1. Perfusion Imaging S. Lalith Talagala, Ph. D
Perfusion Imaging S. Lalith Talagala, Ph.D. NIH MRI Research Facility National Institute for Neurological Disorders and Stroke National Institutes of Health, Bethesda, MD
First, I want to thank the organizers for inviting me to give this talk.
And I appreciate the opp to talk to you today.
In this talk I plan to give an overview of the two perfusion MRI techniques available for clinical studies.

2. Perfusion Imaging: Outline
Introduction
Dynamic Susceptibility Contrast (DSC)
Method
Quantification
Examples
Arterial Spin Labeling (ASL)
Labeling Techniques
So the talk outline quite simply the following:
We will discuss each one of two techniques :DSC and ASL
Talk about
how they work,
how signal changes are converted to perfusion estimates
Along the way I will show some example images from each method

3. Definitions Perfusion – capillary blood flow delivered to the tissue
MRI methods can assess
blood flow – ml blood / min / 100 g of tissue -
blood volume – ml blood /100 g of tissue
mean transit time – seconds
Normal values for brain
To start off lets review the definitions:
Perfusion which refers to capillary blood flow deliv ered to tissue. We are interested in measuring this quantity on a voxel by voxel basis.
In addition to flow, MRI can measure blood volume and mean transi time of i.e. average time taken by blood to travel through the capillary bed.
We will concentrate on measurements in the brain… so these parameter will be referred to as cerebral blood flow/ volum CBF,CBV.
Gray Matter
White Matter
CBF (ml /min/100 g)
CBV (ml / 100 g)
4 - 6
2 - 4
MTT (s)
4 - 5
5 - 6

4. Perfusion MRI Dynamic Susceptibility Contrast (DSC)
Requires contrast injection
Large signal changes, Fast
Single application (clinical)
Arterial Spin Labeling (ASL)
No contrast required
Small signal change, Slow
Multiple measurements (clinical and research)
S there are now two major approaches for Clinical perfusion MRI., DSC and ASL
Here I have listed their major differences.
DSC needs contrast, ….
ASL no contrast ..

5. Dynamic Susceptibility Contrast (DSC)
Monitor passage of Gadolinium contrast through tissue using rapid T2*/T2 weighted MRI
Gradient Echo EPI
TR ~ s TE = ms ~1.5 min
Imaging
Lets look at the DSC technique:
This method is based on monitoring the passage of a paramagnetic contrast agent through the tissue using T2/T2* weighted MRI.- Here we detect the change in water MRI signal due to contrast
So we inject contrast as a bolus and then acquire images through the brain using a fast imaging tech such as echo planar imaging
Gd mmol/kg
Bolus Injection

6. DSC - Mechanism Gd Chelates: Paramagnetic, Intravascular W/ Gd W/O Gd
Tissue/ Blood Dc
Field Inhomogeneity
R2* (=1/T2*)
GE/SE MRI Signal
So what happens when Gd goes through the tissue:?
First it is paramag, and, under normal conditions, stays with in the blood vessels. ..

7. DSC – Passage of Gd through tissue
Here is a series of images acquired following bolus injection of contrast. This particular data set is from a patient with a tumor on the right side.
You are looking at the same slice acquired at different times.
We see high baseline signal before the contrast arrives in the slice,
the signal is decreased when the contrast washes in
and the signal is restored to when the contrast washes out of the slice
1.5 T, 0.1 mmol/kg, GE- EPI, TE = 50 ms, TR = 2 s

8. DSC – Signal vs Time GM pixel Blood pixel
The signal change due to washin and wash out constrat is better seen by looking at the signal timecourse of indiviual pixels.
Here we see the timecourses of a pixels corresponding to a blood vessel and a gray matter
We can appreciate a large of signal loss during the first pass of the contrast agent and a smaller signal loss later due to recirculation of the agent.
Signal loss is greater in blood pixels than in tissue pixels.
Baseline First pass Recirculation

9. DSC: Signal loss to Concentration
k – proportionality constant
C – Gd concentration
Sc – Signal with Gd
S0 – Baseline signal without Gd
So we can see that the signal change depends on concentration, but how are they related?
Early studies have shown that the change in T2 relax rate is linearly proportional to concentration…so we use that.
We can also write that signal in the presence of Gd in terms of the baseline signal and delta R2*
Combining the two allows us to estimate the Gd concentration via the log of the ratio of signal with and without contrast.
Note that k is not known for each subject.

10. DSC: Concentration vs time
So we can pixel by pixel convert signal timecourses to conc timecourses.
We can see the blood pixel show higher concentration than the GM pixel.

