MR-only Radiotherapy:current status and perspectives

MR-only Radiotherapycurrent status & perspectives

Matteo Maspero [email protected] researcher

UMC Utrecht Center for Image Sciences

Dep. Radiotherapy 12th October 2018

necessary to ensure a safe clinical implementation of an MR-only pathway into the clinic.

Content

1. Introduction

• Definition & requirements, rationale

2. The challenges

• MRI for treatment simulation

• Dose accuracy

• Position verification

3. World-wide diffusion: where are we standing?

4. Perspective: what is it rising at the horizon?

Definition

MR-only Radiotherapy

=

Radiotherapy based on magnetic resonance as sole or

primary imaging modality during treatment planning

“MR alone”, “sole MRI”

1. Edmund, JM, & Nyholm, T. 2017, Radiat Oncol, 12(1) https://doi.org/10.1186/s13014-016-0747-y

2. Johnstone, E et al. 2017, Int J Radiat Oncol Phys Biol 100(1) https://doi.org/10.1016/j.ijrobp.2017.08.043

3. Owragi AM et al. 2018, Phys Med Biol, 63(5) https://doi.org/10.1088/1361-6560/aaaca4

4. Wafa B, Moussaoi A. 2018 Med Tech J, 2(1) https://doi.org/10.26415/2572-004X-vol2iss1p150-178

5. Kerkmeijer LGW, et al. 2018 Clin Oncol (in press) https://doi.org/10.1016/j.clon.2018.08.009

Why do we use MRI in Radiotherapy?

• Superb soft tissue visualisation

• Multi-contrast: morphological and functional

MA Schmidt, GF Payne. 2015, Phys Med Biol, 60(22): R323–R361 https://dx.doi.org/10.1088%2F0031-9155%2F60%2F22%2FR323

Rasch, C et al. 2010 Radiat Oncol, 15(5):21

“Tumor delineation, the

weakest link in search

for accuracy”Njeh et al, Med Phys 2008

Why do we use MRI in Radiotherapy?

MA Schmidt, GF Payne. 2015, Phys Med Biol, 60(22): R323–R361 https://dx.doi.org/10.1088%2F0031-9155%2F60%2F22%2FR323

A visual demonstration…

CT vs MR: Prostate

• Prostate anatomy is not well visualized on CT and it can be difficult to

delineate prostate from rectum.

• MRI is helpful for visualization of prostate anatomy and tumor

CT MRI – T1w 3D GRE MRI – 3D bTFE fat sup

Chandarana H et al. 2018 J Magn Reson Imaging https://doi.org/10.1002/jmri.26271

CT vs MR: Head & Neck

T2N2b hypopharynx tumour

Contrast Enhanced-CT MRI – T1w Gd

Head-neck tumor are contoured on MRI and routinely fused with CT

Chandarana H et al. 2018 J Magn Reson Imaging https://doi.org/10.1002/jmri.26271

MRI superior soft tissue contrast: cervix

GTV primary tumor

rectum

bladder

CTVprimary (cervix, corpus uteri)

Pötter R, et al. 2007 Radiother Oncol. 83(2):148-55. https://doi.org/10.1016/j.radonc.2007.04.012

• MRI facilitates treatment adaptation and nodal localization

CTVnodes (path.lymph nodes)

CT vs MR: Oesphagus

CT

1.3 x 1.3 x 3 mm3 0.7 x 0.7 x 3 mm3

MR T2w TSE

van Rossum PS, et al 2013 Europ Radiol 1;23(7):1753-65 https://doi.org/10.1007/s00330-013-2773-6

MRI has the potential to bring improvement in staging, tumour

delineation and real-time guidance for RT and assessment of treatment

response

Positive node

CT MRI –T2

MRI – T1MRI – Diffusion Weighted Imaging

CT vs MRI Pancreas

2. Planning 3. Treatment1. Imaging

A Radiotherapy workflow…

Primary modality

Provide information on

the attenuation of the

radiation in the body

Thanks to its excellent

soft tissue contrast

facilitate tumour

localisation

2. Planning 3. Treatment1. Imaging

A Radiotherapy workflow…

Image

fusion

Reference

Images

Plan &

DoseDelineation

Linac

Position

Verification

Repeated by the nr

of fractions

2. Planning 3. Treatment1. Imaging

An MR-only Radiotherapy workflow

Image

fusion

Reference

Images

Plan &

DoseDelineation

Linac

Position

Verification

2. Planning

Plan &

Dose

3. Treatment1. Imaging

The MR-only Radiotherapy workflow

Reference

Images

Plan &

DoseDelineation

Linac

Position

Verification

Rationale

• Fusion of CT and MRI is never perfect

• Remove systematic error during planning• Prostate: “The systematic uncertainties, affecting all treatment

fractions, are reduced from 3-4 mm (1Sd) with a CT based workflow

to 2-3 mm with a MR based workflow.”*

• Reduce nr CT scans along with

• exposure to ionising radiation

• patient discomfort

• time between imaging and planning

• overall cost of the treatment

*T Nyholm et al. 2009, Radiat Oncol, 17;4:54 https://doi.org/10.1186/1748-717X-4-54 .

