Recent Advances in Cardiac Imaging Using HARP MRI

1
Recent Advances in Cardiac Imaging Using HARP MRI

• Jerry L. Prince
• Johns Hopkins University
• December 2001

2001 MMBIA
2
Acknowledgments

• David Bluemke
• Joao Lima
• Ernesto Castillo
• Dara Kraitchman
• Smita Sampath
• Bernard Gerber
• William Kerwin
• Sandeep Gupta
• Tony Faranesh
• Vijay Parthasarathy

• Nael Osman
• Jerome Garot
• Elias Zerhouni
• Elliot McVeigh
• Ergin Atalar
• Andy Derbyshire
• Tom Foo
• Carlos Rochitte
• Alan Heldman
• Li Pan

NIH/NHLBI, Whitaker Foundation, GEMS, Institut
Roche Cardiovasculaire, Federation Francaise de
Cadiologie
3
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

4
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

5
Magnetic Resonance Tagging

• Goal see motion within the myocardium
• Solution change tissue magnetization in fixed
  pattern watch pattern change.
• Zerhouni et al., 1987
• Axel et al., 1988

6
A Cardiac MRI Movie
7
A Tagged MR Movie
8
Goals in Tagged Cardiac MRI

• Technical aims
• dense motion measurement
• regional/local differentiation of strain
  differences
• develop clinically feasible protocols
• seamless integration with other MRI protocols

• Scientific and clinical aims
• Correlate motion abnormalities to disease
• Derive key indices of healthy or diseased
  myocardium
• Identify early markers of coronary artery disease

9
Conventional Tag Analysis

• Template Matching
• Guttman, et al., 94
• Atalar McVeigh, 94
• Active Geometry (snakes)
• Kumar Goldgof 94
• Young et al., 95
• Amini et al., 98
• Spline Fitting
• Young et al., 92
• Radeva et al., 97
• Kerwin Prince, 97
• Finite Element Modeling
• Young et al., 92

10
Critique of Standard MR Tagging

• Imaging time too long
• many breath-holds required
• Feature detection not robust
• manual interaction is required
• Tag lines are sparse
• interpolation is required
• Processing time too long
• clinically infeasible

11
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

12
What is HARP?

• Phase-based optical flow?
• Fleet and Jepson (1990)
• In many ways HARP is simpler than this
• Frequency or phase demodulation?
• Havlicek, Harding, and Bovik (2000)
• Standard demodulation methods are not adequate

• A new MR imaging method?
• HARP began as a tagged image processing method,
  but has evolved to something more
• Now it is a new way to image cardiac function

myocardial strain
13
TAGGING PROCESS
Kerwin and Prince, 1999
Tagging pulse sequence
Saturation pulse
Tagging is an amplitude modulation process
14
tagged images
t0ms
t390ms
k-space
15
Harmonic Expansion
Osman and Prince, 1999

• The tag pattern harmonics produce multiples of
  spectral peaks.

• The tagged image can be written as

Fourier transform

• Complex image corresponding to a single peak has
  sinusoidal tag pattern.

16
Computing a Harmonic Image
Fourier transform
Bandpass filter
Inverse Fourier Transform
Complex Harmonic Image
17
Motion and Image Model
reference map
p(x(y),t)
h2
image plane
x0
p
y (y1,y2)
h1
Where is y in 3-D?
18
Planar Tag Model
Tag frequency
Tag lines
Image Plane
19
Basic Harp Equations

• Tagged image
• Harmonic image
• Phase image

Motion causes phase modulation of the harmonic
image
20
Principle of HARP-MRI
Reference time sinusoidal tag pattern
Later time Tissue compression increases frequency
Sinusoidal tag pattern
Computed phase of tag pattern
Slope of phase increases Phase values track points
wrapping artifact
21
Harmonic Magnitude Image

• The harmonic magnitude image is a blurred
  MR image without the tag patterns.

• By simply thresholding the magnitude image, a
  segmentation mask is produced.