11. Arterial input function (AIF)
DSC – CBV, CBF, MTT ?
F (ml/min)
Cart (t)
Ctis (t)
Tracer Kinetic Theory
Arterial input function (AIF)
Tissue Response
So here is what we have so far…
We know the Gd concentration timecouse in the arteries—this refered to as the arterial input function.
Then we know the corrs tissue conc timecourse i.e the tissue response.
We would like to use this information to estimate the flow, volume and transit time of blood through the voxel.
We can make use of the well developed tracer kinetic theory to estimate them.
From the central volume theorem, these parameters are related as shown here.

12. DSC- Input/Residue/Output Curves
Cart (t)
Cvein (t)
Ctis (t) F 1
R(t)
Residue function
h(t)
Impulse
Ä F×Cart(t)
Ä F×Cart(t)
Cart(t)
Typical
For estimation of flow and mean transit time we need to consider what the tissue response would be to an impulse input.
If the input is instantaneous, all the tracer arrives in the tissue voxel at the same time so that the tissue concentration is maximum at the beginning and decays with time
Tissue response to an impulse input is referred to as a residue function
-- this function is related to the mean transit time.
--need to find a way determine the residue func.
What we can actually measure is typical input which is wider and the corresponding tissue response.
We know that response to any input is the input broadened by the residue fn. i.e. convolution of input with response to an impulse.
Note also that the tissue response is scaled by the flow to voxel.
h”(t)
R”(t)
INPUT
RESIDUE
OUTPUT

13. Tracer Kinetics – Basic Equations
R(t)
h(t) 1
Frequency function
Residue function
For estimation of flow and mean transit time we need to consider what the tissue response would be to an impulse input.
If the input is instantaneous, all the tracer arrives in the tissue voxel at the same time so that the tissue concentration is maximum at the beginning and decays with time
Tissue response to an impulse input is referred to as a residue function
-- this function is related to the mean transit time.
--need to find a way determine the residue func.
What we can actually measure is typical input which is wider and the corresponding tissue response.
We know that response to any input is the input broadened by the residue fn. i.e. convolution of input with response to an impulse.
Note also that the tissue response is scaled by the flow to voxel.

14. DSC- Calculation of CBV
Ctis (t)
Cart (t)
Estimatition of blood volume is straight forward.
CBV is proportional to the ratio of the areas of the tissue timecourse and the blood vessel as I have indicated here.
To get accurate ratio need to correct for recirculation effect and account for the differences in hematocrit in large arteries and capillaries.

15. DSC- Calculation of CBF and MTT
Cart (t)
Cvein (t)
Ctis (t) F
CBF
Rscl(t)
So in terms of concentrations we have the relationship
--- the tissue timecourse is the convolution between the arterial input and the residue funnction scaled by CBF
To find the residue function , we need to decon the arterial input from the tissue curve.
-- And you get a scaled version of the residue function.
MTT is the normalized area of scaled residue function
CBF is the scaling factor or initial height of the scaled residue.
Deconvolution of the arterial curve from the tissue curve allows us to determine CBF and MTT.

16. DSC – CBV, CBF, MTT maps CBV CBF MTT
If we do this calculation on a pixel by pixel basis on the data I showed before we can get images like this.
Here we elevated CBV and CBF and shorter MTT in the right hemisphere containing the tumor compared.
to the relatively normal left side has lower CBV/CBF and longer MTT.
These maps clearly shows the extended abnormal area in this patient.

17. DSC – CBV, CBF, MTT maps CBV CBF MTT
Here is data from multiple slices.
Combination of CBF and MTT maps give good idea of the extent of abnormal area of in this patient
MTT maps in some cases show clearly the boundary between normal and abnormal tissue.

18. DSC – CBF maps 1.5 T, 0.1 mmol/kg, GE- EPI, TE = 50 ms, TR = 2 s
Here some calculated CBF maps from a different patient.
In this we can see good GM/WM contrast on both hemispheres.
You can also see high signal in the blood vessels which can make it difficult to see abnormalities near by.
1.5 T, 0.1 mmol/kg, GE- EPI, TE = 50 ms, TR = 2 s

19. DSC: Quantification Issues
Accuracy of DR2* Û C relationship
Arteries (quadratic) and tissue (linear)
Arterial input function (AIF) determination
Partial volume , vessel orientation effects
Truncation of the peak
Dispersion between measurement site and tissue (local AIF)
Deconvolution errors
Sensitivity to noise
Sensitivity to bolus arrival times
Absolute CBF/CBV require use of scaling factors determined separately
Although we do have a good foundation for quantification of DSC data, in practice number of difficulties.

