Research at the Interface of Physics and Medicine: An Update on the Development of Proton Computed Tomography Reinhard Schulte Department of Radia@on Medicine Loma Linda University Medical Center Loma Linda, California, USA R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
48
Embed
Research(atthe( Interface(of(Physics(and(Medicine:( An ...scipp.ucsc.edu/.../RSchulte_UCSC_colloquium_talk_Nov_3_2011.pdf · Tomography,(UCSC(Physics(Colloquium,Nov3,2011...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Research at the Interface of Physics and Medicine:
An Update on the Development of Proton Computed Tomography
Reinhard Schulte Department of Radia@on Medicine
Loma Linda University Medical Center Loma Linda, California, USA
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Outline
• The Clinical Perspec@ve: Proton Therapy and Proton Imaging
• Concepts of Proton CT • Technical Realiza@ons of Proton CT – Phase I preclinical head scanner – Phase II clinical head scanner
• Next steps & future applica@ons/Plans
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
THE CLINICAL PERSPECTIVE: PROTON THERAPY AND PROTON IMAGING
SECTION I
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Depth-‐Dose Characteris@cs of Mono-‐Energe@c Protons
• Compared to x-‐rays and electrons (which are also used therapeu@cally) protons have a unique inverted depth dose profile
• The dose peak at the end of the proton range is called “Bragg peak” aXer the physicist William Henry Bragg, who discovered it in 1903
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Protons in Comparison • Proton energy deposi@on per
track length (propor@onal to dose) increases as they slow down
• Bragg peak dose at the end of the proton range
• Depth of peak (proton range) adjustable by choosing right energy
• Ac@ve or passive modula@on generates spread-‐out Bragg peak (SOPBP)
• Dose sparing up-‐ and down-‐stream from tumor
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Protons for Precision Radia@on Therapy / Radiosurgery
• Addi@onal features that make protons an a^rac@ve modality for proton radia@on therapy/radiosurgery: – Pencil beam forma@on with
magne@c lenses – Magne@c tracking of proton
pencil beams – Edge tracking (not possible with
photon beams)
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Proton Therapy: A man – A Vision • Robert Wilson, Ph.D.
(1914-‐200) was the first to publish a paper on the medical uses of protons (Radiology 1946:47:487-‐91)
• It took 45 years before his dream became reality when the first hospital-‐based proton treatment center opened at LLUMC
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Robert R. Wilson, Ph.D., 1914-‐2000
The Early Years of Proton & Heavy Ion Radiosurgery at the
Lawrence Berkeley Laboratory (1948-‐1955) • Star@ng in 1948, John Lawrence
(physician) and Cornelius Tobias (biophysicist) developed biomedical program of heavy ions at the LBL cyclotrons
• In 1954, the LBL group began to direct the high doses of heavy ion beams (protons & helium) at human pituitary glands (about 50 pa@ents)
• Reported successful hormonal abla@on & regression of disease in advanced breast cancer pa@ents
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Large-‐Scale Proton Radiosurgery: The Harvard Experience
1961-‐2001 • In 1961, MGH neurosurgeon Raymond
Kjellberg began trea@ng pa@ents with pituitary adenomas using 160 MeV Bragg peak protons from the Harvard Cyclotron Laboratory (HCL)
• Star@ng in the 1970s, Dr. Kjellberg also treated large, inoperable arteriovenous malforma@ons (AVMs) with Bragg peak protons, despite limita@ons in imaging and planning techniques at that @me
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Proton Therapy at LLUMC
• Proton therapy was used in a hospital sekng first at LLUMC in 1990
• More than 14,000 pa@ents have been treated at LLUMC
• Due to the Bragg peak feature, protons deliver less dose to normal @ssue than IMRT
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
LLUMC Facility Design • Built around a proton accelerator (weak-‐focusing synchrotron) used to deliver
protons of 10-‐300 MeV (Gantry accepts up to 250 MeV).
