Download DSC in brain tumor imaging

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Towards DSC-based CAD applications
in brain tumor imaging
…or…
Get your hands off !
Kyrre E. Emblem, PhD
The Interventional Centre, Rikshospitalet, Oslo University Hospital, Oslo, Norway
A.A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital
and Harvard Medical School, Boston, MA, USA.
[email protected]
Oslo University Hospital
http://www.nmr.mgh.harvard.edu
Glostrup 01/24/2012
Overview – hands off !
• DSC in brain tumor imaging
• Some pitfalls
• An example …
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Pre-surgical glioma characterization
Ø Grading
Ø Survival predictions
Ø CBV / CBF / MTT / Vessel Size
”… As many as 75% of anaplastic astrocytomas fail to enhance on conventional MR images
after administration of a gadolinium-based contrast material”
AJNR 2004;25:214-221
Radiology 2002;224(3):797-803
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
‘… Despite this lengthy history of use, perfusion MR imaging
remains a boutique method in many neuroimaging practices,
used only occasionally. ‘
Greg Sorensen, Radiology 2008
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Pre-surgical glioma characterization - some hurdles…
?
• High degree of user dependency
• What is “true” tumor tissue?
?
• Nearby vessels
• What is correct reference tissue?
• Image artifacts / post-processing
?
• Some tumors can show foci of high CBV
irrespective of tumor grade (* oligodendroglial tumors)
?
• Results (threshold values) are difficult to generalize
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Blood volume heterogeneity
Ø Whole tumor histogram analysis
Ø Improved reproducibility
Normal
cortex
Early stage
astrocytoma
Astrocytoma
grade II / III
Glioblastoma
‘Histogram’ - Almost perfect (К = .81-1.00)
‘Hot-spot’ - Moderate agreement (К = .41-.60)
Frontiers in Bioscience 2004;9(1):3105-23
0,14
0,12
Relative frequency
Pilocytic Astrocytoma, grade I (Fig 2.A)
0,10
Diffuse Astrocytoma, grade II (Fig 2.B)
0,08
Anaplastic Astrocytoma, grade III (Fig 2.C)
Glioblastoma, grade IV (Fig 2.D)
0,06
0,04
0,02
0,00
0,00
1,76
3,61
5,46
7,32
9,17
11,02
12,87
nCBV
JMRI 2007;26(4):1053-63
Radiology 2008;247(3):808-17
Oslo University Hospital
AJNR 2007;28(4):761-6
AJNR 2010;31(9):1699-706
Glostrup 01/24/2012
DSC in brain tumor imaging
• Tumor segmentation
Manual
Automatic
MRI acquisition
cleaning
(skull strip, normalization,…)
Oslo University Hospital
segmentation / clustering knowledge-based morphology
(supervised, unsupervised)
Glostrup 01/24/2012
DSC in brain tumor imaging
• Macroscopic vessels
MRM 2001;45(3):477-85
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Macroscopic vessels
DSC-based vessel segmentation
Magn Reson Med 2006;55(3):524-31
Magn Reson Med 2009;61(5):1210-7
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• CBV normalization
AJNR 2009;30(10):1929-32
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Why is standardization so difficult?
- Contrast agent type
- Contrast agent dosage
- Imaging sequence
- Imaging parameters
- MR machine vendor
- Post-processing
Type II
Type II
Type I
Cut-off
Type I
Ideal
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Why is standardization so difficult?
- Contrast agent type
- Contrast agent dosage
- Imaging sequence
- Imaging parameters
- MR machine vendor
- Post-processing
*
*Support Vector Machines
Machine Learning
- Can handle large and complex datasets
- Fast
- Robust
- Parameter flexibility (generalization)
- ‘Self-learning’
Magn Reson Med 2008;60(4):945-52
Magn Reson Med 2010;64(4):1230-6
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Glioma characterization – it’s all about automatization…
Ø Goal?
An end-to-end automatic,
100%-accuracy,
multi-parametric,
multi-modality,
multi-institutional,
“self-learning”,
PACS-integrated,
bug-free(!)
Computer Aided Diagnostic (CAD) tool…?
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging
• Glioma characterization – it’s all about automatization…
Ø Goal?
PACS report
An end-to-end automatic,
100%-accuracy,
multi-parametric,
multi-modality,
multi-institutional,
“self-learning”,
PACS-integrated,
bug-free(!)
Computer Aided Diagnostic (CAD) tool…?
• Most likely tumor grade
• Survival rate for patients with similar data
• Change from last visit
• Goodness of fit
Oslo University Hospital
Glostrup 01/24/2012
Overview – hands off !
