Multivariate study for newly diagnosed glioma patients with MRIMRSI

1
Multivariate study for newly diagnosed glioma
patients with MRI/MRSI
Xiaojuan Li May 16th, 2001 UCB/UCSF
bioengineering program
2
Introduction

• The glioma is the most common type of primary
  brain tumors with approximately 15,000 newly
  diagnosed each year in US.

• Gliomas ranges from benign to highly malignant
  lesions that have different survival curve.
  Therefore proper classification and grading of
  gliomas is of great importance.
• The clinical practice of classification is almost
  always established by histological examination of
  the biopsy samples, which is invasive and
  sometimes risky, and is subject to location
  error.

3
Goals

• To find a non-invasive method to evaluate
  differences in morphologic and metabolic
  parameters with tumor grade.
• study the spatial distribution of the morphologic
  and metabolic abnormalities within the lesion

4
Method and materials

• Conventional MRI
• MR spectroscopy imaging (MRSI)
• Advanced MRI diffusion and perfusion weighted
  imaging
• 20 normal volunteers
• 60 newly-diagnosed patients
• 20 grade 2, 20 grade 3 and 20 grade 4

5
Method
6
MRI/MRSI protocol (1)

• 1.5T GE Signa Echospeed clinical scanner
• T1-weighted sagital image
• ?
• dual-echo proton density/T2-weighted images
• Te30,80ms, Tr2500ms, 48 interleaved slices
  with 3mm thickness
• T2-weighted FLAIR (Fluid Attenuated Inversion
  Recovery) image
• ?
• pre- and post-contrast 3D SPGR (spoiled gradient
  echo) T1-weighted images
• Te8ms, Tr32ms, flip angle45, 124 slices with
  1.5mm thickness

7
MRI/MRSI protocol (2)

• Multi-slices MRSI
• PRESS (point resolved spectroscopy) localization
• Phase-compensating spatial-spectral pulses for
  better water suppression and decreased chemical
  shift misregistration
• VSS (very selective suppression) pulse for
  better spatial suppression
• Tr1s, Te144ms
• voxel size of 1cc
• 1688 or 12128 or 888 3-D phase encoding

8
Spectral data processing
9
MRI/MRSI
MRSI
Contrast-enhanced T1w MRI
T2w MRI

• Cho component of membrane
• Cr/PCr energy buffer and shuttle
• NAA neuronal marker
• Lactate end product of anaerobic metabolism
• Lipid product of membrane break-down

10
Spectral quantification

• Perform a linear regression of entire 3D-MRSI
  dataset
• Calculate the z-scores (residual/s) of the
  perpendicular residuals
• Define Control Population z ? 2.0
• Exclude Voxels with z gt 2.0
• Repeat 1 - 4 until there are no more points to
  remove

11
Study of spatial distribution

• Resample the metabolic image
• Make contour image based on the resampled images
• Calculate the volume

12
Data acquisition ADC (Apparent Diffusion
coefficient)

• Molecular diffusion refers to the translational
  movement of water and other small molecules in
  tissue caused by thermal processes. The rate of
  water diffusion reflects intrinsic tissue
  properties.

ADC map
13
Data acquisition -- rCBV

• Cerebral blood volume is the fraction of the
  volume of tissue that is occupied by blood
  Elevated relative regional (rCBV) reflects the
  increased micorvascularity which is associated
  with brain tumors.

rCBV map
14
Principle Component Analysis
The PCA seeks linear combinations of the
variables (called principle components) with
maximum variance. Then the PC can be applied to
summary the data, losing in the process as little
information as possible.

• X (N by K matrix, with N samples K variables)
• Variance matrix of X
• Eigenvector of S are PC, with corresponding
  eigenvalue showing the variance accounted for by
  the associated eigenvector

15
Discriminant analysis
The problem that is addressed with discriminant
analysis is how well it is possible to separate
two or more groups of individuals, given
measurements for these individuals on several
variables.

• The maximum likelihood (ML) discriminant rule
  allocates an observation x to one of the
  populations G1,Gg which gives the largest
  likelihood to x, i.e., allocates x to Gr where

In this study, we assume the populations in
normal distribution. Then the likelihood
function for group i can be written as
16
Discriminant analysis

• Fishers linear discriminant rule looks for the
  linear function that maximized the ratio of the
  between-groups sum of squares to the
  within-groups sum of squares. Let

max
Linear function
Once this linear function has been calculated, an
oberservation x can be allocated to one of the g
populations on the basis of its discriminant
score.
17
Discuss

• Summarize any actions required of your audience
• Summarize any follow up action items required of
  you