Worldwide Expansion of Proton Therapy (1990 – 2010)
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
History of Proton Radiography (pRad) • A. Koehler was the first to point
out the poten@al value of pRad and to perform experiments with 160 MeV (Koehler, Science 160, 303–304, 1968)
• The higher density resolu@on but poorer spa@al resolu@on was noted by Koehler and later by Kramer et al. (Radiology, 1980)
• Medical interest in pRad as a QA tool for proton therapy was revived by U. Schneider at PSA during the 1990s
• Another strong mo@va@on of pRad development comes from nuclear weapons tes@ng programs at Los Alamos NL
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Andy Koehler, former director of the Harvard Cyclotron)
Proton Computed Tomography (pCT): A man – A Vision (Harvard, 1963)
• Alan M. Cormack, physicist (1924-‐1998) was the first to publish a paper on the reconstruc@on of tomographic images based on X-‐ray absorp@on and proton degrada@on (J. Appl. Phys. 34, 2722, 1963)
• It took less than 10 years before his idea became reality when the first when Geoffrey Hounsfield constructed the first X-‐ray CT scanner
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Alan M. Cormack, 1924-‐1998 Physics Nobel Laureate 1979
The Proton CT Collabora@on
• Group of scien@sts first met at IEEE in Norfolk, VA in 2002 and BNL in 2003 to outline the goal of building a clinical proton CT (pCT) scanner, many have contributed since then
• We have remaining issues due to – Restric@ons in beam entry direc@ons – CT ar@facts in the presence of metallic
hardware, dental fillings emboliza@on glue etc.
– Intra-‐ and inter-‐treatment changes of proton range (mo@on, weight loss etc.)
– Higher RBE of distal edge when placed into cri@cal normal @ssues
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
CONCEPTS OF PROTON CT SECTION II
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
The Principle of Tomography • The Radon transform (Radon,
1917) relates any 2D func@on f(x,y) to an infinite number of line integrals covering an angular space of 2π
• Radon proved mathema@cally that both of the func@on representa@ons are equivalent
• The inversion of the Radon transform from an infinite set of line integrals forms the basis for tomographic image reconstruc@on
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Computed Tomography: X-‐rays vs. Protons
0.01
0.1
1
10
100
1000
104
0.001 0.01 0.1 1
X-Ray Absorption Coefficient
Muscle
Bone
Water
Air
µ
Energy [MeV]
[1/cm]
A^enua@on of Photons, Z ln(N) = No∫ µ x dx
Energy Loss of Protons, ρ ∫=Δ dx
dxdEE
NIST Data R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011 Courtesy: H. Sadrozinski,
UCSC
pCT Single Proton Concept
• An energe@c low intensity cone beam of protons traverses the pa@ent
• The posi@on and direc@on (entry & exit) and energy loss of each proton is measured
• Proton histories from mul@ple projec@on angles
• Minimal proton loss and high detec@on efficiency make this a low-‐dose imaging modality
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Design of a Proton CT Scanner rota@ng with the proton gantry (R Schulte et al. IEEE Trans. Nucl. Sci., 51(3), 866-‐872, 2004)
Low intensity proton beam
Tracking of individual protons
Principles of the pCT Imaging Process
• The measurement is propor@onal to the outgoing energy of each proton, and thus to energy loss in the phantom
• The energy loss can be converted to a line integral of proton stopping power rela@ve to water along the proton path, RSP is (prac@cally) independent of proton energy
• Solu@on → calculate most probable path of the proton within the object using a reconstruc@on algorithms that takes MCS into account
• Requires reconstruc@on algorithm that can handle non-‐linear paths
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
24
Sta@s@cal Model of a Proton Path
( )( ) ( )(( ) ( )) ⎟
⎟⎟
⎠
⎞
⎜⎜⎜
⎝
⎛
−Σ−
+−Σ−−=
−
−
11212112
00111001
21 21
exp|yRyRyy
yRyRyyyyL
TTT
TTT
Path Reconstruc@on Concepts • Different proton paths may be
used in the reconstruc@on: MLP = Most Likely Path, SLP = straight line path, CSP = cubic spline path
• The MLP is determined by maximizing likelihood (chi square) of output parameters, given entry parameters
• MLP significantly improves spa@al resolu@on compared to SLP, CSP reconstruc@on is nearly as good as MLP reconstruc@on
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
• A total number of m protons trace paths through the voxel space and aij is the intersec@on length of the ith proton, i=1…m with the jth voxel
• In the fully discre@zed model, the object is treated as n-‐dimensional vector xj, j=1…n, where xj is the constant rela@ve stopping power of the jth voxel
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
The Discrete pCT Reconstruc@on Problem
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
First Itera@ve Algorithm Applied: Algebraic Reconstruc@on Technique (ART) (work done by
Tianfang Li & Jerome Liang, SUNY) • Originally developed by
Kaczmarz, 1937 • Sequen@al orthogonal
projec@ons onto hyperplanes
• Works but well for proton CT but is inherently slow
• High frequency noise is present
xk+1
Li et al Med Phys 2006
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Further Development of pCT Reconstruc@on with Geant4 Simula@ons (work done by Sco^ Penfold)
• Simulate pCT system and digital head phantom1 • Test/compare different reconstruc@on algorithms
1G. T. Herman, Image Reconstruction From Projections: The Fundamentals of Computerized Tomography, Academic Press, New York (1980).
