06/10/2009
Ir. Janaki Raman RangarajanPromoter: Prof. Dr. Ir. Frederik Maes
Quantitative analysis of multi-temporal and multi-modal in-vivo images
in small animal models
•
18.12.2008
Janaki Raman R
Slide 2
Overview
• Introduction– Quantitative Image analysis– Small animal models
• Image analysis pipeline:– Multi temporal, multi
modal registration & segmentation methods
– Quantification tools
• Results & Discussion– Applications
• Conclusion– Future work
Image Acquisition (IAq)
Image Pre-processing (IPp)
Image Registration (IRg)
Image Quantification (IQn)
•
30.09.2008
Janaki Raman R
Slide 4
Quantitative image analysis
• Image quantification - From ‘seeing’ to ‘measuring’
Requires object delineation or “image segmentation”
5w PI 8w PI
14w PI 30w PI
•
30.09.2008
Janaki Raman R
Slide 5
Fusion of complimentary information
Requires spatial alignment or “image registration”
5w PI
30w PI
Multi-temporal: Follow up over time
Multi-modal: Anatomical & functional info
•
30.09.2008
Janaki Raman R
Slide 6
In vivo Small Animal Image Analysis
• Image artifacts– RF in homogeneity
• Animal models– Transgenic/Wild-type– Rats - Wistor, SD..– Mice - C57BL6J,
ob/ob, ..
• Images– Group: 10– Sequence : 2– Time point: 5 + 1(exvivo)– Images to analyze: 60 x 2
•
30.09.2008
Janaki Raman R
Slide 7
MoSAIC – QUANTIVIAM collaborations
•
06/10/2009
Janaki Raman R
Slide 9
Image analysis pipeline: ESAT/PSI
• Automated (semi) methods– Segmentation, registration, quantification of small animal images
• Multi-temporal & multi-modal– µMR-µMR, µMR-µMRTemplate, µMR-µPET, …..
• Applications– MRI reporters, Morphological phenotyping, Planning acquisition
•
30.09.2008
Janaki Raman R
Slide 10
PhD Goal
• Develop (semi) automated methods– Segmentation– Registration– Quantification in small animals
• Multi-temporal & multi-modal– µMR-µMR, µMR-µPET, µMR-µCT…..
• Registration– Rigid, affine, non-rigid registration
(mesh..)• Segmentation
– Atlas/Template based segmentation
• Application driven– MR reporters, Fat quantification,
morphological phenotyping, ……
Year 1: Build pipeline, evaluate existing methods
Year 2: Multi-temporal to Multi-modality Rigid to Non-rigid registration
Year 3:Multi-modality extension, Applications
Year 4:Validation, applications, report
•
06/10/2009
Janaki Raman R
Slide 11
Micro-MRI acquisition:
Bruker
conversion:bruker-dicom-
analyze
Bias field correction
Segmentation:Brain mask
Motion correction:
bt. repetitions
Normalization:Follow up to
base line
Co-registration:template / atlas
Atlas based segmentation
Image Analysis Pipeline
Image Acquisition (IAq)
Image Pre-processing (IPp)
Image Registration (IRg)
Image Quantification (IQn)
Intensity normalization
Segmentation:VOI‘s
Quantification: MR constrast
TOPIM, Jan 2009
•
06/10/2009
Janaki Raman R
Slide 12
Micro-MRI acquisition:Bruker
9.4T
conversion:bruker-dicom-analyze
Image Analysis Pipeline - Acquisition
Image Acquisition (IAq)
– Resolution ~156 µm or 32 µm (ex-vivo)
– FOV [3.0x5.0x1.5] cm
@ µNMR lab MoSAIC, KUL
•
06/10/2009
Janaki Raman R
Slide 13
Micro-MRI acquisition:
Bruker
conversion:bruker-dicom-
analyze
Bias field correction
Segmentation:Brain mask
Motion correction:
bt. repetitions
Normalization:Follow up to
base line
Co-registration:template / atlas
Atlas based segmentation
Image Analysis Pipeline
Image Acquisition (IAq)
Image Pre-processing (IPp)
Image Registration (IRg)
Image Quantification (IQn)
Intensity normalization
Segmentation:VOI‘s
Quantification: MR constrast
TOPIM, Jan 2009
•
06/10/2009
Janaki Raman R
Slide 14
Bias field correction
Segmentation:Brain mask
Motion correction:
bt. repetitions
Image Analysis PipelineImage Pre-processing
(IPp)
Raw data
Motion corrected,averaged,N=12
• RF field in-homogeneity– Source: RF coil /static field in-homogeneity, patient anatomy or
position
– Effect : Intensity variations of same tissue type
• Derails - segmentation, registration or quantification
• MR Inhomogeneity correction
06/10/2009
Janaki Raman R
Slide 15
Original Corrected(masked)
Bias field
OriginalCorrected
Multiplicative model,3D 4th order polynomial,entropy minimization,mean preserving(Likar et al.)
