[1]	
  Stark	
  DD.	
  Radiology,	
  1991;	
  179:333–5.	
  
[2]	
  Anderson	
  LJ,	
  et.	
  al.	
  Eur	
  Heart	
  J.,	
  2001;	
  22(23):2171-­‐9.	
  
[3]	
  Kellman	
  PK,	
  et.	
  al.	
  J	
  Cardiovasc	
  Magn	
  Reson.,	
  2013;	
  15(1):
56.	
  
The	
  esSmate	
  SD’s	
  turned	
  out	
  to	
  be	
  significantly	
  close	
  to	
  the	
  measured	
  
SD’s.	
  Below	
  are	
  the	
  acquired	
  phantom	
  images	
  and	
  their	
  associated	
  T2*,	
  
measured	
  SD,	
  and	
  esSmated	
  SD	
  maps.	
  
	
  
	
  
	
  
	
  
Reliable	
  idenSficaSon	
  of	
  T2*	
  
abnormaliSes	
  is	
  limited	
  by	
  	
  
measurement	
  noise	
  which	
  	
  
Reduces	
  the	
  precision	
  of	
  T2*	
  	
  
esSmaSon.	
  Knowing	
  how	
  precise	
  	
  
a	
  T2*	
  measurement	
  is	
  important	
  	
  
for	
  prototyping	
  new	
  techniques	
  or	
  	
  
if	
  a	
  clinician	
  is	
  deciding	
  how	
  	
  
aggressively	
  to	
  treat	
  a	
  paSent	
  	
  
with	
  a	
  border	
  line	
  case.	
  We	
  	
  
propose	
  a	
  method	
  for	
  	
  
evaluaSng	
  the	
  precision	
  of	
  T2*	
  	
  
mapping	
  techniques	
  by	
  esSmaSng	
  	
  
the	
  standard	
  deviaSon	
  (SD)	
  of	
  
each	
  measurement	
  thereby	
  
creaSng	
  an	
  SD	
  map.	
  
Evaluating precision of T2* mapping methods
for MRI tissue characterization
Christopher Sandino1, Hui Xue2, Peter Kellman2
	
  
	
  
The	
  standard	
  deviaSon	
  for	
  each	
  T2*	
  can	
  be	
  esSmated	
  by	
  esSmaSng	
  the	
  
parameters’	
  (A,	
  T2*)	
  covariance	
  matrix	
  [3]:	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
All	
  methods	
  were	
  implemented	
  in	
  C++	
  so	
  that	
  they	
  can	
  be	
  tested	
  and	
  ran	
  
on	
  an	
  MRI	
  scanner	
  using	
  medical	
  image	
  reconstrucSon	
  framework	
  
Gadgetron.	
  
	
  
Experimental	
  Valida/on	
  
In	
  order	
  to	
  validate	
  the	
  SD	
  map	
  formulaSon,	
  	
  
we	
  made	
  six	
  gel	
  phantoms	
  made	
  of	
  50	
  mL	
  	
  
soluSons	
  of	
  1.5%	
  agarose	
  gel,	
  cupric	
  sulfate	
  	
  
(CuSO4),	
  saline	
  and	
  varying	
  concentraSons	
  of	
  	
  
Feridex	
  I.V.,	
  an	
  intravenous	
  contrast	
  agent	
  	
  
containing	
  colloids	
  of	
  iron	
  oxide.	
  The	
  	
  
phantoms	
  were	
  scanned	
  N=64	
  Smes	
  to	
  measure	
  the	
  “true”	
  SD	
  and	
  
compare	
  them	
  to	
  the	
  esSmated	
  SD	
  on	
  a	
  per-­‐pixel	
  basis.	
  
	
  
Measured	
  and	
  esSmated	
  	
  
SD’s	
  were	
  compared	
  for	
  
two	
  methods	
  of	
  T2*	
  	
  
mapping	
  (convenSonal	
  
and	
  truncated)	
  and	
  	
  
both	
  showed	
  >99%	
  	
  
correlaSon.	
  Note	
  that	
  	
  
higher	
  SDs	
  are	
  biased	
  	
  
upwards.	
  These	
  correspond	
  	
  
to	
  curves	
  with	
  longer	
  T2*	
  	
  
which	
  aren’t	
  sampled	
  at	
  long	
  enough	
  echo	
  Smes.	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
In-­‐vivo	
  studies	
  showed	
  relaSvely	
  high	
  SD	
  in	
  the	
  
bloodpool,	
  the	
  heart-­‐lung	
  interface,	
  and	
  areas	
  where	
  
moSon	
  occurred.	
  The	
  figure	
  above	
  shows	
  a	
  possible	
  
clinical	
  applicaSon	
  where	
  the	
  SD	
  map	
  is	
  used	
  to	
  mask	
  
unreliable	
  pixels.	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  	
  	
  	
  	