Mask
Magnitude
22
Harmonic Phase (HARP) Image

• The harmonic phase image can only be
  computed between and .
• The computed harmonic phase (HARP) angle image is

• The wrapping function is

23
2-D Motion

• Two tag orientations are required for 2-D motion.
• Produces vector-valued HARP angle images
• Relation to true harmonic phase pair

24
Principle of Point Tracking
Osman, Kerwin, McVeigh, Prince, 1999

• A point has two HARP angles
• Search in next image for same two values

Initial time
Later time
25
CINE HARP (CHARP)

• Objective find the position of at time
• Standard multivariate fixed point problem
• Solved efficiently by Newton-Raphson

gradients of HARP images
target angles
wrapping operator
Can track any point in image, forward or backward
in time
26
Trajectories of Selected Points
27
Geometry for Lagrangian Strain
Octants tracked to end-diastole
Octants placed at end-systole
t1
t20
D t19.5ms
28
Computing Lagrangian Strain

• Simple Lagrangian strain is change in length per
  unit length.
• Grid provides points for circumferential and
  radial strain for sub-endocardium, midwall, and
  sub-epicardium
• The strain between and is

29
Lagrangian Circ Strain Profiles
data courtesy of Elliot McVeigh, 1998
Stretching gt0
Shortening lt0
pacemech9
30
Computing Eulerian Strain

• Slope of the harmonic phase gives strain
• The local elongation in the direction of n
  is computed using

• Define matrix of tag frequencies

31
Eulerian Circumferential Strain

• red no change
• black shortening
• yellow stretching
• pacemaker shows abnormality

32
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

33
Tag Lines HARP vs. FINDTAGS
FINDTAGS black dots HARP (pi) white lines
Overall RMS 0.15 pixels
34
Tag Crossings HARP vs. FINDTAGS
35
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

36
Size/Shape/Style of Filter
?

• Overall shape determined by anatomy
• Motion spectrum is largely circular
• Small size reduces quality of result
• Tradeoff resolution versus interference

?
37
Location and Shape of a Harmonic Peak

• Location is defined as the center of mass
• Shape is defined using the covariance matrix
• The following can be shown

Is related to the DC peak
Is related to the motion
38
Filter Size for Minimal Error

• r is filter diameter as a fraction of wl
• How to choose r to minimize error for a given tag
  period
• Filter should be smaller when the noise is higher
• Filter should be smaller to avoid interference

39
Resolution vs. Filter Bandwidth

• Typical parameters
• FOV 320 mm
• Tag period 4.5 mm
• BPF 32 x 32
• Intrinsic Fourier resolution
• Strain resolution

Adequate to capture normal and abnormal cardiac
strain patterns
40
Smoothed Gradient Computation

• Subject to noise and interference
• Need regularized gradient computation
• Outline of solution
• shift to DC
• compute finite difference
• 1D local phase unwrap
• 1D average filter
• shift back up to tag frequency

Before
After
41
Smoothed Circumferential Strain
Before
After
20
-20
Negligible degradation of transmural strain
42
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

43
Conventional Tagged MRI
16 heartbeat breath-hold
Horizontal Tags
Vertical Tags
APPLY TAGS
IMAGING
End Diastole
1 heartbeat
44
HARP-MRI Requirements
45
256x128 to 322
46
FastHARP Pulse Sequence
Sampath, Derbyshire et at, ISMRM 2001
Time to acquire one image 45 ms
47
FastHARP Imaging Protocol
Complete acquisition in 2 heartbeats
Horizontal Tags
Vertical Tags
APPLY TAGS
IMAGING
End Diastole
1 heartbeat
48
Breath-hold FastHARP
Normal male Age 26 60 bpm
Eulerian Circumferential Strain
49
Non-breath-hold FastHARP
Normal male Age 26 60 bpm
Eulerian Circumferential Strain
50
Circumferential Strain Comparison
dot-dash breath-held 1-1 SPAMM (16 HB)
dot breath-held FastHARP
solid non-breath-held FastHARP
51
Real-time FastHARP MRI
Kraitchman et al, ISMRM 2001

• Canine ischemia model
• LAD occlusion
• continuous acquisition FastHARP
• Image pairs to display tagged images
• Occlusion around frame 80
• breathing apparent after frame 80

52
Systolic Referenced Strain

• Reference strain to end-systole -- early image
• Display only end-systolic strain images
• Occlusion event is very apparent
• approx 6 second delay in strain abnormality

53
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

54
Recall Harmonic Expansion

• A tagged image can be written as
• Images Il are related by Fourier series
  expansion

A non-DC harmonic image can be used to
synthesize a new Fourier series expansion to
create sharp tag lines at multiples of the tag
frequency
55
Fourier Synthesis of Tag Patterns
Osman, Prince, ISMRM 2001

• Acquire harmonic phase image
• Demodulation recovers DC peak
• Remodulation synthesizes harmonics
• Combine harmonics in Fourier series to generate
  sharp tag lines
• Motion is automatically included in the
  synthesized harmonics

