Radiation Protection and Dosimetry in Medicine ...cmpwg.ans.org/mc2015/Vaz.pdf · Radiation Protection and Dosimetry in Medicine Computational Issues: An Overview Pedro Vaz, Wayne

Radiation Protection and Dosimetry in Medicine Computational Issues: An Overview

Pedro Vaz, Wayne Newhauser and Bernadette Kirk

@ Workshop on Computational and Mathematical Challenges in Particle Therapy Nashville, TN, 19th April 2015

Outline

• The system of Radiological Protection – where are we and where do we go from here ?

• Computational issues & challenges in Medicine

Monte Carlo Treatment Planning

Grand Challenges in High Performance Computing for Medicine Some Examples from Particle Therapy

• Computational issues & challenges in (Radiation) Biology and Biosciences

Nanodosimetry and track structure simulation

• Assessment of the computational state-of-the-art

• Available computing power vs complexity (of the problems to be modelled)

• Conclusions

The system of Radiological Protection

The Future System of Radiological Protection

“Radiation protection standards rely on current knowledge of the risks from radiation exposure. Any over-, or under-, estimation of these risks could lead either to unnecessary restriction or to a lower level of health protection than intended.”

Sou

rce:

HLE

G r

epo

rt (

20

08

)

Computation of Effective Dose

Sou

rce:

IC

RP-

103

repo

rt (

20

07

)

ICRP reference voxel phantoms (report ICRP-110)

Adult Male phantom

Steps towards patient-tailored therapies (and diagnostic),

& “individual” risk estimation

Computational Phantoms for medical dosimetry (1)

• Three different formats for computational anatomic phantoms:

stylized (or mathematical)

• Flexible, allowing changes in organ size, body shape, and extremity positioning, but generally deficient with respect to anatomic realism

voxel (or tomographic)

• Three dimensional array of voxels, each with a unique organ identity, elemental composition, and density. Very difficult to alter to represent the body morphometry

hybrid

• Based upon NURBS and/or polygon mesh surfaces. Preserve both the anatomic realism of voxel phantoms and the mathematical flexibility of stylized phantoms

Computational Phantoms for medical dosimetry (2)

• Phantom Morphometric Categories:

– Reference

• reference phantom defined typically as an individual at 50th height/weight percentile in a given human population ICRP 110

– Patient-specific

• Uniquely match the body morphometry and organ anatomy of an individual medical patient

– Patient-Dependent Phantoms

• match patient to phantom using a large library of phantoms covering a broad range of body shapes and sizes

OECD/NEA + UF study Changing the paradigm…?

Effective dose

Effective individual risk

?

Reference individual

Adult male, Adult female, Pediatric male,

Pediatric female

ICRP phantom library

Broad range of body sizes

(height/ weight)

Patient dependent phantoms

From Ted Lazo (NEA) @ Article 31 (2014) meeting

Use of non-reference phantoms Accuracy of dose calculations (1)

From Ted Lazo (NEA) @ Article 31 (2014) meeting

Use of non-reference phantoms Accuracy of dose calculations (2)

From Ted Lazo (NEA) @ Article 31 (2014) meeting

Computational issues in radiation therapy

Uncertainties in dose delivery

Monte Carlo Treatment Planning (MCTP) Rationale

Section II.A.1 Slopes of dose-effect curves “At this point, a 5% change in dose may result in a 10% to 20% change in tumor control probability at a TCP of 50%. Similarly, a 5% change in dose may result in a 20% to 30% impact on complication rates in normal tissues.”

Section II.A. Required Dose Accuracy

Section II.A.2 The level of dose differences that can be detected clinically “Thus it could be concluded that at least a 7% difference in dose delivered is manifested in the patient’s response to radiation treatment and is detectable clinically by a radiation oncologist.”

Use of Monte Carlo in Radiotherapy Treatment Planning

• Nowadays, most manufacturers deliver Monte Carlo Treatment Planning Systems – however their validation against state-of-the-art Monte Carlo programs still far from achieved

• Monte Carlo programs for Clinical Dose Calculation: – PEREGRINE used in CORVUS inverse TPS (NOMOS, Pittsburgh, PA);

– VMC++ used in Oncentra TPS (Nucletron B.V., Veenendaal, The Netherlands)

– Macro MC used in Eclipse TPS (Varian Medical Systems Inc., Palo Alto, CA)

– DPM used in Pinnacle TPS (Philips Radiation Oncology Systems, Madison, WI);

– PENFAST(PENELOPE) used in ISOgray TPS (DOSIsoft, Cachan, France);

– XVMC used in iPlan (BrainLAB AG, Feldkirchen, Germany), in XiO and Monaco (CMS Inc, StLouis, MO) and PrecisePLAN (Elekta Inc., Norcross, GA) TPS.

Commercial Treatment Planning

algorithms

Monte Carlo Treatment Planning

systems

Real Time Treatment Planning

(using MC ?)