Treated first patient in February 2014

February 2017: FDA clearance for Linac

Additional rationale: MR-guided RT

“Seeing what you treat”

1 Mutic S, Dempsey JF. Semin Radiat Oncol. 2014 https://doi.org/10.1016/j.semradonc.2014.02.0082 Lagendijk JJ, Raaymakers BW, van Vulpen M. Semin Radiat Oncol. 2014 Jul;24(3):207-9 https://doi.org/10.1016/j.semradonc.2014.02.009

1.5T Unity Elekta20.35T MRIdian Viewray1

Treated first patient in May 2017

June 2018 CE clearance

Intermezzo

plan

pre-beam

MRI

beam on

MRI

contourMR-sim plan

1 week per treatment session

contour

MR-only in MRgRT workflow

Courtesy: R Tijssen, UMCU

A clarifying intermezzo

MR-only

RT

MR-guided

RT

MR-based

Dose calculation &

Position Verification

PET-MRI*

Hofmann, M et al. 2009, Eur J Nucl Med Mol Imaging 36(1): 93

MR-based

Attenuation

Correction

The challenges

1) MR for treatment simulation

2) MR-based dose calculations

3) MRI as reference images for position verification

MRI for treatment

simulation1

➢ MRI is generally seen as a diagnostic tool, but..

➢ MRI in radiotherapy ≠ MRI in radiology

➢ Need to re-optimise exam card

➢ Geometric accuracy

➢ Large field of view (FOV)

➢ Hardware and software adaptation to accommodate

o RT positioning

MRI for treatment simulation

McWilliam A, et al. 2018 Clin Oncol 30(11):680-685 https://doi.org/10.1016/j.clon.2018.08.004

Wide bore MRI

Magnetom Espree

GE RT system

Philips Ingenia RT system

Canon Vantage Galan

Setup adaptation MRI: scan in treatment position

MR-RT simulator At Treatment (Linac)

• MR-compatible immobilization/set-up devices, e.g. no carbon

• Positioning lasers to record patient position

Adaptations imaging in treatment position

flat table-top

Flat table top instead of

Concave MR diagnostic table top

Coil bridge not deform body contour

A crucial aspect: Geometric Fidelity

Characterization, prediction, and correction of geometric distortion in 3T MR images

Baldwin, Lesley N. and Wachowicz, Keith and Thomas, Steven D. and Rivest, Ryan and

Fallone, B. Gino, Medical Physics, 34, 388-399 (2007)

“Armed with detailed knowledge of spatial distortion, MR images can be undistorted…”

Geometric accuracy: system imperfections

• Two system related causes

– Gradient non-linearity

– Magnetic field inhomogeneity of the magnet

• Vendor make corrections for gradient non-linearity

B0

>>>

B0

>>>

Geometric accuracy

• Residual corrections can be checked by special geometric

QA tests using dedicated phantom

• Automated procedure

Philips Ingenia RT geometric QA phantom & analysis

E.g. vendor provided geometric accuracy:

• < ± 1 mm total geometric accuracy of image data in < 20 cm Diameter Spherical Volume (DSV).

• < ± 2 mm total geometric accuracy of image data in < 40 cm DSV.

Patient induced magnetic field distortions

• Regions with transitions in magnetic susceptibility

causes magnetic field distortions.

5.8 ppm

-4.0 ppm

2.0 ppm

How to mitigate B0 distortions: high bandwidth

Which is the price? Decrease of signal-to-noise ratio

Quantifying total geometrical distortions

EP2147 R.L. Goodburn, R. Tijssen, M. Phillipens ESTRO 2018

https://doi.org/10.1016/S0167-8140(18)32456-3

Total (system + patient) distortion mapB0 map

pelvisbrain

• Philips 1.5 T MR-sim

– Patient +machine B0 distortion

– Gradient non-linearity

In 35 cm VOI, stable over time (2 yrs)

MRI for treatment

simulation1• Adaptations are necessary but possible

• Vendors have seen the potential of MRI in RT and

facilitate MR simulation

• MRI expertise is needed

MR-based dose

calculations2

Solution

Generate synthetic/pseudo CT images

MRI CT

Challenge

MRI does not provide electron density information that are necessary

to plan radiation treatment

Approaches so far proposed: in numbers

• Meeting point with MR-PET

• Voxel-based

• Atlas

• Hybrid

Brain/H&N

Prostate

TorsoPhantom

75.3%

12.4%

7.4%

Pu

bli

cati

on

s RT

MR/PET

Both

1998 2004 2006 2007 2008 2010 2011 2012 2013 2014 2015

Vandenberghe, S and Mardsen PK, 2015, Phys Med Biol 60:R115

Until 2015 ~ 50 papers (RT only)

Edmund, JM, & Nyholm, T. 2017, Radiat Oncol, 12(1) https://doi.org/10.1186/s13014-016-0747-y

sCT generation: voxel-based, the initial experiments

• Based on manual segmentation or threshold followed by bulk

assignment

• Investigate requirements/feasibility

– How many tissue classes?