20. Arterial Spin Labeling (ASL)
Measure the change in MRI signal due to magnetic labeling (tagging) of inflowing blood
Tag
Control
Label
=>
Perfusion
Maps
Now lets look at the ASL technique.
ASL tech is based on measuring the signal change due to magnetic labeling of inflowing blood….labeling/tagging generally refers to inversion of blood magnetization
Subtraction of images acquired without and with labeling allows us to see the signal change due to flow
The ASL expt consists of time period for labeling, delay for labeled blood to flow into the tissuse and an imaging period.
Imaging can be done with any rapid imaging method , use of EPI is common.
Here, unlike in DSC, we can use min TE bec we want to changes in longituditunal magnetization .
GE/SE EPI
TR ~ s
TE = minimum
~ 4-5 minutes
LABELING
IMAGING
0.5 – 2.5 s s
0.75 s

21. ASL: Control/Label/Difference
Difference (DM) images - S
Control
Images in Mydocs/data/ASLdiffExample
/008NIfTI/
1 pair
(10 sec)
24 pairs
(4 min)
DM: GM – 0.9 %, WM – 0.15 %
Label

22. ASL – One Compartment Kinetic Model
Cart (t)
Cvein (t)
Ctis (t) F
R(t)
h(t)
Impulse
Ä F×Cart(t)
Ä F×Cart(t)
Cont
ASL
R”(t)
h”(t)
Pulsed
ASL

23. ASL- Quantification of CBF
One compartment model: Labeled blood stays in the vasculature
CASL
PASL
R1a – relaxation rate of arterial blood
a0 – labeling efficiency
t – labeling time
w –post labeling delay
l – brain/blood partition coefficient of water

24. ASL: Labeling Strategies
Continuous ASL (CASL)
Narrow labeling plane
Long duration (seconds)
Input function – constant
Pulsed ASL (PASL)
Wide labeling slab
Created by a short pulse (milliseconds)
Input function – decaying exponential
(T1 of blood)
CASL
Labeling
Plane
PASL
ASL can be performed in two ways continuous and pulsed and they differ in how the inflowing blood is tagged.
In cont ASL , a spatially narrow inversion plane is defined with approp RF and Gradients and it is applied for a long time ..several seconds.
And in pulsed ASL, we invert blood (and tissue) with in a wide region proximal to the region being imaged. The inversion is accomplish in a v short time.
Labeling
Slab

25. ASL: Pulsed Labeling (QUIPSS II)
Proximal
Inversion
Proximal
Saturation
Image
RF/Signal …
Gradient Gi …
Label
Gi¹0
Label
Voxel
View

26. ASL: Pulsed Labeling (QUIPSS II)
Gi¹0
Control
Gi = 0
Voxel
View
Label
Control

27. ASL: Pulsed Labeling (Q2TIPS)
Control
Difference
Advantages:
High tagging efficiency
Low SAR
Ease of implementation
Disadvantage:
Limited coverage

28. ASL: Continuous Labeling (Flow-driven Adiabatic Fast Passage)
Image
Label
RF/Signal …
Control
Gradient Gl …
Label
Gl¹0
RF=0
Control
Voxel
View
Label
Control

29. Continuous ASL: Neck Labeling Coil
Plane
8 Ch Rx
Advantages:
Whole brain coverage
High labeling efficiency
Lower SAR
Neck Labeling Coil
Although these studies were encouraging, SNR of perfusion images were not sufficient for any higher resolution images. Availability of multi channel coils in the last few years helped to this problem.
Here we see the 8 or the 16 channel coils that we normally use for our studies.
Disadvantage:
Requires special hardware

30. CASL with a Neck Labeling Coil: Multi-shot 3D-FSE Spiral
Shows perfusion data with 2 coil method on the left is the difference signal (control-label) ..obtained using a 3d sequence spiral FSE with background suppression..resolution …. 6 min …Feature to note is the whole brain coverage..i.e good perfusion signal throughout the brain. Indicates that tha labeling coil is well balanced. Quantified by acquiring R1 maps…middle and CBF on the right.
% DS R1 map CBF
3T, Head Coil, 3D-FSE, 3.7 x 3.7 x 5 mm3
8 shots, TR 5.9 s, Label dur 4.1 s, PL delay 1.64 s, Backgr supp
6 min 22 sec
Talagala et al, MRM 52: (2004)

31. Continuous ASL: Neck Labeling Coil
3T, 8 Ch Rx, 2D EPI, 3 x 3 x 3 mm3, TE/TR 13 ms/5 s, 4.5 minutes

32. CASL with a Neck Labeling Coil: Hemangioblastomas
3T, 8 Ch Rx, 2D EPI, 1.5 x 1.5 x 3 mm3
TE/TR 16 ms/5 s, LD/PLD 3 s/1.6 s
10 minutes

33. CASL Perfusion MRI at 7 T Volume Tx Coil Surface Labeling Coil
Head Rx Array Coil
Tx Volume / 8 Ch Rx Array
Neck Labeling Coil

34. 7T CASL with a Neck Labeling Coil
Control (RF off) - Label (RF +ve offset)
Control (RF off) - Label (RF –ve offset)
7T, 8 Ch Rx, 2D EPI, 2 x 2 x 3 mm3
TE/TR 13 ms/5 s, ASSET X2, LD = 3 s, PLD 1.5 s
8 minutes
Talagala et al ISMRM 2008