Wayne Newhauser

Grand Challenges in High Performance Computing

for Medicine: Some Examples from Particle Therapy

Some Grand Challenges

Optimize outcomes

Model how physical dose regulates biologic outcomes

Visualize all the radiation exposure (dose and quality) to all the tissues

Model interaction of radiation, drugs,

Personalize medicine (e.g., model inter-patient variations)

19

Newhauser and Durante (Nature Reviews Cancer 2011)

Radiation Exposure to Patients

20

Therapeutic

Leakage (Challenging)

Scatter (Challenging)

(Easy)

21

Calculate Radiation Exposures

Proton absorbed dose

(easy)

Neutron absorbed dose

(challenging)

Newhauser et al (2008; 2009), Taddei et al (2009, 2010); Zhang et al (2014); Giebeler et al (2013); Perez-Andujar et al (2012)

1 day w/ 1 CPU

3 wks w/ 1072 CPUs

From Newhauser et al, PMB (2009) and Miralbell et al., IJROBP (2002). See related studies by Taddei et al PMB (2009), Brodin et al Acta Oncologica (2011), Zhang et al (2014)

Photon IMRT (15 MV, 9 field)

Photon CRT (6 MV, 1 field)

Protons (SOBP, 1 field)

Risk: 55% 31% 4-5% Rel risk: 12 7 1

Compare Treatment Strategies: Predicted Doses and Risks of SMN after Photon vs Proton Therapies

Routine Calculation of All the Dose

to All the Tissues

23

Sagittal equivalent dose planes overlaying a thoracic

CT image of the HL patient showing (a) proton

equivalent dose and (b) combined proton and neutron

equivalent dose. Equivalent dose values are

percentages of the prescribed target equivalent dose,

i.e., 36 Sv. The mediastinal tumor and healthy

thyroid are contoured in black.

Eley, Newhauser, Homann, Howell, Schneider, Durante Bert.

Cancers 2015, 7, 427-438

Easy

Challenging

Moving Organs and Beam

24

Easy Challenging

Challenging

Eley, Newhauser, Luchtenborg, Graeff, Bert. Phys. Med. Biol. 59 (2014) 3431–3452

“4D optimization of scanned ion beam tracking therapy for moving tumors”

From Newhauser and Durante, Nature Rev Ca, 2011

Radiation Absorbed Dose

Risk of SMN Incidence

Risk of SMN Mortality

Visualize Dose & Risk Is “dose” enough? Absorbed dose? Equivalent dose? Effective dose? Integral dose? Ambient dose equivalent? Is “risk” enough? Incidence? Mortality? Absolute? Relative? Timepoint?

Algorithmically Optimize Outcomes

26

Rechner, Eley, Howell, Zhang, Mirkovic, Newhauser, Risk-optimized proton therapy to minimize radiogenic second cancers (in review)

Axial slice showing risk-optimized proton therapy (ROPT) treatment plans with and without DVH constraints applied during the planning process

Algorithmically Optimize Outcomes

27

Predicted excess relative risk (ERR) versus beam angle (θ) for second cancer in the bladder and rectum using the linear-non-threshold risk model

Rechner, Eley, Howell, Zhang, Mirkovic, Newhauser, Risk-optimized proton therapy to minimize radiogenic second cancers (in review)

Technical Computing Challenges

Reproducibility: Osterwiel et al., Science, 325 1622 (2009)

Scalability (software development and parallelization)

Heat generation and removal (operating costs)

Bandwidth associated with input and output of data

Latency as information travels between parts of a supercomputer

29

Summary: Computing Aspects

Now feasible to reconstruct whole body radiation doses and risks of second cancer

Despite rapid progress in hardware and software, still many large gaps in knowledge

Increasingly personalized medicine will require huge increase in computing in radiotherapy

Computational issues & challenges in (Radiation) Biology and Biosciences

Nanodosimetry and track structure simulation

Modelling Radiation Biology

From: Carmen Villagrasa (IRSN), EURADOS Winter school : “Status and Future Perspectives of Computational Micro- and Nanodosimetry”

Issues (1)

De Broglie wavelength for a 10 eV electron: =h

mv= 0.39 nm

DNA transverse dimension: 2-3 nm

Inelastic cross-sections of low energy electrons (and other particles) MUST

Some data for water. How about other “materials” ?

Tracking particles down to the few eV energy range

For each particle fully simulate the ionization pattern

track structure Monte Carlo simulation !

Issues (2)

• Track structure Monte Carlo simulation programs:

Perform the transport of particles simulating each particle´s interaction

Time consuming

Limited to microscopic spatial dimensions

Utilize DNA models of different complexity

Some simulate processes such as DNA damage repair

Insufficient benchmarking and validation?

