Development of a Head Scanner for Proton CT Hartmut F.-W. - - PowerPoint PPT Presentation

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 1

Development of a Head Scanner for Proton CT

Hartmut F.-W. Sadrozinski

• R. P. Johnson, S. Macafee, A. Plumb, H. F.-W. Sadrozinski, D. Steinberg, A. Zatserklanyi

SCIPP, UC Santa Cruz, CA 95064 USA

• V. Bashkirov, F. Hurley, R. Schulte

Loma Linda University Medical Center, CA 92354 USA

• K. Schubert, M. Witt

CSU San Bernardino, San Bernardino, CA 92407, USA

• S. Penfold

CMR, Univ. of Wollongong NSW 2522, Australia

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 2

Large-scale Imaging with Silicon Sensors

1 10 100 10 100 1000

Stopping Power for Protons

Bone Muscle H2O Fat

dE/dl [MeV/cm] Proton Energy E [MeV]

0.01 1 100 10

4

1 10 100 1000

X-Ray Absorption Coefficient

Bone Muscle H2O Fat

 X-Ray Energy [keV] [1/cm]

Attenuation of Photons, Z N(x) = Noe-  x Energy Loss of Protons, r

NIST Data

dx dE dl dE r 

Measure statistical process of X-ray removal Measure energy loss on individual protons

 

    l dx dE dx dx dE E r

Bethe-Bloch

keV 300mu 6 cm 1 1.000 1.000 10 0.165 1.000 100 0.006 0.699 1000 0.002 0.381

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 3

Alderson Head Phantom

Range Uncertainties (measured with PTR) > 5 mm > 10 mm > 15 mm

Schneider U. (1994), “Proton radiography as a tool for quality control in proton therapy,” Med Phys. 22, 353.

RSP H

Proton CT Basics

Proton therapy and treatment planning requires the knowledge of the stopping power in the patient, so that the Bragg peak can be located within the tumor. X-ray CT has been shown to give insufficiently accurate stopping power (S.P.) maps in complicated phantoms or from uncertainty in converting Hounsfield values to S.P. The goal of Proton CT is to reconstruct a 3D map of the stopping power within the patient with as fine a voxel size as practical at a minimum dose, using protons (instead of x-rays) in transmission. In a rotational scan the integrated stopping power is determined for every view by a measurement of the energy loss.

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 4

pCT Challenge #1: Multiple Coulomb Scattering

D C Williams Phys. Med. Biol. 49 (2004) 2899–2911

The proton path inside the patient/phantom is not straight  the path of every proton before and after the phantom has to be measured and its path inside the patient reconstructed.

.5

.4

.3

.2

.1 .1 .2 .3 .4 .5 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2

D e p th in s id e A b s

• rb

e r [c m ] D isplacem ent [cm ] RMS = 490um MLP width = 380 um

.5

.4

.3

.2

.1 .1 .2 .3 .4 .5 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2

D e p th in s id e A b s

• rb

e r [c m ] D isplacem ent [cm ] RMS = 490um MLP width = 380 um

• M. Bruzzi et al IEEE Trans. Nucl. Sci.,54, 140 (2007)

From deflection and displacement, calculate the “Most Likely Path MLP” Beam test with sub-divided phantom:

MLP can be predicted with sub-mm precision using tracking detectors with ~ 80m resolution

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 5

Tracking and measuring the residual energy of every proton requires fast sensors and fast data acquisition (DAQ). Data Flow math: Assuming 100 protons / 1mm voxel and 180 views requires ~ 7*108 protons. With 10 kHz data rate, one pCT scan will take 20 hrs (requiring a very patient patient!). A scan with a proton rate of 2 MHz takes 6 min. N.B. such a scan will deliver a dose of 1.5 mGy. Image Reconstruction To reconstruct images with > 107 voxels using ~109 protons is NOT trivial. Our reconstruction code is already running on GPU’s in anticipation of the much higher data rates of the future.

pCT Challenge #1a: Proton Data rate

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 6

The proton energy loss is not fixed, but is a stochastic process. The straggling error is a function of depth, irreducible when the energy is not measured.  the straggling within the phantom limits the precision of the energy loss measurement.

Challenge #2 to pCT: Range / Energy Straggling

WEPL Resolution vs. WEPL for different Plate Thickness' (200 MeV Protons)

3 3.2 3.4 3.6 3.8 4 4.2 4.4 4.6 4.8 5 50 100 150 200 250 WEPL [mm] WEPL RMS [mm]

1 mm 3 mm 4 mm 6 mm 5 mm proj. 6 mm proj. 8 mm proj. 10 mm proj

Range straggling ~ 1% of range

~ 1mm for 100 MeV, ~ 3mm for 200 MeV

Range counter always encounters the maximum range straggling: the error is independent of the WEPL of phantom (depends on proton energy) WEPL = Water equivalent Path Length (of proton in phantom)

Geant4 Study:

Range Straggling vs. Energy

0.0 0.1 0.2 0.3 0.4 0.5 50 100 150 200 250 300 Proton Energy [MeV]

