Automated Software Framework for Voxelized Absorbed Dose Estimation in

Radionuclide Therapy P Jackson, PhD; J Beauregard, MD, MSc; T Kron, PhD, DIPLPHYS; M S Hofman, MBBS; R Hicks, MBBS

price.jackson@petermac.org

Centre for Molecular Imaging

Peter MacCallum Cancer Centre

East Melbourne, Australia

Purpose:

• Accurately map distribution of radiation dose to healthy & target tissues in Peptide Receptor Radionuclide Therapy (PRRT)– Treatment given over 4 or more

cycles– Opportunity to adjust prescription

based on absorbed dose estimate?

Small volume disease, limited

absorption in lesion, high

SUV in kidneys & spleen

Bulky & highly octreotate-avid

liver lesions, limited uptake

observed in healthy organs

Internal Emitter Dosimetry• Radionuclide Dosimetry as function of:

– Pharmacokinetics• Biological half-life, uptake & clearance (multiple phases)

– Physical half-life– Emission particles

• Type & energy– Absorbing medium

• Generally soft-tissue (high-dose regions)

• MIRD: OLINDA/EXM (Gold Standard)– Male/Female/child MIRD phantoms with customizable organ

mass– Absorbed dose values accurate when considered over large

population• Difficult to account for absorbed dose to tumour/target

volume• Not intended for individualised dosimetry

“NOTE: This code gives doses for stylized models of average individuals -

results should be applied with caution to specific subjects.”

Not an issue for short-range beta emitters, though• Significant Manual input• No accounting for sub-organ kinetics (renal pelvis vs.

cortex)

Voxelized Registration and Kinetics (VRAK)• In-house dosimetry protocol based on

serial SPECT/CT:– Deformable Registration– Automatic Kinetics fitting & activity

integration– Voxel S-Values (EGSnrc)– Output: dicom dose & cumulated

activity volumes, data file with per voxel kinetics parameters

– All scripted• Called from command-line w/text input

file• Python-based (pydicom, numpy)

– Open-source dependencies– Cross-platform (Win, Mac, Unix)

4-Hour

Fused SPECT/CT

24-Hour

Fused SPECT/CT

72-Hour

Fused SPECT/CT

Image Registration

4-Hour CT

Fused to

4-, 24- & 72-hour

SPECT

Voxelized Kinetics

4-Hour CT

Fused to

Cumulated Activity

Image

(Bq*Hr/mL)

Voxel S-Value & Gamma

Convolution Kernel

4-Hour CT

Fused to

Absorbed Dose

Image

(mGy)

Curve Fitting

parameters per

voxel

Visual Analysis

Video:

Interpolated activity &

Cumulated activity

Quantitative SPECT CT:

• 177Lu: 112 keV (6.2%) & 208.4 keV (10.4%) γ’s, 497 keVmax β- (100%)

• 3 Quantitative SPECT/CT series– 4, 24, 72 hours Post-injection– SPECT quantitation previously

calibrated*• 177Lu-specific attenuation & dead-time

correction– 2 couch position (chest+abdomen)

• Full datasets for 28 LuTate Rx’s (18 different patients) used for validation

*Beauregard, J.-M. et al., 2011. Quantitative (177)Lu SPECT (QSPECT) imaging using a

commercially available SPECT/CT system. Cancer imaging: the official publication of the

International Cancer Imaging Society, 11, pp.56-66.

Image Registration : SPECT/CT• Sequential Rigid & Deformable

– Elastix* (ITK-based, adjustable parameters, scriptable)

• Multi-resolution B-Spline Deformation• Mutual Information Metric (80%) +

deformation penalty (20%)• Robust, few ‘optical flow’-type artifacts

– Co-register Anatomical (CT) volumes• Warp functional (SPECT) images

– Inspect registered SPECTs• Apply translation where necessary

– Output: 3 SPECT volumes aligned to 4-hour CT (fixed image) array

space• Same resolution & origin (Image Position

Patient

* Klein, S., Staring, M. & Pluim, J.P.W., 2007. Evaluation of optimization methods for

nonrigid medical image registration using mutual information and B-splines. IEEE

transactions on image processing : a publication of the IEEE Signal Processing

Society, 16(12), pp.2879-90.