Intensity distribution
=> Background separation
Speed up from polynomial order 1 to 4
Corrected
Bias field correction
Segmentation:Brain mask
Motion correction
bt. repetitions
Image Pre-processing
•
06/10/2009
Janaki Raman R
Slide 16
Multi-temporal study – ROI delineation
Baseline
Follow-up
ROI manually delineated in
baseline/atlas scan and automatically
propagated to co-registered follow-up
scans
allows voxel-wise analysis
MIRIT - affineMaximization of Mutual Information (Collignon A, Maes F, et. al)
•
06/10/2009
Janaki Raman R
Slide 17
Alignment:Follow up to
base line
Normalization:template / atlas
Atlas based segmentation
Image registration – Multi-temporal
TOPIM, Jan 2009
Image Registration
Before registration After registration
Baseline
Follow-upImage difference:- different animal position- anatomical difference
5w PI
8w PI
30w PI
Maximization of Mutual Information (Collignon A, Maes F, et. al)- Mutual information of corresponding voxel pairs is maximal if the images are geometrically aligned. -12 parameter(translation/rotation/shear/scaling) affine transformation
3w PI
• Spatial alignment – multi-temporal
06/10/2009
Janaki Raman R
Slide 18
5 months
MI based registration (affine): µMRI-µMRI (after bias correction)
Alignment:Follow up to base line
Normalization:template / atlas
Atlas based segmentation
Image Registration
Rat brain - Sprague Dawley
2 months
•
02/06/2009
Janaki Raman R
Slide 19
Micro-MRI acquisition:
Bruker
conversion:bruker-analyze
Bias field correction
Multi-modality registration: µPET- µMR
Image Analysis Pipeline
Image Acquisition (IAq)
Image Pre-processing (IPp)
Image Registration (IRg)
Image Quantification (IQn)
Coregistration/Normalization
MR as prior:PET reco.
(J Nuyts, K.U.L)
Motion correction:
bt. repetitions
Segmentation:Brain mask
QuantificationµPET- µMR
Intensity normalization
(MR)
Segmentation:VOI‘s
• Multi-modal registration
06/10/2009
Janaki Raman R
Slide 20
10 months
2 months
10 months
2 months
Without bias field correction & mask
With bias field correction & mask
J. Nuyts, A. Atre, K.Vunckx Nuclear medicine, KUL
ML reconstruction + MRI (prior) -> Bowscher Reconstruction
• Spatial normalization to template
06/10/2009
Janaki Raman R
Slide 21
Alignment:Follow up to base line
Normalization:template / atlas
Atlas based segmentation
Image Registration
Study
Atlas
MI based registration (affine): µMRI-µMRT (after bias correction)
Atlas
AtlasP. Schweinhardt 2003, rat brain templateMBIRN, MDA 2006, mouse brain template
• Atlas based segmentation
04/01/2010
Janaki Raman R
Slide 22
MBIRN data base – C57BL6 MDA
Control injection
Reference region
Test injection
Study
Atlas
TOPIM, Jan 2009
• Inter-scan intensity normalization
06/10/2009
Janaki Raman R
Slide 23
Control injection Test injection
Reference region
•
02/06/2009
Janaki Raman R
Slide 24
Micro-MRI acquisition:
Bruker
conversion:bruker-analyze
Bias field correction
Multi-modality registration: µPET- µMR
Image Analysis Pipeline
Image Acquisition (IAq)
Image Pre-processing (IPp)
Image Registration (IRg)
Image Quantification (IQn)
Coregistration/Normalization
MR as prior:PET reco.