TECHNICAL REALIZATIONS OF PROTON CT
SECTION III
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Phase I pCT Scanner Design (completed in 2010)
• Horizontal beamline setup
• Rota@onal stage for object rota@on
• Upstream and downstream tracker modules
• Downstream energy detector (calorimeter)
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Phase I pCT Tracker (UCSC) • The Phase I pCT tracker consists of
front and rear module for loca@on and direc@on measurements
• Modules: two detector boards measuring the X-‐Y posi@on in two loca@ons => direc@on
• Detector boards: 4 Si Strip Detectors (SSDs), 9 cm x 9 cm, 384 strips, 0.23 mm pitch
• Strips oriented in horizontal or ver@cal direc@on (X and Y sensi@vity)
• Total sensi@ve area 9 cm x 18 cm • Modified GLAST/Fermi readout chip,
max rate 100 kHz
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Phase I pCT scanner front module Detector board with 2 SSDs in the front (visible) and 2 SSDs in the back of the board
Phase I pCT Energy Detector (Calorimeter) (NIU-‐LLUMC, UCSC)
• Crystal matrix with 18 thallium-‐doped cesium-‐iodide (CsI(Tl)) crystals (~3.6 cm x 3.6 cm x 12.5 cm)
• Each crystal read out by area-‐matched Si photodiode
• Si photodiode => preamp/shaper => ADC
• Excellent linearity and energy resolu@on < 1% above 40 MeV
• Integrated with rear tracker module
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Phase I pCT DAQ Hardware (UCSC – LLU)
• Design based on field-‐programmable gate arrays (FPGAs) -‐ Master & Slave
• Reads and processes data from 16 tracker SSDs & 18 crystal frontends in parallel
• Data rate has been op@mized
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Phase I pCT DAQ SoXware/GUI (Ford Hurley, LLU)
• User-‐friendly user interface for main pCT func@ons
• Online display of tracker and calorimeter response
• Root-‐based reconstruc@on of proton histories
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Phase I pCT Scanner at LLUMC: Timeline
• System component integra@on & moun@ng (April 2010)
• Tes@ng with radioac@ve source and cosmic rays (muons)
• Installa@on on proton research beam line & 1st test runs (May 2010)
• Spill uniformity op@miza@on (June 2010)
• Scanner calibra@on (July 2010) • Phantom scans since Dec 2010
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Crystal Calorimeter Calibra@on • The rela@ve sensi@vity
(individual weigh@ng factors) for individual crystals are measured at the scan energy before each scan
• Only central proton histories that most likely did not leave the individual crystal are used
• Weigh@ng factors between different crystals vary within +/-‐ 15%, and between different energies <1%
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Proton tracks used for to derive a rela@ve weigh@ng factor for each crystal
Phase I pCT Scanner WET Calibra@on • The calorimeter response is
calibrated against polystyrene plates of known thickness and rela@ve stopping power, resul@ng in individual response curves, to which a Gaussian was fi^ed
• The Gaussian peaks are used to construct a curve that converts calorimeter response to a water-‐equivalent path length (WEPL)
• The response vs. WEPL curve is fi^ed to a second order polynomial, providing fast conversion to WEPL
• This inverse calibra@on was verified by comparing the measured stopping power of @ssue equivalent plates to known values
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Developing Advanced Reconstruc@on Algorithms with GEANT4 Simula@ons • Reconstruc@on of the
original object (a) with a constant intersec@on length (equal to the voxel size) leads to “noisy” images (b)
• Using a refined intersec@on length depending on projec@on angle (c) or varying from voxel to voxel (d) leads to significantly improved images