• Primary target for radiation-induced damage – DNA molecule

– single and clustered damage

From: Fundamentals of micro and nanodosimetry, Hans Rabus, 2011

Nanodosimetry and biological effectiveness

Nanodosimetry Track structure

Particle track-structure analysis

From: Fundamentals of micro and nanodosimetry, Hans Rabus, 2011

Nanodosimetry Cluster size distributions

• Ionization cluster size distributions ()

– Number of ionizations produced by a single-particle track in the DNA segment

• Diffusion and recombination of radiation-induced water radicals

Track structure / nanodosimetry Monte Carlo simulations

+ GEANT4-DNA (http://geant4-dna.org)

Fro

m:

H. N

ikjo

o, R

ad. M

eas

. (2

00

6)

Assessment of the computational state-of-the-art

Shielding design Accuracy of deep penetration simulations

10-14

10-13

10-12

10-11

10-10

10-9

10-8

10 100 1000

Fig. 8 Neutron spectra inside iron for 1 GeV neutrons.

GEANT-4(4m)(SATIF-10)

PHITS(4m,R=3m)(SATIF-8)

FLUKA(4m)(SATIF-10)

ROZ-6.6(4m)(SATIF-8)

MARS(4m)(SATIF-8)

HETC-3STEP(4m,SATIF-6)Neu

tron

s/M

eV

/cm

2 p

er

n/c

m2

Neutron Energy (MeV)

Results presented at SATIF-10

From H. Hirayama (KEK) @ SATIF-12 Meeting

Shape of FLUKA is different from others.

MCNPX larger than others.

10-5

10-4

10-3

10 100 1000

Al, 1GeV proton at 15 degrees

FLUKA 2011MARS 1514 (LAQGSM)PHITSMCNPX Version2.7Geant4 V10.00p01FLUKA 2011 2b5

Neu

tro

ns/M

eV

/sr

Energy (MeV)

From H. Hirayama (KEK) @ SATIF-12 Meeting

10-2

10-1

0 20 40 60 80 100 120 140 160

Al, 1 GeV proton

FLUKA 2011MARS 1514(LAQGSM)PHITSGeant4 V10.00p01 Version2.7FLUKA 2011 2b5

Neu

tro

n flu

en

ce a

bo

ve

20

MeV

(neu

tro

ns/s

r)

Angle (Degrees)

From H. Hirayama (KEK) @ SATIF-12 Meeting

10-2

10-1

100

0 20 40 60 80 100 120 140 160

Al, 10GeV proton

FLUKA 2011MARS 1514(LAQGSM)PHITSGeant4 V10.00p01MCNPX Version2.7FLUKA 2011 2b5

Ne

utr

on

flu

en

ce

ab

ove

20

Me

V(n

eu

tro

ns/s

r)

Angle (Degrees)

Differences between

code become larger at

large angle.

From H. Hirayama (KEK) @ SATIF-12 Meeting

10-1

100

0 20 40 60 80 100 120 140 160

Al, 100GeV proton

FLUKA 2011MARS 1514(LAQGSM)PHITSGeant4 V10.00p01MCNPX Version2.7FLUKA 2011 2b5

Ne

utr

on

flu

en

ce

ab

ove

20

Me

V(n

eu

tro

ns/s

r)

Angle (Degrees)

Differences between

code become larger at

large angle.

MCNPX results are larger and PHITS

results are smaller than others.

From H. Hirayama (KEK) @ SATIF-12 Meeting

Available computing power vs complexity (of the problems to be modelled

Computing Power vs. Monte Carlo simulations

http://www.intel.com/technology/mooreslaw/index.htm

100

0+ line

s of cod

ing

100000+

lines of

coding

500000+

lines of

coding

Computing Power vs. Complexity

From: Bernie Kirk

Co

mp

lexit

y

of

Pro

ble

m

Computers

PC

Clusters of PCs

Highly parallel computers

M

E

M

O

R

Y

Megabyte/Megavoxel 220

Gigabyte/Gigavoxel 230

Terabyte/Teravoxel 240

Petabyte/Petavoxel 250

ICRP

MIRD

Rigid 3-D

Moving 4-D

Computational Challenge for Radiation Therapy and Imaging

From: Bernie Kirk, Source: George Xu

Outlook and Conclusions (1)

• Advances in Medicine and in the Biosciences (at large) impose:

– Specific computational requirements concerning the modelling and simulation of the interaction of radiation with matter

– Challenging approaches to the understanding of the biological effects of ionizing radiation Microdosimetry and Nanodosimetry

• The way forward encompasses:

– Individual dose and risk assessment in radiation therapy

– Tailoring treatments for the individual patient

– Real Time Tretatment Planning Systems

Outlook and Conclusions (2)

• Applications driving computational requirements (Monte Carlo, deterministic and hybrid methods and programs): – Medical uses of ionizing radiation (diagnostic, therapy, nuclear

medicine and interventional procedures) – Emerging and innovative nuclear technology systems

• Future evolution calls for:

– Effective hybrid methods – More and better cross-section data – Efficient tools for:

• Sensitivity/uncertainty analysis, • Variance reduction, • Tallying, • Input and output

– Full 3D and time-dependent capabilities and calculations