Sigma (R) [g/cm

2]

0.0% 0.5% 1.0% 1.5% 2.0%

Sigma(R)/R

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 7

Efficiency of energy measurement In addition to ionization processes described by the Bethe-Bloch equation, protons undergo processes which remove protons from the peak in the energy spectra useful for the energy determination.

pCT Challenge #2a: useful Proton Rate

Fraction of Ionization events vs. Energy 0.0 0.2 0.4 0.6 0.8 1.0 50 100 150 200 250 Proton Energy [MeV] Fraction of Ionization events in Polystyrene CsI data Geant4 Polystyrene Geant4 BaF

Simulation and data agree well: At 200 MeV, only ~60% of the protons entering the phantom will be in the quasi-Gaussian end peak

• f the spectrum.

Because of non-Gaussian tails, the energy distributions at present are fitted only at the high side, which comes with a loss of precision. With improved modeling of the tails, this might be recoverable.

Geant4 Range Counter Data CsI Calorimeter

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 8

Instrument Solutions to the pCT Challenge:

Experiment Tracker Energy Detector Proton Energy [MeV]

TERA / CERN

• U. Amaldi et al., NIM A 629

(2011) pp 337-344

GEM ~100 m Range (3mm) + WLSF + MPPC 100 upgrade Firenze / LNS

• V. Sipala et al., IEEE NSS-MIC

2011, MIC15.S-305

SI SSD 80 m Fast crystal calorimeter + P.D. 68 LLU / UCSC / NIU

• F. Hurley et al., subm. to

MEDICAL PHYSICS

Si SSD 80 m CsI + P.D. 100 - 200 NIU / FNAL SciFi +MPCC 0.3-0.5 mm Range (3mm) + WLSF + MPCC 100 - 200 under construction LLU /UCSC Si SSD “Slim edges” 80 m Range (>3mm) + direct MPCC or Polystyrene Calorimeter + PMT 100 - 200 under construction

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 9

Imaging Results (LLU-UCSC-NIU)

• B. Colby, D. Fusi, R. Johnson, S. Kashiguine, F. Martinez-McKinney, J. Missaghian, H.

F.-W. Sadrozinski, M. Scaringella SCIPP, UC Santa Cruz, CA 95064 USA

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

Loma Linda University Medical Center, CA 92354 USA

• G. Coutrakon, B. Erdelyi, V. Rykalin

Northern Illinois University

• S. McAllister, K. Schubert

CSU San Bernardino

2003 2010

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 10

The LLU-UCSC-NIU Prototype Scanner

Optical Interface Photodiode

• R. W. Schulte, et al.,,

IEEE Trans. Nucl. Sci., 51,, pp 866, 2004.

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 11

CT Image Reconstruction

• 1. WEPL calibration and cut
• 2. Correction for overlaps in Si

tracker

• 3. Correction matrix with

Calorimeter response

• 4. Angular and spatial binning
• 5. Filtered Back Projection and

Iterative Algebraic reconstruction

• 6. MLP formalism for final

reconstruction

2.5 mm slice 0.65 mm voxels

Reality Check: We accumulated data for this reconstructed image during 4 hours at 20 kHz trigger rate. This is not acceptable for clinical applications ! Next development step: 50x faster pCT scanner

11

Material Predicted RSP RSP reconstructed from Measurement Polystyrene 1.037 1.035 Bone 1.70 1.68 Lucite 1.20 1.19 Air 0.004 0.05

air bone lucite polyst.

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 12

 Increase Size 2x : 40 cm x 10 cm  Improve data throughput 50x:

 2MHz sustained proton rate with minimal pile-up  Si sensors are intrinsically fast, built faster readout ASIC and distributed DAQ  Data stream uses local FPGA for data collection, formatting and transmission

 Improve speed of energy detector:

 CsI calorimeter replaced with faster plastic scintillator  Both range counter and range counter-calorimeter hybrid under test  Polystyrene Range Counter with direct MPPC readout looks very promising

(~3x p.e. wrt to WLSF readout?)

 Geant4 results on Range Counter with thicker tiles is intriguing

 Improve tiling of Si sensors:

 Si SSD are attractive since they have low noise at good efficiency,

an important factor in a sparse system (no redundant space points)

 “slim edges” allow tiling without overlap

LLU-UCSC-CSUSB Head Scanner

• R. Johnson, H. F.-W. Sadrozinski, D. Steinberg, A. Zatserklanyi,
• V. Bashkirov, F. Hurley, S. Penfold, R. Schulte, S. McAllister, K. Schubert

NIH Grant 1R01EB013118-01

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 13

WEPL (not Energy!) Detector Choices

Range Counter Direct MCPP readout

(signal 3-5x of 3mm+WLSF)

Range-Calorimeter Hybrid “Bulky”: 3 Polystyrene 10cmx10cmx40cm + PM

3

• 4”

PMT

~ 30 p.e.

70 plates, 4 mm, Polystyrene Scint.

Hodoscopic CsI Calorimeter P.D. Readout Too slow!