4-hour 24-Hour Difference/Overlay

Post-registration SPECT/CTs:

4-hour: Red

24-hour: Green

Aligned Images Voxelized Kinetics• 3 Co-registered SPECT Series 1 Volume

(cumulated activity)– Read Time P.I. From DICOM header– Decay-correct activity values

• Voxel-by-voxel– Analytical fit of 3-phase exponential

pharmacokinetics• 1 Uptake, 2 Clearance

– -A1=A2+A3 (C=0 at t=0)– K1>k2>k3

• Ignore values ~0 (out of patient or not relevant for dosimetry)

• Adjust unrealistic, noisy values• Calculate for slope (as exponential) between

measurements– Solve k3, A3– Solve k2, A2– Solve k1– Limit Rate constants (k) to realistic range

where necessary• Weighted for final time point• Single-threaded, but efficient (50-100x

improvement over iterative curve-fitting routine: Scipy.Optimize)

• 25M Voxels/Hr– 2-Position Chest/Abdo SPECT: 1.5 Hours

tktktk eAeAeAC *3

*2

*1

321 ***

4 Hr SPECT 24 Hr SPECT 72 Hr SPECT

Cumulated Activity

Kinetics Processing• 3x SPECT Array of kinetics parameters

(A1, k1, A2, k2, A3, k3): [x,y,z,6]• Integrate decay-adjusted curves for

disintegrations per unit volume– Output: Cumulated Activity (Bq*Hrs)/(mL)– Save as dicom (.dcm) or ITK (.mhd/.raw) file

• Visual Output– Can be use to create image sequence of

uptake & clearance• Arguments: slice #,

projection (axial, sagittal, coronal), time window, # of frames

– Save frames at times t• Interpolated Activity• Interpolated cumulated activity

– Informative evaluation of relative uptake & clearance

• Contribution of activity at time t to total # of disintegrations

Activity t=0-100Hrs Cumulated Activity

(Bq/mL) (Bq*Hr/mL)

Dose CalculationCumulated Activity Absorbed

Dose– EGSnrc Simulation for beta- and

photon components of 177LuDecay

• All tissue assumed to be H2O equivalent• Voxel S-Values for beta- component

– Local energy deposition (only voxel self-dose)

– Agreement with publishedS-values*

• Long-range Gamma Voxel Dose Kernel– Low-resolution (high-efficiency)

» 1.6*1.6*2.0 cm Voxels– Kernel calculated by dosxyz (EGSnrc)– Convolved through activity array

*Lanconelli, N. et al., 2012. A free database of radionuclide voxel S values for the dosimetry

of nonuniform activity distributions. Physics in medicine and biology, 57(2), pp.517-33.

Radiation Absorbed Dose (point spread function, in H20)

1.00E-19

1.00E-17

1.00E-15

1.00E-13

1.00E-11

1.00E-09

1.00E-02 1.00E-01 1.00E+00 1.00E+01 1.00E+02

Radial Distance (cm)

Abs

orbe

d D

ose

(Gy/

deca

y)

Beta DoseGamma DoseTotal Dose

Point Spread Function

EDKnrc

Sphere

Model

VSV Calculation

dosxyz (EGSnrc)

Beta- Range vs. SPECT Resolution• Range of beta electron transport shorter than range of

SPECT Partial Volume effect– SPECT Partial Volume effect for point source:

• Gaussian spread w/FWHM 7-15 mm*– Absorbed Dose range for 177Lu beta-: 1-2 mm– Activity heterogeneity clearly evident in PET

• \ Beta- energy from apparent activity (as seen on SPECT) considered to be deposited locally (in same voxel)

177Lu-Octreotate SPECT Imaging

24 Hrs PI

68Ga-Octreotate PET Imaging

1.1 Hrs PI

* Gear, J. et al., 2011. Monte Carlo

verification of polymer gel dosimetry applied

to radionuclide therapy: a phantom study.

Physics in Medicine and Biology, 56, p.7273.

Gamma Dose• Small proportion of total dose

– ~5-20% depending on region• Greater Range

– Soft dose gradient• Dose Kernel from DOSxyz

– Low-resolution (voxels 1.6*1.6*2.0 cm3)

– Long range (21 cm max)• In processing:

– Convolve through activity array– Combine with beta component for total

absorbed dose• Output can be written as dose volume

from beta-, gamma, combined

– maintains alignment to CT dataBeta

- dose

(87%)

Gamma dose

(13%)

Gamma Kernel (dosxyz)

Convolved Cumulated Activity Array

Validation Methods• Coregister Serial SPECT/CT Volumes• Segment fixed CT volume in each

study• Lower Large Intestine, Small intestine,

Stomach, Upper Large Intestine, Heart , Kidneys, Liver, Lungs, Muscle, Pancreas, Marrow, Spleen, Bladder, Lesion

• OLINDA Analysis– Apply segmentation to serial SPECT

scans– Input mean organ activity values– Compute mean organ dose

• VRAK Analysis– Apply segmentation to dose volume– Report Mean organ dose

Results: Cumulated Activity & Organ Doseas compared to OLINDA

*Compartmental Organs in MIRD model (separate source & target regions; contents + wall)

At Risk organs (Somatostatin Analog PRRT)

Results: Cumulated Activity & Organ Doseas compared to OLINDA

• Estimate of decays and total dose to kidneys, liver & spleen in the range of 5% with respect to conventional technique

• Segmenting both contents and wall of compartmental organs overestimates dose by 30-100%