(J Nuyts, K.U.L)
Motion correction:
bt. repetitions
Segmentation:Brain mask
QuantificationµPET- µMR
Intensity normalization
(MR)
Segmentation:VOI‘s
•
06/10/2009
Janaki Raman R
Slide 25
Case Study1: Imaging Neurogenesis
– Viral vector based genetic labeling
• Luciferase (BLI) vs Ferritin (MRI)
G Vande Velde, A. Ibrahimi, V. Baekelandt, Z Debyser
5wPI
Division of Molecular MedicineDivision of Molecular Medicine
•
06.10.2009
Janaki Raman R
Slide 26
Viral vector based MRI reporter genesFerritin Study
E.No. Vector Left/control Right
8 LV eGFP eGFP+Ferritin
11 LV eGFP FerrH-I-FerrL eGFP
PBSLentiviral Vector(LV)
Adeno associated Vector(AAV)
NOD-SCID
12 LV eGFP Ferritin
13 LV PBS eGFP
14 AAV PBS Ferritin
15 AAV PBS eGFP-T2A-fLuc
16 AAV eGFP-T2A-fLuc FerrH-T2A-fLuc
17 AAV PBS Medium- OMEM
18 LV eGFP-T2A-fLuc FerrH-T2A-fLuc
19 AAV eGFP FerrH-T2A-fLuc
20 AAV eGFP FerrH-T2A-fLuc
ISMRM, Hawaai Apr 2009; WMIC Sep 2009.
• Image analysis of MRI reporters
• Conventional quantification– Qualitative visual examination– Manual delineation of ROI’s – Parametric maps
• Disadvantages– Manual analysis is tedious and error-prone (user,
artifacts)– Low resolution of parametric maps & hypo-intense
contrast of MRI reporters
06/10/2009
Janaki Raman R
Slide 27
5w PI 8w PI 14w PI 30w PI
• MRI reporter - Pre-processing step
06/10/2009
Janaki Raman R
Slide 28
Bias field correction
Segmentation:Brain mask
Source Initial mask Bias field Bias corrected MRI Final mask Brain mask
C57bL6/J black mice exp 15: AAV06c
• MRI reporter - Registration step
06/10/2009
Janaki Raman R
Slide 29
MI based Image alignment & normalization ofC57bL6/J black mice with MBIRN atlas exp 15: AAV06a vs. AAV 06c
• MRI reporter : Quantification step
• Visualization of segmented hypo-intense MR contrast (Ferritin) in axial & coronal planes
06/10/2009
Janaki Raman R
Slide 30
Visualization of segmented hypo-intense MR contrast (Ferritin) in Paxinos reference frame
• Results : LV immune response
06/10/2009
Janaki Raman R
Slide 31
World Molecular Imaging Conference, Sep 2009
• Results : LV immune response
06/10/2009
Janaki Raman R
Slide 32
World Molecular Imaging Conference, Sep 2009
• Results: LV vs AAV background
06/10/2009
Janaki Raman R
Slide 33
World Molecular Imaging Conference, Sep 2009
• Results: AAV immune response
06/10/2009
Janaki Raman R
Slide 34
World Molecular Imaging Conference, Sep 2009
• Conclusions of MRI reporter study
• LV vector contributes significantly to background contrast
• Backgound contrast challenges SNR of potential vector based MR reporters(e.g. Ferritin)
• AAV vector results in very low background contrast in comparison to LV
AAV is promising for other potential MR reporter genes!