56
Tag Synthesis Example
Can synthesize tags at higher frequency
57
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

58
Out-of-Plane Tagging
w tag frequency

• The initial longitudinal magnetization is

59
Effect of Longitudinal Strain

• Harmonics appear because of tag modulation
• Harmonics shift because of tissue longitudinal
  strain
• DC appears because of tag fading
• Local frequency is related to strain

60
Strain-Encoded Imaging (SENC)

• The intensity of an MR image is determined by
• s(z) is the slice profile
• kz is a phase encode parameter

Strain can be dialed in by choosing kz
61
Systolic Image Different z-Encode
62
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

63
Normals in Stress and MI Patients
Garot et al., Circulation 2000

• 10 healthy male volunteers
• 37 ? 10 years
• 5 and 20 ?g/kg/min dobutamine
• standard SPAMM-tagged MRI breath-hold acquisition

• 9 patients
• 3?2 days after acute MI
• 4 anterior, 5 inferior
• Age 43?11 8 men
• standard SPAMM-tagged MRI breath-hold acquisition

64
Dobutamine Stress Data
65
Evolution of Circ Shortening
N10
66
Acute MI Data
5-segment division
Systolic sequence Recent anterior AMI
Circumferential strain
67
Circ Shortening in CAD
akinetic
disfunctional
Circumferential Shortening
remote
Classified by cine-MR wall-thickening measures
68
Circ Shortening (pooled data)
Circum. strain (HARP)
(1 hour per study)
Circumferential strain (FindTags)
(10 hours per study)
69
Coronary Occlusion Model
Garot, et al., 2000

• 7 mongrel dogs
• LV catheter for micro-spheres and pressure
• balloon catheter in midpoint of LAD
• 90 min at 8 atm inflation
• MR images acquired 48 hours after reflow
• SPAMM tagged MRI at 7mm tag separation

• 5-7 short axis cross sections breath-holds
• baseline and 5 ?g/kg/min dobutamine
• TTC staining and gamma spectrometer counting
• classification into
• infarcted
• at risk
• remote

70
Circumferential Strain
strain
71
HARP Reserve Conclusions

• Risk regions retained maximal contractile reserve
  after acute MI
• No functional reserve was found in segments with
  transmural infarction
• Under dobutamine stimulation, there is
  significant augmentation of strain in
  subendocardial infarcts.

72
Outline

• Magnetic Resonance Tagging
• Basics of HARP
• HARP Advances
• Validation studies
• Improved filtering
• FastHARP MRI pulse sequence
• Synthetic Tags
• SENC strain encoded imaging
• Clinical Applications Results
• Future Directions

73
Future Directions

• FastHARP
• Add CSPAMM and slice following
• Spiral acquisition parallel acquisition (SENSE)
• Multislice 3D strain imaging protocol
• HARP processing
• reduce computation time
• improve bandpass filtering (tracking motion)
• exploit refinement for phase unwrapping
• Clinical and medical science
• detect subclinical ischemic heart disease
• certify FastHARP for dobutamine stress tests

74
The End
75
(No Transcript)
76
HARP Tracking Algorithm
Incremental displacement
77
Two HARP angles per pixel
Time sequence of HARP images
78
3D HARP-MRI
Pan, 2001, unpublished

• Acquire multislice tagged MRI data
• Grid each slice
• Calculate Lagrangian strain on octants
• Stack grids,display Lagrangian strain
• Lacks longitudinal compression

79
Eulerian Strain Time Evolution
Early stretching
Early contraction
Strain before first image!
pacer lead
80
Maximum Thickening

• Matrix of tag frequencies
• Left Cauchy-Green strain tensor
• Angle of maximum thickening is
• is max eigenvector of B
• is radial direction

81
Max Thickening Geometry
abnormal (in systole)
normal (in systole)
82
Max Thickening in Paced Canine
A contraction at pacer lead B contraction
begins around LV C stretching on opposite
wall D resting muscle E contraction on opposite
wall F all contraction
83
Computing n(y,t)

• Compute the image
• Estimate the local frequency using
• Coefficients a and b depend on the slice profile

84
Longitudinal Strain Map

• Normal human, male age 28
• End-systole
• Dark gray 9 strain
• White 12 strain

85
SENC Pulse Sequence
z tagging
kz encoded imaging
86
Full Cycle in Cardiomyopathy
Beache, CVMR 2001