Proton path

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 14

WEPL Calibration of “Bulky”

200 MeV Protons, Polystyrene Degraders of known Water Equivalent Thickness WET

Comparison of WEPL Resolution

Range Counter WEPL RMS is constant ~ 4 mm (as expected, since range counter is dominated by straggling in phantom/degrader + range counter (expect ~ 1mm from plate thickness) “Bulky” appears to be a good choice, if spill-over can be dealt with when range is close to calorimeter interface

WEPL Error vs. Degrader Thickness

1 2 3 4 5 6 7 8 9 50 100 150 200 250 300 WET (mm) WEPL RMS (mm)

Bulky CsI Calorimeter 4mm Range Counter (MC) Bulk2'

Goal of WEPL calibration:

Determination of WEPL directly from calorimeter response, without converting first to MeV

"Bulky" WET Calibration Curve

50 100 150 200 250 300 500 1000 1500 2000 2500 3000 Response [a.u.] WET (mm)

Bulky2 Bulky 1

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 15

Silicon Tracker Improvements

Large area coverage requires tiling of sensors. Sensors have ~ 1mm inactive edges which create image artifacts.

In the present prototype tracker this is dealt with by shingling sensors, but a reconstruction nightmare. Overlapping sensors introduces additional, non-uniform energy corrections

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 16

For Tiling with no Overlap: “Slim Edges”

see M. Christophersen’s Talk

Si SSD with 900m dead edge XeF2 scribing + Cleaving + N2 PECVD with guard ring Cut within 50 m

• f Guard Ring

Guard Ring Cut (!) without guard ring

Reduce dead edge from 1mm to ~ 200 m Excellent breakdown behavior Current at 150V: ~10 nA/cm with guard ring ~100 nA/cm without guard ring

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 17

Charge Collection of “Slim Edge” HPK Sensor

GLAST2000, p-on-n; t = 400 m; L = 10 cm; Pitch: 228 m; # of strips:8

Treatment:

Laser scribed w 10% intensity Cleaved with tweezers Oxygen PECVD on Sidewall Cut ~ 100 mm from guard ring

1) Measure i-V for entire sensor and for every strip: Is any current leaking into the active region? 2) Investigation of the Charge collection pre- and post- cutting using the AliBaVa analog readout system

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 18

Currents on Individual Strips

The currents on the cut sensor are 1000x larger wrt un-cut. The currents on individual strips are measured through the bias resistor voltage measurements. They are generally consistent for both cut and un-cut

• devices. We do not see an abnormal behavior for the edge strips.

Device i-V uncut-cut Strip i-V uncut-cut

i-V of Edge Strips for Cut vs Uncut

0.E+00 2.E-10 4.E-10 6.E-10 8.E-10 50 100 150 200 250 300 350 Vbias [V] Current [A]

Uncut Strip 1 Current Uncut Strip 8 Current Cut Strip 1 Current Cut Strip 8 Current Uncut Strip 5 Cut strip 5

Cut / Uncut Skinny: IV Curve 1.E-09 1.E-08 1.E-07 1.E-06 1.E-05 1.E-04 100 200 300 400 Vbias [V] Total Current [A]

Cut Skinny Uncut Skinny IV

P-on-n HPK (GLAST), laser scribed, PECVD Oxygen, 96m from guard

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 19

Outer Strip Signal pre-cut / post-cut

Strip # 203 is next to the cut edge Median does not change by more than 5% from before to after cut

Outer-Inner Signal, 150 V Bias

10 20 30 40 50 60 70 80 100 200 300

Signal [ADC value] #

100 200 300 400 500 600

• Ch. 202

Ch 203 cut Inner cut

Outer Noise

1 10 100 1000 10000

• 30
• 20
• 10

10 20 30 Channel # # of eents

• Ch. 202
• Ch. 203 cut

150 V bias, 90Sr source Cutting does not change the noise in the adjacent outer strip Tail in pulse height distribution from partially measured tracks close to the bias ring is unchanged before/after cut

Data taken and analyzed by R. Mori (Florence U.) & M. Cartiglia (Milano H.S.)

Hartmut F.-W. Sadrozinski: pCT HSTD8 Dec. 2011 20

Conclusions



Proton CT has come a long way since I first talked about it at the 2002 IEEE NSS-MIC Symposium in Norfolk, VA.



We see very different approaches on instruments, motivated in part by a technology transfer from HEP/Space. This has come with severe limitations (proton rate!).



Using our prototype scanner, we are starting to reconstruct very clear and sophisticated radiographs AND CT images, and are actively improving reconstruction algorithms.



We are now arriving at a new phase in pCT: we have a dedicated detector development, with focus on speeding up the data taking to be useful in clinical applications, and optimizing the detector systems based on lessons learned.



End-to-end simulation of the instrument has been essential for our understanding of the requirements and proper choice of the technical solution, yet many lessons were learned during operation of the scanners



Next (big) step: clinical application. Ongoing and unwavering support by Prof. James M. Slater (LLUMC) made this project possible. We acknowledge support from the US National Institute of Health under the grant NIH Grant 1R01EB013118-01