  T2*	
  mapping	
  is	
  a	
  widely	
  used,	
  non-­‐invasive	
  
magneSc	
  resonance	
  imaging	
  (MRI)	
  technique	
  used	
  to	
  
detect	
  iron	
  overload	
  in	
  Sssue.	
  High	
  concentraSons	
  of	
  
iron	
  change	
  the	
  paramagneSc	
  properSes	
  of	
  the	
  Sssue	
  
around	
  it,	
  including	
  its	
  T2*	
  -­‐	
  the	
  transverse	
  relaxaSon	
  
Sme	
  aler	
  an	
  electromagneSc	
  excitaSon	
  [1].	
  
	
  
	
  	
  
	
  
	
  
	
  
	
  
	
  
The	
  T2*	
  relaxaSon	
  curve	
  
is	
  measured	
  by	
  sampling	
  	
  
MR	
  signal	
  intensiSes	
  at	
  	
  
various	
  echo	
  Smes	
  (TE)	
  [2].	
  
We	
  know	
  that	
  this	
  curve	
  is	
  
of	
  the	
  form	
  	
  
Yi=A*exp(-­‐TEi/T2*),	
  so	
  we	
  	
  
can	
  use	
  an	
  exponenSal	
  	
  
regression	
  to	
  esSmate	
  T2*	
  
	
  for	
  each	
  pixel	
  thereby	
  	
  
creaSng	
  a	
  map	
  such	
  as	
  the	
  	
  
one	
  shown	
  to	
  the	
  right.	
  	
  
This	
  parScular	
  paSent	
  has	
  low	
  T2*	
  in	
  the	
  liver,	
  a	
  sign	
  
of	
  iron	
  overload	
  due	
  to	
  β-­‐thalassemia.	
  
Introduc/on	
  
Objec/ve	
  
Methodology	
  
0 5 10 15 20
0
1
2
3
4
5
6
7
8
Echo Time (ms)
SNR
Observed
Truth
Fitted
0 5 10 15 20
0
1
2
3
4
5
6
7
Echo Time (ms)
SNR
Observed
Truth
Fitted
Low	
  SNR	
  
Low	
  T2*	
  
y	
  =	
  1.0446x	
  -­‐	
  0.0004	
  
R²	
  =	
  0.99959	
  
0	
  
1	
  
2	
  
3	
  
0	
   1	
   2	
   3	
  
Es/mated	
  SD	
  (ms)	
  
Measured	
  SD	
  (ms)	
  
T2*
meas.
SD
meas.
SD
est.
time
(ms)
RF
1.6 3.9 6.2 8.5 10.8 13.2 15.5 17.8
Mul -echo phantom images
Observed
Non-Linear
Regression
Estimated
TE
y
TE
y'
TE
y-y'
Residuals
Robust
Estimation
of variance
med(abs(residuals))
Compute
Covariance
Matrix
Results	
  
Good	
  Breath-­‐hold	
  Poor	
  Breath-­‐hold	
  
T2*	
  map	
   SD	
  map	
   T2*	
  map	
  w/	
  mask	
  
	
  
	
  
•  SD	
  esSmaSon	
  efficiently	
  and	
  accurately	
  provides	
  a	
  
way	
  to	
  analyze	
  T2*	
  precision	
  
•  Useful	
  for	
  prototyping	
  new	
  T2*	
  esSmaSon	
  
methods	
  
•  PotenSally	
  useful	
  for	
  clinical	
  use	
  and	
  re-­‐
determining	
  thresholds	
  for	
  disease	
  management	
  
Conclusions	
  
References	
  
1Department of Electrical Engineering, University of Southern California, Los Angeles, CA 2Laboratory of Cardiac Energetics, National Heart, Lung, and Blood Institute, Bethesda, MD