• Publications- Ferritin case study– G. Vande Velde, J.R. Rangarajan, T. Dresselaers et. al Evaluation of lentiviral and adeno-associated viral vector systems for ferritin
expression as MRI reporter gene in mouse brain. Journal of NeuroImaging (in preparation)
– G. Vande Velde, J.R. Rangarajan, T. Dresselaers, J. Toelen, Z. Debyser, V. Baekelandt, U. Himmelreich, Quantification of 3D T2*-weighted MR images allows evaluation of different viral vectors for stable MR reporter gene expression in the rodent brain , ISMRM - ESMRMB joint annual eeting, May 1-7, 2010, Stockholm, Sweden (accepted)
– G. Vande Velde, J.R. Rangarajan, T. Dresselaers, A. Ibrahimi, Z. Debyser, V. Baekelandt, U. Himmelreich, Comparison of lentiviral and adeno-associated viral vectors for stable MRI reporter gene expression in the rodent brain, 2009 world molecular imaging congress Sep. 2009, Montréal, Canada
– G. Vande Velde, J.R. Rangarajan, T. Dresselaers, O. Krylyshkina, A. Ibrahimi, Z. Debyser, V. Baekelandt, U. Himmelreich, Evaluation of LV and AAV vector systems for stable delivery of MRI reporter genes to the rodent brain, ISMRM April 2009, Honolulu, Hawaii
– J.R. Rangarajan, G. Vande Velde, U. Himmelreich, T. Dresselaers, C. Casteels, A. Atre, D. Loeckx, J. Nuyts, F. Maes, An image analysis pipeline for quantitative analysis of multi-temporal and multi-modal in vivo small animal images , TOPIM 2009 - dual and innovative imaging modalities, January 26-30, 2009, Les Houches, France
– “Quantitative multi-temporal image analysis of MRI reporters in rodent brain” (in preparation)
06/10/2009
Janaki Raman R
Slide 35
• Longitudinal MPIO quantification
06/10/2009
Janaki Raman R
Slide 37
NeuroImage 2009
• Pilot study – MPIO vs. reporter gene
– MPIO in RMS using the pipeline, on best known protocol– MPIO quantification both within in vivo and ex vivo – Compare MPIO vs. Vector based MRI reporter – In progress :
• Pre-processing (done); Registration & Quantification (ongoing);
06/10/2009
Janaki Raman R
Slide 38
• MPIO @Rostral Migratory System
06/10/2009
Janaki Raman R
Slide 39
Registration to MBIRN MDA atlas, overlayed with labels
• Atlas based segmentation of RMS
06/10/2009
Janaki Raman R
Slide 40
LabelOutline
LabelMask
•
06/10/2009
Janaki Raman R
Slide 41
Case study 2: Multimodal imaging in HR
• Morphological phentoyping– Transgenic (HD)– Wild type (control)
• Multi-temporal quantification– 2M, 5M, 10M, 18M…– Morphological changes
• Within & between phenotypes
• Multi-modal study– µMR-µPET (rigid)– µMR-µMR template (affine,
non-rigid)
Wild type (control) Transgenic (HD)
2 M 5 M 10 M
WT
Tg
µMRI : RARE 3D, T2
µPET : FDG, CB1
HD
von Horsten, S. et al. Hum. Mol. Genet. 2003
18 M
C Casteels, J Nuyts, K Van Laere, Nuclear Medicine, KU Leuven
• Image analysis Huntington rats
06/10/2009
Janaki Raman R
Slide 42
10 months
2 months
10 months
2 months
Casteels C 2006
Casteels C
•
06/10/2009
Janaki Raman R
Slide 43
Multi-temporal, multi-modal registration
MMI based Registration
MR atlas in Paxinos space
A) RIGID(R1) B) AFFINE (R2) NON-RIGID (R3)
C) R1 (R2, R3)
PET MR
• Pre-processing step
06/10/2009
Janaki Raman R
Slide 44
2 months 5 months 10 months 18 months
2 months 5 months 10 months 18 months
No Bias correction
Bias corrected & brain mask
• Multi-temporal spatial alignment in HR
06/10/2009
Janaki Raman R
Slide 45
10 months
2 months
•
06/10/2009
Janaki Raman R
Slide 46
Multi-modal registration: PET-MR
MI based registration (rigid): µPET-µMR Influence of bias field correction and brain mask selection
µMRI µPET (no BFC) µPET (with BFC)
• Morphological phenotyping
06/10/2009
Janaki Raman R
Slide 47
18M
5M
Transgenic – Huntington rats
Wild Type – control group
18M
I > mean + 2*stddev (- not robust )- Good BFC & registration can help better segmentation
• Results
• Initial results– Successful PET- MRI registration– Benefits PET reconstruction– MRI - MRI temporal & template
registration– A good brain mask & bias field
correction(BFC) is important
• In progress– Improve BFC in cooperation with VisionLab– Building individual template/ time point–
affine/NRR– Robust segmentation of both striatum &
ventricles
06/10/2009
Janaki Raman R
Slide 49
• Future work: Image analysis
• Summary– The image analysis pipeline with image registration
framework facilitates quantification of multi-temporal & multi-modal study in small animal models
• Future work• Planning of stereotactic surgery
– Plan optimal trajectory.
