A Low-dose, Accurate Medical A Low-dose, Accurate Medical Imaging - - PowerPoint PPT Presentation

A Low-dose, Accurate Medical Imaging Method for Proton Therapy: Proton Computed Tomography A Low-dose, Accurate Medical Imaging Method for Proton Therapy: Proton Computed Tomography

Bela Erdelyi

Department of Physics, Northern I llinois University, and Physics Division, Argonne National Laboratory FFAG’09, Fermilab, Batavia, September 21-25, 2009

September 21-25, 2009 Proton Computed Tomography 2

Acknowledgments

• Northern I llinois University

– Physics: G. Coutrakon, V. Rykalin, K. Wong – Computer Science: N. Karonis, K. Duffin, K. Naglich, J. Panici

• Loma Linda University Medical Center

– R. Schulte, V. Bashkirov, F. Hurley, S. Penfold

• Santa Cruz I nstitute for Particle Physics

– H. Sadrozinski, M. Petterson, N. Blumenkrantz,

• B. Colby, J. Feldt, J. Heimann, R. Johnson, D.

Lucia, D. C. Williams

September 21-25, 2009 Proton Computed Tomography 3

Outline

• The main idea
• A little history
• The fundamentals
• Current status
• Summary

September 21-25, 2009 Proton Computed Tomography 4

The Main Idea

• For proton therapy, one positions the

Bragg peak onto the tumor

• For pCT, raise the initial energy so protons

traverse the object to be imaged

• Measure the phase space data of each

proton individually

• Use this data to construct the electron

density map of the object traversed by protons

• Use the resulting electron density map for

diagnosis, proton therapy treatment plan, adaptive treatment, positioning verification, etc.

September 21-25, 2009 Proton Computed Tomography 5

Motivation: Range Uncertainties

September 21-25, 2009 Proton Computed Tomography 6

Overview

September 21-25, 2009 Proton Computed Tomography 7

Advantages and Benefits

• Practically eliminates range uncertainties,

therefore allowing very accurate and precise proton treatment plans

• Provides fast patient positioning

verification and adaptive treatments, if necessary

• Achieves a reduced dose necessary for

imaging relative to XCT

• Provides a quantification of the range

uncertainties as a function of tumor site, type, etc. that will be useful to any proton therapy facility in operation that lacks a pCT system.

September 21-25, 2009 Proton Computed Tomography 8

History of pCT (1)

September 21-25, 2009 Proton Computed Tomography 9

History of pCT (2)

• I n the late 1970s, Ken Hanson (LANL) and

Kramer et al. (ANL) experimentally explored the advantages of pCT and proton radiography

• They pointed out the dose reduction w.r.t. XCT

and the problem of limited spatial resolution due to proton scattering

• I n the 1990s Ron Martin (ANL) proposed building

a proton CT system using a scanning beam proton gantry

• During the 1990s Uwe Schneider (PSI ) further

developed the idea of proton radiography as a tool for quality control in proton therapy

• I n the late 1990s Piotr Zygmanski (MGH) Harvard

Cyclotron tested a cone beam CT system with protons

• pCT Collaboration (2003- )

September 21-25, 2009 Proton Computed Tomography 10

The Fundamentals (1)

The Bethe-Bloch formula gives the mean energy loss rate

• f protons in a medium

−dE (~ r) dl = ηe (~ r) F (I (~ r) , E (~ r)) F (I (~ r) , E (~ r)) = K 1 β2 (E) · ln µ2mec2 I (~ r) β2 (E) 1 − β2 (E) ¶ − β2 (E) ¸ Z L ηe (~ r) dl = − Z E(L)

E0

dE F (I (~ r) , E (~ r)) I(~ r) 7→ Iwater Discretize ηe (~ r) over some basis functions (typically voxels) ⇒ xi, i = 1, n. Each proton will determine a linear equation in the variables xi

A~ x = ~ b

⇒ rhs is a numerical integration ⇒ bi, i = 1, m.

