Department of Radiology UNIVERSITY OF PENNSYLVANIA

Image Reconstruction inPositron Emission Tomography

Robert M. Lewitt Samuel Matej

Ivan G. Kazantsev Lucretiu Popescu

PET Reconstruction TeamMedical Image Processing Group

2

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Examples of clinical PET images

S.H.

P1309

clf2716

M.B.

P1310

clf2717

M.G.

P1311

clf2720

Female82 years

1.70 m (5’7”)101 kg

Lung carcinoma

Male81 years

1.90 m (6’3”)105 kg

Lymphoma + Lung carcinoma

Female24 years

1.55 m (5’1”)45 kg

Lymphoma

Advantages: high sensitivity to cancer Problems: resolution, noise, attenuation, scatter, randoms

(Illustrated images lack proper treatment of those effects)

3

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Whole-bodyPET study –Fully 3D reconstruction

4

Department of Radiology UNIVERSITY OF PENNSYLVANIA

PET data - classification of events

false coincidence lines

1) True event

2) Scatter event

3’, 3”) Single events

3’+3”) Random event

detector crystal

detector crystal

3”3’ 2

2

Side shield

Patient shield

1

5

Department of Radiology UNIVERSITY OF PENNSYLVANIA

PET data - 2D vs. 3D scanner geometry

“2D” detector crystal

detector crystal

“3D”

⇒ The larger the acceptance angle, the more (good and bad) events are accepted.

6

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Fully 3D reconstruction

Fully 3D PET reconstruction flowchart

4DAC, Scatter, Randoms,…

Forward-projection

Fully 3D dataEmision data

x

x

x

x

3D image

Emission image

Attenuation imageTransmission data Transmission reconstruction

Scatter, randomsestimation

7

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Fast Fully 3D Reconstruction – WHY?

Modern emission tomography systems:

Fast increase of data sizes (exceeding Moore’s law)→ needed – reduction of computation demands of reconstruction

Low counts per data bin – noisy dataData attenuation, scatter and contamination

→ needed – reconstruction techniques with better modeling

Conflicting demands→ needed – very fast reconstruction approaches

8

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Studied 3D PET reconstruction approaches

3D non-iterative analytical techniques (3DRP, 3D-FRP)

3D iterative techniques (3D RAMLA, ...)

Rebinning (into non-oblique data) followed by multislice2D or 2.5D iterative reconstruction

List mode reconstruction

Time-of-flight reconstruction

Dynamic list mode reconstruction

9

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Favorite tools

Fourier-based approachesAnalytical reconstructionForward and back-projectors for attenuation correction and iterative reconstruction

Kaiser-Bessel window functionsImage basis functionInterpolatorsDisplay 0

0.25

0.5

0.75

1

0 1 2 3

Radius

αα

m

Efficient gridsReconstruction Display

10

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Analytical 3D/2.5D reconstructions

3D-FRP

3DRP

FORE reconstruction

11

Department of Radiology UNIVERSITY OF PENNSYLVANIA

AllegroIterative

approaches

3D recon2.5D recon

12

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Fourier-based projection

Projection Difference No zeropadding 100% zeropadding

1% scale 0.5% scale

58 ms/view 96 ms/view

13

Department of Radiology UNIVERSITY OF PENNSYLVANIA

Fourier-based iterative reconstruction

14Department of Radiology UNIVERSITY OF PENNSYLVANIA

The End

Image Reconstruction inPositron Emission TomographyWhole-bodyPET study – Fully 3D reconstructionFully 3D PET reconstruction flowchartFast Fully 3D Reconstruction – WHY?Studied 3D PET reconstruction approachesFavorite toolsAnalytical 3D/2.5D reconstructionsAllegroFourier-based projectionFourier-based iterative reconstructionThe End