Evaluating precision of T2* mapping methods for MRI tissue characterization

  • 1.
      [1]  Stark  DD.  Radiology,  1991;  179:333–5.   [2]  Anderson  LJ,  et.  al.  Eur  Heart  J.,  2001;  22(23):2171-­‐9.   [3]  Kellman  PK,  et.  al.  J  Cardiovasc  Magn  Reson.,  2013;  15(1): 56.   The  esSmate  SD’s  turned  out  to  be  significantly  close  to  the  measured   SD’s.  Below  are  the  acquired  phantom  images  and  their  associated  T2*,   measured  SD,  and  esSmated  SD  maps.           Reliable  idenSficaSon  of  T2*   abnormaliSes  is  limited  by     measurement  noise  which     Reduces  the  precision  of  T2*     esSmaSon.  Knowing  how  precise     a  T2*  measurement  is  important     for  prototyping  new  techniques  or     if  a  clinician  is  deciding  how     aggressively  to  treat  a  paSent     with  a  border  line  case.  We     propose  a  method  for     evaluaSng  the  precision  of  T2*     mapping  techniques  by  esSmaSng     the  standard  deviaSon  (SD)  of   each  measurement  thereby   creaSng  an  SD  map.   Evaluating precision of T2* mapping methods for MRI tissue characterization Christopher Sandino1, Hui Xue2, Peter Kellman2     The  standard  deviaSon  for  each  T2*  can  be  esSmated  by  esSmaSng  the   parameters’  (A,  T2*)  covariance  matrix  [3]:                                   All  methods  were  implemented  in  C++  so  that  they  can  be  tested  and  ran   on  an  MRI  scanner  using  medical  image  reconstrucSon  framework   Gadgetron.     Experimental  Valida/on   In  order  to  validate  the  SD  map  formulaSon,     we  made  six  gel  phantoms  made  of  50  mL     soluSons  of  1.5%  agarose  gel,  cupric  sulfate     (CuSO4),  saline  and  varying  concentraSons  of     Feridex  I.V.,  an  intravenous  contrast  agent     containing  colloids  of  iron  oxide.  The     phantoms  were  scanned  N=64  Smes  to  measure  the  “true”  SD  and   compare  them  to  the  esSmated  SD  on  a  per-­‐pixel  basis.     Measured  and  esSmated     SD’s  were  compared  for   two  methods  of  T2*     mapping  (convenSonal   and  truncated)  and     both  showed  >99%     correlaSon.  Note  that     higher  SDs  are  biased     upwards.  These  correspond     to  curves  with  longer  T2*     which  aren’t  sampled  at  long  enough  echo  Smes.                                     In-­‐vivo  studies  showed  relaSvely  high  SD  in  the   bloodpool,  the  heart-­‐lung  interface,  and  areas  where   moSon  occurred.  The  figure  above  shows  a  possible   clinical  applicaSon  where  the  SD  map  is  used  to  mask   unreliable  pixels.                        T2*  mapping  is  a  widely  used,  non-­‐invasive   magneSc  resonance  imaging  (MRI)  technique  used  to   detect  iron  overload  in  Sssue.  High  concentraSons  of   iron  change  the  paramagneSc  properSes  of  the  Sssue   around  it,  including  its  T2*  -­‐  the  transverse  relaxaSon   Sme  aler  an  electromagneSc  excitaSon  [1].                   The  T2*  relaxaSon  curve   is  measured  by  sampling     MR  signal  intensiSes  at     various  echo  Smes  (TE)  [2].   We  know  that  this  curve  is   of  the  form     Yi=A*exp(-­‐TEi/T2*),  so  we     can  use  an  exponenSal     regression  to  esSmate  T2*    for  each  pixel  thereby     creaSng  a  map  such  as  the     one  shown  to  the  right.     This  parScular  paSent  has  low  T2*  in  the  liver,  a  sign   of  iron  overload  due  to  β-­‐thalassemia.   Introduc/on   Objec/ve   Methodology   0 5 10 15 20 0 1 2 3 4 5 6 7 8 Echo Time (ms) SNR Observed Truth Fitted 0 5 10 15 20 0 1 2 3 4 5 6 7 Echo Time (ms) SNR Observed Truth Fitted Low  SNR   Low  T2*   y  =  1.0446x  -­‐  0.0004   R²  =  0.99959   0   1   2   3   0   1   2   3   Es/mated  SD  (ms)   Measured  SD  (ms)   T2* meas. SD meas. SD est. time (ms) RF 1.6 3.9 6.2 8.5 10.8 13.2 15.5 17.8 Mul -echo phantom images Observed Non-Linear Regression Estimated TE y TE y' TE y-y' Residuals Robust Estimation of variance med(abs(residuals)) Compute Covariance Matrix Results   Good  Breath-­‐hold  Poor  Breath-­‐hold   T2*  map   SD  map   T2*  map  w/  mask       •  SD  esSmaSon  efficiently  and  accurately  provides  a   way  to  analyze  T2*  precision   •  Useful  for  prototyping  new  T2*  esSmaSon   methods   •  PotenSally  useful  for  clinical  use  and  re-­‐ determining  thresholds  for  disease  management   Conclusions   References   1Department of Electrical Engineering, University of Southern California, Los Angeles, CA 2Laboratory of Cardiac Energetics, National Heart, Lung, and Blood Institute, Bethesda, MD