• Variability in needle tracts– Possibility of missing anatomical ROI– Trajectory could hit vasculature=> immune response
• MR acquisition– Localize better acquisition plane during image acquisition of
time series images.
04/01/2010
Janaki Raman R
Slide 51
•
04/01/2010
Janaki Raman R
Slide 52
Neuro Anatomical Surgical Planning
• Functional neurosurgery– Anorexia (e.g. Septal
nucleus)
• Neuromodulation– Injections
– Electrical stimulation
– Lesions
• Bleeding– Burr holes (Visible)
– Other brain regions(?)
• Influences..– Systematic unwanted side
effectsSource: U. Himmelreich, KUL
Kris van Kuyck, Bart Nuttin. Lab. of experimental & functional neurosurgery, KUL
• Planning optimal trajectory using MRA
06/10/2009
Janaki Raman R
Slide 53
• Register/Overlay– Anatomical MRI with MRA
– Anatomical MRI with Paxinos Atlas
– Anatomical variability
– Probabilistic atlas of vasculature in PaxinosKris van Kuyck, Bart Nuttin. Lab. of experimental & functional neurosurgery, KUL
•
04/01/2010
Janaki Raman R
Slide 54
Neuronal connectivity @Neuromodulation
1. Optimal trajectory in Paxinos space
• Lesions (e.g. septal nucleus)
2. ROI based image plane positioning
3. Tractography along Neuromodulation regions
Kris van Kuyck, Bart Nuttin. Lab. of experimental & functional neurosurgery, KUL
1. Trajectory: Lesion/Stimulation Electrode
2. ROI: Lesion
3. Fiber tract running along ROI
Investigate
• Registration (ESAT)
• Atlas construction (ESAT)
• Quantification of brain connnectivity(VLAB/ESAT)
•
04/01/2010
Janaki Raman R
Slide 55
3 mm
2 mm
Lateral 1.6 mm Lateral 1.6 mm
1 1
2 2
Localization: SOI in paxinos2MR
SOI – Site of injection
• Planned vs actual trajectory
04/01/2010
Janaki Raman R
Slide 56
• Asymmetric needle tracts
04/01/2010
Janaki Raman R
Slide 57
• Variability in Injection trajectory
• Localization of injection axis in 3D– Principal component analysis on voxel coordinates(P)
– First Eigen vector(A) => axis direction– Visualize within & across group
• Variability of tip of axis, height, radius
04/01/2010
Janaki Raman R
Slide 58
C+max[P.A]
H
C+min[P.A]
C
• Next 3-6 months: Prospective registration
04/01/2010
Janaki Raman R
Slide 59
• Need: Image plane positioning– Localization of region of interests– Define image plane using baseline/atlas as reference
Tripilot 2D 3D MSME DWI
Tripilot 2D 3D MSME DWI
5w PI
8w PI
•
06/10/2009
Janaki Raman R
Slide 60
Acknowledgment
• KU Leuven– Prof. Dr. Ir. F. Maes, Prof. Dr. Ir. P. Suetens– Prof. Dr. U. Himmelreich, Dr. Ir. T. Dresselaers– Prof. Dr. V. Baekelandt, G. Vande Velde– Prof. Dr. Z. Debyser, Dr. A Ibrahimi– Prof. Dr. Ir. J. Nuyts, A. Atre– Prof. Dr. K. Van Laere, C. Casteels
• University of Antwerp– Prof. Dr. A. Van der Linden, Ruth Vreys, Dr. M. Verhoye,
Ir. J. Van Audekerke– Prof. Dr. Ir. J. Sijbers, M. Zhenhua