35. CASL Perfusion MRI at 7 T %DS T1 CBF Gray Matter White Matter
ml/
(100g.
min) ms
%DS T1
CBF
Gray Matter
White Matter
Mean T1 (ms)
1940
1363
DS/S (%)
1.43 +/- 0.25
0.3 +/- 0.04
CBF (ml/min.100g)
76 +/- 11
27 +/- 2.4
7T, 8Ch Rx, 2.1 x 2.1 x 3 mm3, 9 minutes, n=5

36. ASL: Pseudo Continuous labeling
0 f 2f 3f nf
Image …
Label
RF/Signal …
f = g Gz t z
Control …
Gradient
Advantages:
Whole brain coverage
High labeling efficiency
Use standard hardware
Disadvantages:
Higher SAR
Labeling sensitive to off-resonance effects
2cm

37. Pseudo Continuous ASL: 3T data
3.6 x 3.6 x 5 mm3
Gradient-echo EPI
TE/TR = 20.8/500 ms
t/w = 2500/1700 ms
Scan time 5:00

38. Pseudo Continuous ASL: 7T data
2.3 x 2.3 x 3 mm3
Gradient-echo EPI
TE/TR = 20.8/5100 ms
t/w = 3000/1200 ms
SENSE 3x
Scan time 4:15
Luh et al, MRM 69:402 (2013)

39. CASL fMRI with a Neck Labeling Coil
3T, Head Coil
Finger movement (0.5 Hz),
{48 s Task / 48 Rest} X 6, 10 min
GE EPI, 3.75 x 3.75 x 5 mm3
12 s per Cont/Label pair
SPM, Spatial normalization smoothing (8 mm), N= 15
Perfusion based fMRI can be performed using this methood by using 2D EPI multi slice imaging.
Here we see group average data showing CBF changes in response to a finger tapping task. CBF increases are seen the cortical and subcortical regions and the cerebellum. We were able to identify decreases in CBF in the ipsilateral hemisphere.
Garraux et al, NeuroImage 25: (2005)

40. 3T CASL Perfusion fMRI with 16 Rx
CBF
75 ± 11
ml/(min.100g)
DCBF
78 ± 7%
3 x 3 x 3 mm3
CBF
92 ± 16
ml/(min.100g)
DCBF
102 ± 10%
1.5 x 1.5 x 3 mm3
Improved SNR allows Perfusion based fMRI studies also can be conducted at higher resolution
Here are examples at 3X3 and 1.5X 1.5 inplane resolution shows activation localized to gray matter.
The resting CBF and %change in CBF is about 20% higher on the average at high-resolution, which is most possibly due to reduced partial volume effect.
Finger movement (2 Hz), {40 s Rest / 40 Task} X 8, N = 6
GE EPI, TE 26 ms, 10 s per Control/Label pair, 10 min 40 sec

41. Functional Connectivity with ASL Perfusion
In addition to high res studies, increased sensitivity from multi channel coils make studies that are otherwise difficult more robust and reproducible.
CBF based resting state connectivity mapping shown here is one example.
On top you see the spectra of resting state time courses from the motor ROI. Spectra show expected low frequency BOLD oscillations. Because in the CBF timcourse, tagged and untagged images, CBF resting state oscillations are modulated at nyquist frequency and therefore appear at the high frequency region of the spectrum. This allows CBF fluctuations to be extracted without BOLD contamination.
CBF images can then be subjected to correlation analysis to generate functional connectivity maps as shown here.
Here we see bilateral CBF correlation in L/R motor areas across multiple slices.
We found that CBF fluctuation about ~24% of the baseline CBF in the resting state.
DCBF = 29 ± 19% DBOLD = 0.26 ± 0.14% (N=13)
3T, GE EPI, 16 Ch Rx, 3.75 X 3.75 X 3 mm3
TE/TR 12.5/3200 ms, 10 min 40 sec
Chuang et al., NeuroImage 40, 1595 (2008)

42. Functional Connectivity:
BOLD, Perfusion, CMRO2
Wu et al., NeuroImage 45, 694 (2009)

43. Perfusion MRI: Summary
Dynamic Susceptibility Contrast (DSC)
Requires contrast administration
Fast acquisition (< 2 min), whole brain coverage
Readily performed in clinical scanners
Online/Offline processing software available
Absolute quantification is difficult
Arterial Spin Labeling (ASL)
No contrast required
4-5 min acquisition, whole brain coverage
Absolute quantification is possible
Robust sequences becoming available
Useful for clinical and research work
S there are now two major approaches for Clinical perfusion MRI., DSC and ASL
Here I have listed their major differences.
DSC needs contrast, ….
ASL no contrast ..