32
The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL Adaptive Weighted Deconvolution Model to Estimate the Cerebral Blood Flow Function in Dynamic Susceptibility Contrast MRI Jiaping Wang, Ph.D Department of Mathematical Science University of North Texas at Denton Joint work with Drs. Hongtu Zhu and Hongyu An from UNC-CH

Jiaping Wang, Ph.D Department of Mathematical Science University of North Texas at Denton

  • Upload
    shima

  • View
    51

  • Download
    0

Embed Size (px)

DESCRIPTION

Adaptive Weighted Deconvolution Model to Estimate the Cerebral Blood Flow Function in Dynamic Susceptibility Contrast MRI. Jiaping Wang, Ph.D Department of Mathematical Science University of North Texas at Denton Joint work with Drs. Hongtu Zhu and Hongyu An f rom UNC-CH. Outline. - PowerPoint PPT Presentation

Citation preview

Page 1: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Adaptive Weighted DeconvolutionModel to Estimate the Cerebral Blood Flow Function in Dynamic Susceptibility Contrast MRI

Jiaping Wang, Ph.D

Department of Mathematical ScienceUniversity of North Texas at Denton

Joint work with Drs. Hongtu Zhu and Hongyu An

from UNC-CH

Page 2: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Outline

Background and Motivation

Adaptive Weighted De-convolution Model Simulation Studies

Real Data Analysis

Page 3: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Part 1. Background and Motivation

Page 4: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Dynamic Susceptibility Contrast (DSC) Perfusion MRI measures the passage of a bolus of a non-diffusible contrast through the brain. The signal decreases as the bolus passes through the imaging slices.

Background

Page 5: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Convolution Relationship

= R(t) where Ca(t) is the given AIF, C(t) is the observed concentrationfunction, which is computed as S(t)/S0. We are interested in estimating the residue function R(t).

Page 6: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Deconvolution Techniques

Fourier Transformation SVD TSVD at 0.01

TSVD at 0.05 TSVD at 0.1 TSVD at 0.2

Page 7: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Part 2. Adaptive Weighted Deconvolution Model

Page 8: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Notations

D : 3D volume

N : the number of points on D

d : a voxel in D

: : spatial-temporal process

: error process

: AIF function, constant along space : Residue function

}],,0[:),({ DdTtdtC

}],,0[:),({ DdTtdt )(tCa),( dtR

Page 9: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Voxel-wise Approach

),()(),(),()(),(),( dtduuCdutRdttCdRdtC aa

Frequency-Domain

dtTiftdtCdfFT

C ))(/2exp(),(),(0

fdtFfFdfFdfFaCRC for ),()(),(),(

Temporal-Domain

Page 10: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

for ),()(),(),( fdfFfFdfFdfFaCRC

fdffdfdfaCRC for ),()(),(),(

))/(2exp(),(),(0

TiftdtCdfT

tC

( f ,d) ~ (0,1( f f ') ( f , f ';d,d'))

Discrete

Continuous

),(),(for smooth y piecewisel is ),( dfNdfdfH

Key Assumptions:

Page 11: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Two Main Steps (Spatial Adaptive Approach): 1. Transform the time series into the Fourier or Wavelet domain.2. Smoothing the curves in the frequency domain by involving the local neighborhood information.

Voxel-wise vs. Spatio-Interdependence

Jumping Space Irregular Boundary

Page 12: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

),(),(for ),()(),(),( dfNdfdffdfdfaCRC

)','()'(),(

)','()'()','()','(

dffdf

dffdfdf

a

a

CR

CRC

UnknownApproximation

),(),(),);,(()','( hdBrfrfrhdfBdf

Spatial-Adaptive Approach

Page 13: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

),;',',,()]'(),()','([)),);,(();,(( 2

)','(

rhdfdfwfdfdfhrdfBdfLaCRC

dfR

Weighted LSE

),;',',,()'()'(/),;',',,()'()','(),(ˆ)','()','(

rhdfdfwffrhdfdfwfdfdfaaa CC

dfCC

dfR

)),(ˆ(Var dfR

)/2/()]/2cos(1)[2exp(),(ˆ),(ˆ 221

0

TtTttfidfdtRT

kkkR

Estimated HRF

Page 14: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Being HierarchicalDrawing nested spheres with increasing radiuses at each voxel and each frequency

S

S

rrrhhh

10

10

Page 15: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

• Sequentially determine weights • Adaptively update

Being Adaptive

)(),(ˆ iR df

)(),;',',,( irhdfdfw

))','(ˆ,),(ˆ(|)'(||)'(|),;',',,( )1()1(2,1,

)( iR

iRhlochloc

i dfdfKddKffKrhdfdfw

)1(),;',',,( irhdfdfw

)1(),(ˆ iR df

)(),;',',,( irhdfdfw

)(),(ˆ iR df

)),(ˆ),(ˆ( )1()( iR

iR dfdfS

Stopping Statistics

Page 16: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

How to determine ?

Page 17: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Page 18: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Page 19: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Page 20: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Page 21: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Page 22: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Page 23: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Part 3. Simulations

Page 24: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

(i) A temporal cut of the true images;(ii) The true curves C(t)

(iii) The true curves R(t)

Simulation Set-up

The true residue curves

(iv) The AIF Curve

(i) (ii) (iii) (iv)

Page 25: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Simulation Results

Result from SWADM Cluster Result Mean Curves of Clusters from SWADM

Comparison Statistics: Dd=

Where Xd is the estimated curve from the proposed method, Yd is from other methods including voxel-wise inverse Fourier Transformation (IFT), SVD, TSVD at thresholds 0.01, 0.05, 0.1 and 0.2, respectively. ||•|| is a norm operator.

Page 26: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

(1) Comparison with SVD;

Comparison Results

(2) Comparison with TSVD at 0.01; (3) Comparison with TSVD at 0.05;

(4) Comparison with TSVD at 0.1;(5) Comparison with TSVD at 0.2; (6) Comparison with Voxel-wise IFT;

Page 27: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Comparison along SNRs

Average of Dd along different SNRs One sample t test for Dd

TSVD at 0.01

TSVDat 0.05

TSVDat 0.1

TSVDat 0.2

Voxel-wiseIFT

SVD

Page 28: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Part 4. Real Data

Page 29: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

(e)

The DSC PWI data set obtained from an acute ischemic strokepatient at Washington University in St. Louis after receiving a signed consent form with Institutional Review Board approval.

MR images were acquired on a 3T Siemens whole body Trio system (Siemens Medical Systems, Erlangen, Germany). PWI imageswere acquired with a T2*-weighted gradient echo EPI sequence (TR/TE= 1500/43 ms,14 slices with a slice thickness of 5 mm, matrix= 128x128). This sequence was repeated 50 times and Gadolinium diethylenetriamine penta-acetic acid (Gd-DTPA, 0.1 mmol/kg) was injected at the completion of the 5th measure.

Data Description

Page 30: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Slices from C(t) images

Sample of C(t) curves, the largest one can be considered as AIF.

Clustered pattern

Mean curves of clusters

Clustering Results

Page 31: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Estimation Results from Different Methods

The curves from same voxel in Cluster I The curves from same voxel in Cluster II

Page 32: Jiaping  Wang,  Ph.D Department of Mathematical Science University of North Texas at Denton

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Thank You!