September 21-25, 2009 Proton Computed Tomography 11

The Fundamentals (2)

A~ x = ~ b

Determined by proton paths Electron densities Determined by average proton energy loss over paths Not known exactly due to MCS Not known exactly due to energy loss straggling From measurements

A(ξ)~ x(ξ) = ~ b(ξ)

Random realization with ξ drawn from a probability distribution

September 21-25, 2009 Proton Computed Tomography 12

The Most Likely Path

Deterministic systems

Transfer map known fixed computed

September 21-25, 2009 Proton Computed Tomography 13

The Most Likely Path

Stochastic systems

Transfer map known computed known

September 21-25, 2009 Proton Computed Tomography 14

Spatial Resolution (1)

A measure of our ability to predict each individual proton’s trajectory inside the object to be imaged is the spatial resolution A measure of our ability to predict each individual proton’s trajectory inside the object to be imaged is the spatial resolution

σ (l, E) = 13.6 MeV

c

β (E) p (E) r l X · 1 + 0.038 ln µ l X ¶¸

Multiple Coulomb Scattering (MCS)

Example: 200 MeV protons in 20cm of water have σ=39mrad -> σlat=3.5mm

Use constrains to reduce uncertainty: position, direction, energy

September 21-25, 2009 Proton Computed Tomography 15

Spatial Resolution (2)

• Developed new formalism to include energy as a constraint
• Equivalently, it fixes the trajectory length
• Example: 2 protons, with exactly the same incoming energy,

position, direction, and outgoing position and direction – but different outgoing energy

• Previous MLP formalism gives the same MLP, new one is different

due to the different path lengths

• Also implies improved spatial resolution – difficult to compute

analytically

September 21-25, 2009 Proton Computed Tomography 16

Electron Density Resolution (1)

A(ξ)~ x(ξ) = ~ b(ξ)

A measure of our ability to predict from the random vector is the electron density resolution A measure of our ability to predict from the random vector is the electron density resolution

~ x(ξ) ~ x

σx = sPn

i=0 σ2 xi(ξ)

n

Definition: Definition:

σx = g σhEoutikA vF ( hEouti) √mv

September 21-25, 2009 Proton Computed Tomography 17

Electron Density Resolution (2)

September 21-25, 2009 Proton Computed Tomography 18

Reconstruction Methods Projection Methods

Basic property: To reach any goal that is related to the whole family of sets by performing projections onto the individual sets. Basic ability: To handle huge-size problems whose dimensions are beyond the capabilities of current, more sophisticated, methods.

September 21-25, 2009 Proton Computed Tomography 19

Algebraic Reconstruction Technique

H1 H2 H3 H4 H5

k

x

1 k

x +

September 21-25, 2009 Proton Computed Tomography 20

Block Iterative Projection

1

(1,2,3) B =

2

(4,5,6) B =

H1 H2 H3 H4 H5

k

x H6

1 k

x +

September 21-25, 2009 Proton Computed Tomography 21

String Averaging

1

(1,3,5,6) I =

2

(2) I =

3

(6,4) I =

H1 H2 H3 H4 H5

k

x

1 k

x + H6

September 21-25, 2009 Proton Computed Tomography 22

Hardware Implementation

• Desktop/ laptop
• Compute Clusters
• GP-GPU
• GP-GPU clusters
• Hybrid: multi-core CPUs + GP-GPUs (clusters)

Speedup of the rel. electron density calculation when performed with a NVIDIA GTX280 GPU relative to a Intel Q6600 quad core CPU (Scott McAllister, Master’s Thesis, Cal State SB, 2009)

September 21-25, 2009 Proton Computed Tomography 23

GEANT4 Simulations

September 21-25, 2009 Proton Computed Tomography 24

Reconstruction Results From Simulations

1 cycle 5 cycles 10 cycles BICAV DROP OS-SART CARP

September 21-25, 2009 Proton Computed Tomography 25

Reconstruction Results From Real Data

pCT prototype pCT prototype pCT phantom pCT phantom

September 21-25, 2009 Proton Computed Tomography 26

The pCT System Prototype Schematic

Ready by early 2010

September 21-25, 2009 Proton Computed Tomography 27

System Components

September 21-25, 2009 Proton Computed Tomography 28

Next Phase

September 21-25, 2009 Proton Computed Tomography 29

Summary

• pCT is a new medical imaging method that will

greatly benefit proton therapy in general and will

• ffer a low-dose diagnostic imaging modality
• The project is truly interdisciplinary involving

physics, mathematics, computer science and engineering

• The NI U-LLUMC-SCI PP Collaboration is well

underway; the first pCT prototype system capable of imaging head-sized objects will be ready by early 2010

• Further work is necessary towards a fully clinical
• peration-ready pCT system