– Greatest discrepancy in bladder– Similar effect observed in marrow

• Comparable estimate of cumulated activity, but higher dose in segmented volume

• Geometric effect accounted for by organ S-Value– Selective segmentation required

• But susceptible to partial volume effect

Compartmental Organ Model

Source Volume

Tissue (Bladder Wall)

MIRD Bladder:

Ellipsoid Source (Radius x,y,z: 4.71, 3.21, 3.21 cm)

Outer Wall (Tissue) 2.5 mm thickness

Results: Organ S-Value

• VRAK beta- VSV + gamma kernel closely matches MIRD S-values for solid organs

– Common at risk organs in PRRT– Lesions are solid volumes too

• Mean Lesion Dose (between all studies): 24.5 ± 9.7 Gy

– ~10:1 ratio lesion to kidney dose (10.1±6.0)• Fraction of total patient dose from beta-

emission: 87.1±0.75%• MIRD model better suited to compartmental

organs– But MIRD schema does not account for self-

dose to GI wall, only from contents• Relevant in somatostatin analog therapy

Discussion• Efficient, automated tool for PRRT dosimetry using

serial functional images– Close agreement with OLINDA in both kinetics and

dose estimates at organ level for 177Lu-octreotate dataset

– Smooth, predictable kinetics estimation• No artifacts or noise from processing• Visually clear dosimetry data that can be overlaid with CT

volume– Non-specific in design: can be applied to other

isotopes, quantitative imaging (PET)• Need to know: physical half-life, isotope VSV & Gamma

Kernel (dosxyz)– Personalized dosimetry data from initial cycle may be

used to inform subsequent therapies• Radionuclide dosage• Extended renal-protective measures• Monitor relevant biochemical markers for at-risk organs

HU

mGy

4-Hour CT

VRAK Dose

Discussion• Limitations:

– Homogenous tissue (H2O) assumed– Resolution of SPECT (1-2 cm)

• Detection of highly heterogeneous uptake?• Disintegrations in tumour spread across larger

volume– Reduces estimate of absorbed dose in lesion

– Registration accuracy varies• In range of 5 mm for most organs

– Other algorithms (Demons, etc) can be more precise, but prone to artifacts

• Inconsistent breath hold can shift activity near diaphragm (liver & metastases, spleen)

– Kinetics Fitting may overestimate cumulated activity to organs with transient uptake at early time point (bladder, bowel)

• Not observed in critical organs, tumour volumes

Anatomical Volume

Approx. Margin of SPECT volume

Project Homepage• VRAK: http://code.google.com/p/vrak-dosimetry/

– Dependencies:• Elastix (http://elastix.isi.uu.nl/about.php)• Python 2.7 (http://www.python.org/download/releases/2.7.3/)• Pydicom (0.9.7) (http://code.google.com/p/pydicom/downloads/list)• Numpy (1.6.2) (http://sourceforge.net/projects/numpy/files/)• For Video:

– ffmpeg (compiled from source) (http://ffmpeg.org/download.html)– PIL (http://www.pythonware.com/products/pil/)– Matplotlib (1.1.1) (

http://sourceforge.net/projects/matplotlib/files/matplotlib/matplotlib-1.1.1/)– Recommended Viewer: 3D Slicer (http://www.slicer.org/)

References• Beauregard, J.-M. et al., 2011. Quantitative (177)Lu SPECT (QSPECT) imaging using a commercially available

SPECT/CT system. Cancer imaging: the official publication of the International Cancer Imaging Society , 11, pp.56-66.

• Lanconelli, N. et al., 2012. A free database of radionuclide voxel S values for the dosimetry of nonuniform activity distributions. Physics in medicine and biology, 57(2), pp.517-33.

• ICRP, 1983. Radionuclide Transformations: Energy and Intensity of Emissions. In ICRP Publication 38. Pergamon Press.

• Kawrakow, I. & Walters, B.R.B., 2006. Efficient photon beam dose calculations using DOSXYZnrc with BEAMnrc. Medical Physics, 33(8), p.3046.

• Klein, S., Staring, M. & Pluim, J.P.W., 2007. Evaluation of optimization methods for nonrigid medical image registration using mutual information and B-splines. IEEE transactions on image processing: a publication of the IEEE Signal Processing Society, 16(12), pp.2879-90.

• Stabin, M.G., Sparks, R.B. & Crowe, E., 2005. OLINDA/EXM: the second-generation personal computer software for internal dose assessment in nuclear medicine. Journal of nuclear medicine: official publication, Society of Nuclear Medicine, 46(6), pp.1023-7.

• Gear, J. et al., 2011. Monte Carlo verification of polymer gel dosimetry applied to radionuclide therapy: a phantom study. Physics in Medicine and Biology, 56, p.7273.

• Eckerman, K. et al., 1994. Availability of nuclear decay data in electronic form, including beta spectra not previously published. Health physics, 67(4), pp.338-345.