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Status of Proton CT Reconstruc@on
• A series of pCT scans of the Lucy radiosurgery phantom was started in Oct 2010
• Phantom images were reconstructed with our newly developed 3D reconstruc@on algorithm
• Good image quality has been achieved
• A sytema@c evalua@on using standard CT phantom modules and a realis@c anatomical head phantom is underway
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
GPU-‐Accelerated Reconstruc@on • Reconstruc@on is
performed using an NVIDIA Tesla general purpose graphics processing unit (GP-‐GPU)
• This state of the art inexpensive computer cluster technology speeds up reconstruc@on @me by orders of magnitude, compared to tradi@onal CPUs
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
NVIDIA Tesla C1060 GPU: 240 stream processors with 4 GB of device RAM
Phase I pCT Reconstruc@on of Lucy
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
14 cm diameter polystyrene sphere with @ssue equivalent inserts
64 slice Mul@-‐Slice X-‐ray CT, 0.53 x 0.53 x 0.625 mm3
pCT Phase I, 0.63 x 0.63 x 2.5 mm3
Phase I pCT Reconstruc@on of the Catphan® Uniformity Module
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
• A series of pCT scans of standard CT performance phantom modules is underway
• The uniformity module tests hypothesis that pCT RSP reconstruc@on is not affected by energy loss
• The noise power analysis shows interes@ng characteris@cs of pCT reconstruc@on algorithms
NEXT STEPS & FUTURE APPLICATIONS/PLANS
SECTION IV
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
• Stepwise improvement of reconstruc@on, image quality & data rates
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Rat scanned on the Phase I pCT scanner at LLUMC
Development of Phase II Clinical pCT: Issues and Possible Choices
• Size of System – Large area tracker suitable for head & neck, thorax (36 cm x 9 cm), may move longitudinally during scan
• Type of Tracker – Silicon (larger, thinner, ac@ve edge = “edge less” (implemented by SCIPP)
– Scin@lla@ng fiber with Si Photomul@plier readout (implemented by NIU)
• Type of Energy/Range Detector – “Modern” crystals (e.g. YAG:Ce, Menichelli et al 2010 IEEE TNS) – Range detector: Stack of plas@c scin@llators with direct or fiber-‐mediated SIPM readout (implemented by NIU, tested at LLU)
– Segmented scin@llators (tested at LLU)
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Primary Applica@ons for pCT • Higher planning accuracy/precision needed – Radiosurgery for vascular
malforma@ons, pituitary adenomas, meningiomas, etc.
– High-‐dose boosts to tumors near organs at risk for damage
– Crea@ng lesions in defined loca@ons for pain treatment
of vertebral bodies in children • X-‐ray CT ar@facts
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
Where is the Target?
Future Physics Contribu@ons to Proton Therapy
• Adap@ve proton treatments with fast-‐replanning
• On-‐line verifica@on of p pencil beams using vertex tracking
• Post-‐treatment dose verifica@on
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011
What have we learned?
• pCT is an interes@ng and challenging project that requires input/collabora@on from/with many scien@fic different fields
• Interest in the physics community is quite large, but there is s@ll skep@cism in the medical community
• Proof of principle with current Phase I prototype (beyond simula@on) will be crucial for clinical acceptance
• Besides usefulness for proton therapy, low-‐dose aspect, faithful reproduc@on of density, and freedom from ar@facts should be inves@gated & stressed
R Schulte, Proton Computed Tomography, UCSC Physics Colloquium, Nov 3, 2011