The Forefront in Image Processing: PETMolecular Approaches

1
The Forefront in Image Processing PET/Molecular
Approaches
Joel Karp University of Pennsylvania
Sixth Annual NCI-Industry Forum Quantitative
Oncologic Imaging April 7-8, 2005
2
Issues of Performance, Image Processing,
Quantification

• Performance of current-generation PET
  scanners Global effects - data
  correction Local effects - image
  reconstruction Statistical and count-rate
  effects
• Self-consistency instrument performs same
  day-to-day
• Cross-consistency all instruments produce same
  result
• Comparing images (PET and CT) from different
  patients, different instruments, and different
  institutes

3
What is Measured with PET
b
Random coincidence (2t . singles2)
True coincidence
a
Scattered coincidence
Coincidence?
Yab Nab(AabTab Sab Rab)
Record event
Trues
What is measured
Normalization
Attenuation
Randoms
Scatter
4
Reconstruct image from line-of-response (LOR)
projection data
If there are N counts in the image, SNR ? N /
(N)1/2
Signals from Different Voxels are Coupled ?
Statistical Noise Does Not Obey Counting
Statistics
5
Iterative reconstruction
Data
y
A x y
Back-projection
c(k)
Differ ence
d(k)
Correction for Attenuation, Scatter, Randoms
Start here
Update
Forward projection
y(k)
x(k)

Image
6
Data Flow
DETECTOR
Philips Allegro 616 x 29 crystals
RAWVIEW (52 bytes/event) For
A,B side of event 26 PMT energies/zone (26
bytes) 100M events 5200 Mbytes
DIGITIZER
POSITION CALCULATOR
LISTVIEW (8 bytes/event) For
A,B side of event 2D position (3 bytes)
timestamp Energy (1 byte) TOF
(1 byte) 100M events 800 Mbytes
BINNER
SINOGRAM (80 Mbytes/frame) R,
Phi (295x161x2 95 Kbytes) Slice
(292 841) 100M events 560 Mbytes
(7 frames)
COMPUTER
IMAGE X,Y
(128x128 16 Kbytes) 100M events 4 Mbytes (250
slices)
reconstruction
PACS archive
7
2D (septa) vs. 3D (no septa)
2D Imaging
3D Imaging
S
T
R
Low geometric sensitivity
High geometric sensitivity
Scatter decreases with high energy threshold -
depends on energy resolution
8
Out-of-field activity increases randoms in 3D
Problem increases as bore size increases -gt less
shielding

• Randoms 2t . Singles2
• decreases with narrow timing window (2t)
• decreases with high energy threshold
• estimated (and subtracted) with 2nd (delayed)
  timing window

9
Count-rate Performance
10 mCi dose
70-cm long x 20-cm diameter NEMA 2001 (body)
Noise Equivalent Count-rate NEC
T/(1S/TR/T) NEC SNR2
Philips Allegro
10
PET Imaging Performance

• Spatial resolution -gt partial volume
  effect intrinsic 4-6 mm reconstructed gt1
  0 mm
• Scatter fraction -gt noise and bias (after
  correction) 2D 10-20 SF 3D 30-60 SF
• Sensitivity and count-rate capability -gt
  statistical quality 25 - 100 kcps or 5 M -
  20 Mevents per 3 min frame

11
Scatter Correction
Before Scatter correction
After Scatter correction
Single Scatter - Model based correction Calculate
the contribution for an arbitrary scatter point
using the Klein-Nishina equation
12
Attenuation correction with radioisotope
transmission scan
20 mCi 137Cs source - 662 keV
A 1 / e -md d length of chord through
tissue m attenuation coefficient
13
Attenuation correction for PET
Types of transmission images
Single photon Cs-137 (662 keV) lower noise 5-10
min scan time some bias lower contrast
X-ray (30-140kVp) no noise 1 min scan
time potential for bias high contrast
Coincident photon Ge-68/Ga-68 (511 keV) high
noise 15-30 min scan time low bias low contrast
14
Attenuation/Scatter correction
No AC or Scatter Corr
AC and Scatter Corr
University of Pennsylvania PET Center
Philips Allegro
15
Fully 3D Iterative Reconstruction improves image
quality
Fore-FBP
3D Ramla
How about quantification?
16
NEMA NU2-2001 Image Quality Phantom
foam
Out-of-field Activity
17
Partial Volume Effect
18
NEMA IEC Phantom
LOR RAMLA reconstruction
Vary relaxation parameter l from 0.00025 (top
left) to 0.075 (bottom right)
19
Contrast vs. Noise
Iterative - Ramla Filtered Backprojection (FBP)
1.7 cm hot sphere
2.8 cm cold sphere
20
Image processing Filters for restoring the
spatial frequency components
Low (left) - Maximum gain
2.5 Medium(middle)- Maximum gain 3.5 High
gain (right) - Maximum gain 4.5
k - parameter describing the Gaussian
roll-off fcut - cutoff frequency K, fcut -were
bracketed from an analysis of phantom
data
WF(f) 1/MTF(f) for
fltfcut WF(f) 1/MTF(fcut) exp-kf 2
for fgtfcut
21
Profile through the lesion
Lesion contrast improves with filtering
no
low
med
high
22
Time-of-Flight list-mode iterative
reconstruction
no TOF
300 ps TOF
1 Mcts
5 Mcts
10 Mcts
23
Challenges in comparing images

• Spatial resolution differences partial
  volume - simple (approximate) correction spatial
  recovery in reconstruction model adds noise
• Reconstruction algorithm local convergence
  depends on algorithm and activity
• Accuracy of corrections - randoms, scatter,
  attenuation depends on patient size and activity
  distribution
• Imaging protocol scan acquisition time
  and delay post-injection
• Quantification - typically based on simple
  cylinder QC - monitor and correct daily
  drifts Activity calibration - counts/voxel/min
  -gt nCi/ml Count-rate corrections - dead-time

24
Challenges in comparing images

• Instrumentation in PET is constantly
  evolving performance of new scanner gtgt older
  scanner
• Image data size is large - data transfer and
  archiving PET 4 Mbyte (with 4 mm3
  voxels) CT 64 Mbyte (with 1 mm3 voxels)
• DICOM quantification (SUV) requires PT
  format (not NM) manufacturers workstations still
  most practical
• Data analysis tools must be standardized and
  validated region-of-interest
• Image processing behavior must be
  understood - difficult to standardize