kwan, MRI Stimulation for Quantitative Evaluation of Image-Processing Methods, www. bic. mni. mcgill. ca/users/rkwan • Remi K. -S. Kwan, Alan C. Evans, G. Bruce Pike. An Extensible MRI Simulator for Post-Processing Evaluation. Visualization in Biomedical Computing (VBC’ 96). Proceedings. Lecture Notes in Computer Science, vol. 1131. Springer-Verlag, 1996. 135 -140.

Solutions • Creation of a physical phantom – Expensive – Time consuming • Multiple image relationships – Expensive – Invasive – Time Consuming • Simulation of MRI images to create a “absolute base-line” for studies

Simulation of MRI images • Program developed using Object Oriented Design techniques • Simulation involves two different aspects: – Signal Production – Image Production

Simulator Design Spin Model Phantom Pulse Sequence Scanner image Signal Production RF Coil Image Production

Signal Production • Timing of events in the signal production are described by the Pulse Sequence model – RF pulses • Message sent to Spin Model as a pulse is applied to an event

The Spin Model • Current state of tissue magnetization • Illustrates behavior under influence of events: – RF pulses, gradient fields, relaxation • Interface: defines everything that must be implemented in all subsequent models • All extraneous data is hidden so that the behavior of the model can be determined by only the model being used

Image Production • Signal Production Models -> Image Production Models -> MRI Volumes • Phantom Model: spatial distribution of tissues and properties of the tissues • Scanner Model: coordination of all components, interface to the Pulse Sequence Model

• RF Coil Model: control of signal reception – Noise control • Different RF Coil Models: – Simulate noiseless conditions – Noise level depending on imaging parameters • Slice thickness

Creating Realistic Images • To create realistic phantoms from the MRI simulator, the author input pre-labeled data set generated from a MRI volumetric data set – 3 D brain model pre-labeled • Signal Production Simulation: – Signal intensities are calculated from the data – Mapped to create a pseudo-MRI volume

Basic Results

Method Evaluation • Sharp tissue boundaries - possible to smooth using higher resolution or blurring the edges of the data set • Highly accurate reconstruction of the original image • Useful in the evaluation of image contrast and image slice size

f. MRI Results

No motion Motion Corrected

Evaluation • This information was the result of Kwan’s masters project • Little other information on the subject was found. • Most of the information is old- the latest information that was used was published in 1997.

• This method is potentially very useful in the creation of a database of brain function • Extremely important to validate the results of the testing as the goal is to create an atlas. • The creation of a simulation program would be very time consuming but validation would be necessary for the success of the long term goals of the project.