• DSC in brain tumor imaging
• Some pitfalls
• An example …
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
• Oligodendroglial subtypes
WHO Grade II Oligodendroglioma
WITH
LOH on chromosome 1p/19q
Chicken wire
’classic’ histopathologic features
(~90%)
Neuroradiology 2006;48(10): 703-13
Magn Reson Imaging 2007;25(6):1113-9
Oslo University Hospital
J Neurosurg 2007;107(3):600-9
AJNR Am J Neuroradiol 2008;29(9):1664-70
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
• Oligodendroglial subtypes
WHO Grade II Oligodendroglioma
WITHOUT
LOH on chromosome 1p/19q
Chicken wire
Neuroradiology 2006;48(10): 703-13
Magn Reson Imaging 2007;25(6):1113-9
Oslo University Hospital
J Neurosurg 2007;107(3):600-9
AJNR Am J Neuroradiol 2008;29(9):1664-70
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
• Oligodendroglial subtypes
WHO Grade II Oligodendroglioma
WITHOUT
LOH on chromosome 1p/19q
PPV =
NPV =
100%
90%
Neuroradiology 2006;48(10): 703-13
Magn Reson Imaging 2007;25(6):1113-9
Oslo University Hospital
J Neurosurg 2007;107(3):600-9
AJNR Am J Neuroradiol 2008;29(9):1664-70
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
•
Non-linear dose-response in blood
Ø
Blood - water diffusion, dynamic
dephasing, frequency shifts
Ø
Tissue - static intra-voxel dephasing
Magn Reson Med 2002;47:461–71
JMRI 2005;22:693–96
…overestimation of perfusion values
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
•
… corrections?
Ø
Calibration curves
Ø
Double-echo imaging (SE / GE)
Ø
AIF outside bulk blood
Magn Reson Med 2003;49:1067-76
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
•
Contrast agent extravasation
Ø
Blood-Brain Barrier (BBB) breakdown
Ø
DSC values can be
underestimated (T1 dominant) - or overestimated (T2* dominant)
1/T1
leakage
No leakage
Plasma [CA]
Oslo University Hospital
Pre-contrast T1-weighted
Post-contrast T1-weighted
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
•
… corrections?
Ø
Sequence optimization (multi-echo, low
flip angle – minimize T1 effects)
Ø
Pre-bolus (T1-saturation)
Ø
Post-processing schemes (gamma fitting,
correction algorithms)
R2*
T2*-effect
time
T1-effect
Magn Reson Med 2004;51:961-68
Oslo University Hospital
JCBFM 2011;31(10):2041-53
AJNR 2006;27(4):859-67
Glostrup 01/24/2012
DSC in brain tumor imaging - pitfalls!
•
…bottom line?
DSC estimates in brain tumor MRI are ‘OK’ !
- just make sure to do the following:
1. Don’t trust quantitative values (in bulk blood) – relative values fairly OK
2. Select an AIF close-to, but outside bulk blood
3. Correct for contrast agent leakage
4. User-independent methods preferred + VALIDATION NEEDED!
Oslo University Hospital
Glostrup 01/24/2012
Overview – hands off !
• DSC in brain tumor imaging
• Some pitfalls
• An example …
(end-to-end automatic…)
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - anti-angiogenic therapy
• Vascular normalization
Vessels in normal tissue
Tumor Vessels
Nat Med 2001;7(9):987-9
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - anti-angiogenic therapy
• Vascular normalization
Vessels in normal tissue
Tumor Vessels
Nat Med 2001;7(9):987-9
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - anti-angiogenic therapy
Contrast-enhanced MRI
Vessel Permeability
Vessel Size
Blood Flow
Cancer Cell 2007;11(1):83-95
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - anti-angiogenic therapy
+1.96SD
-1.96SD
Cancer Res 2012;72(2):402-7
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - anti-angiogenic therapy
Median OS 348d versus 169d
(P<0.001)
Cancer Res 2012;72(2):402-7
Oslo University Hospital
Glostrup 01/24/2012
DSC in brain tumor imaging - anti-angiogenic therapy
PACS report
+1.96SD
-1.96SD
• Most likely tumor grade
• Survival rate for patients with similar data
• Change from last visit
• Goodness of fit
Oslo University Hospital
Glostrup 01/24/2012
Alternative use of Neuro Perfusion?
•
Caffeine (abuse?)
≈
Oslo University Hospital
Glostrup 01/24/2012
Related documents