– Which HU should we assign?

Eilertsen, K et al 2008, Acta Oncol 47(7) https://doi.org/10.1080/02841860802256426

Pros

• Easily implemented

Cons

• Laborious/time-expensive

methods

sCT generation: voxel-based, regression

Nyholm, T, Jonsson, J. 2014 Sem Radiat Oncol, 24(3) https://doi.org/10.1016/j.semradonc.2014.02.005

Gaussian regression to map MRI

voxel into CT voxel

• Use non-standard sequence to

facilitate the regression

Pros

• Fast, more

elegant

Cons

• Robustness can

require non-

standard MR

sequences

sCT generation: atlas

Jens Sjölund et al, 2015. Phys Med Biol, 60(2). http://dx.doi.org/10.1088/0031-9155/60/2/825

Dowling JA et al. 2012 Int J Radiat Oncol Biol Phys 83(1) https://doi.org/10.1016/j.ijrobp.2011.11.056

Pros

• Robust methods, given the a-

priori knowledge that an

atlas incorporate

Cons

• A-priori knowledge does not

deal with abnormal

anatomies

• 3D dual-echo Cartesian GRE with Dixon reconstruction

• Intensity & shape-based model for bone-only

sCT generation: hybrid tissue classification

ED Map

W

IP

F

M Kohler et al, Philips MRCAT for prostate dose calculations using only MRI data, 2015, White paper, 4522991 11581

• MRCAT, Philips Healthcare

– prostate

– CE and FDA approved on 2016-03-21 http://goo.gl/jtyX8H

• MriPlanner, Spectronic Medical AB

– prostate

– CE approved pending 2016-06-16 https://goo.gl/cY5Zrn

• Syngo.via.syntheticCT, SIEMENS

– Vendor specific platform for translational research

– brain

– CE FDA approved from 2018-01-12 https://goo.gl/37vEv6

Approaches so far proposed: commercial solutions

How to check the dose accuracy

MRI

CTsCT

Treatment Planning System

sCT

DoseCT

Dose

Rigid

Registration

Image and dose comparison

MR

-deriv

edC

T-d

eri

ved

Average CT-sCT dose differences over 14 pts

+ RTSS

MRCATCTreg

Rigid

Registration

Clinical Plan

CT

Dose

Difference

DCT – DMRCAT

-0.33 ±0.39 % in CTV

Dose accuracy of MRCAT : ~ 0.3 %

sCT generation: dose accuracy, e.g. prostate

Kerkmeijer LGW et al. 2018 Clin Oncol (in press) https://doi.org/10.1016/j.clon.2018.08.009;

Maspero M. 2018, ISNB: 978-90-393-6953-1 https://dspace.library.uu.nl/handle/1874/362987.

Dose differences < 1% between CT and MR-based dose calculations

sCT generation, the latest trend

Han X 2017 Med Phys 44(4) https://doi.org/10.1002/mp.12155 ; Nie D 2017 MICCAI;

Deep learning: U-net

Deep Learning based sCT generation for brain

• Cross validation 2 x 26 patients

• < 1 % dose difference PTV

• 1 minute to generate sCT

Deep Learning based sCT generation for pelvis

Maspero et al. 2018, Phys Med Biol 63(18):185001 https://doi.org/10.1088/1361-6560/aada6d

pix2pix

500

-500

HU

MRICT sCT

Dose differences in target/OAR < 0.5%

Pros

• Do not require non-standard

sequence

• No need of a deterministic

image processing pipeline

• Cons

• Steep learning curve to start

working with these methods

• Need of GPU

MR-based

attenuation

correction2

• For brain and prostate, sCT generation seem to be a

solved problem

• Commercial solutions are available and can facilitate

the implementation

MRI as reference for

position verification3

Image guided dose delivery

Cone beam CT, or EPID

• Patient anatomy aligned to radiation

beam by registration of daily CBCT

to planning CT = reference

• Provides translation and rotation of

patient table

MRI for position verification

• Ideally, when switching to an MR-only we would like to

maintain the current IGRT regime

– Anatomical site

– Institute

• Possibilities:

– MR as reference images

– Synthetic-CT as references

• Exemplum gratia: rectal & prostate cancer

Commercial software are optimised for CT-based

registration

Position verification for rectal cancer @ UMCU

De Boer HCJ and Heijmen BJM, 2007, Int J Radiat Oncol Biol Phys 67(5):1586-95

Match

based on

CBCT on

bones in a

clipbox

The dataset available

CT sCT CBCTs

x 5-12 per patient

• Prone to residual error

between sCT-> CT

registration

• If conducted with clinical

software: trustworthy

simulation of the clinical

situation

• Results: mean difference of

translations and rotations

over ten patients (76 CBCTs)

was <1 mm and <0.5o in all

directions

Study design

CTsCT

T,RCBCT→sCT

Registration

CBCT

T,RCBCT→CT

CBCTCBCTCBCT

T,RCBCT→sCTT,RCBCT→sCTT,RCBCT→CTT,RCBCT→CT

Maspero M, et al. Phys Imaging Radiat Oncol (7):58-64 https://doi.org/10.1016/j.phro.2018.09.002

3/4 gold Fiducial Markers

Template

Prostate

1

𝐿 = 5 mm

Ø = 1 mm

Prostate position verification @ UMC Utrecht

1 week before

planning…

OutputImage Processing

At

the L

inac

Pla

nn

ing

Im

ag

ing

Position verification

CT

Localisation

prostate

Alignment to reference

PretreatmentImage

Gold Fiducial Marker: a problem of visibility…

CT

• FM as hyperintense

signal

• Easy to localise

MRI

• FM as hypointense signal

• Easy to misclassify

MRI

CT

bSSFP fatsat Multi echo GRE

CT

Inter-observer study to locate the fiducials

• 5 RTT

• 17 patients with each 3 fiducials

• 2 MRI sequences (one locate)

Maspero M et al. 2018 Radiat Oncol 5;13(1):105. https://doi.org/10.1186/s13014-018-1029-7

Difference inter-marker distances on CT and MR

Sequence Observer Mean Median STD Range

1 0.8 0.6 0.7 [0.1, 3.1]

2 0.6 0.5 0.5 [0.0, 2.5] Single 3 0.7 0.6 0.5 [0.0, 2.1]

4 0.7 0.6 0.6 [0.1, 2.9]

5 0.7 0.5 0.6 [0.1, 2.5]

All 0.7 0.6 0.6 [0.0, 3.1]

1 0.7 0.4 0.6 [0.0, 2.7]

2 0.6 0.4 0.6 [0.0, 3.0] Multiple 3 0.7 0.5 0.7 [0.0, 2.5]

4 0.7 0.5 0.6 [0.0, 2.8]

5 0.7 0.6 0.5 [0.0, 2.5]

All 0.6 0.5 0.6 [0.0, 3.0]

Accuracy, Precision in mm

Manual MR localization as accurate as CT based localization,but 3/240 times fiducials were misclassified (3/17 pts)• We quantified risks adding redundancy to the procedure…

Maspero M et al. 2018 Radiat Oncol 5;13(1):105. https://doi.org/10.1186/s13014-018-1029-7

Manual FM localisation: a difficult case

Calcification

Fiducial

Marker

Calcification occurrence ~ 40%

Hong CG et al, Korean J Urol (2012); 53:330-334

Calcification occurrence (>2mm)

39/60 pts~ 65%Gustafsson C et al, Med Phys (2017); in press

CT bTFE FFE

Maspero M et al. 2018 Radiat Oncol 5;13(1):105. https://doi.org/10.1186/s13014-018-1029-7

Manual FM localisation: a difficult case

1/3 FM has 1/5 observer in disagreement

FM 2

Maspero M et al. 2018 Radiat Oncol 5;13(1):105. https://doi.org/10.1186/s13014-018-1029-7

Automatic methods to localize fiducials

Maspero M et al Phys Med Biol. 2017 Oct 3;62(20):7981-

8002 https://doi.org/10.1088/1361-6560/aa875f

Gustafsson C et al. Med Phys 2017;44(11):5563-5574.

https://doi.org/10.1002/mp.12516

Ghose S et al. Med Phys 2016;43(5):2218

https://doi.org/10.1118/1.4944871

Fernandes C et al. PHIRO 2017;1:14-20

https://doi.org/10.1016/j.phro.2017.02.001

Up to ~98% accurate FM classification with

automatic method

Maspero, M. 2018, PhD Thesis, ISBN: 978-90-393-6953-1

https://dspace.library.uu.nl/handle/1874/362987

MRI as reference for

position verification3• MRI-based position verification is feasible

• No general recipe: every clinic should verify according

to its IGRT standard

+1

3Workflow

adaptation

for clinical

introduction

Clinical introduction: simulation is a complex serial workflow

Multi-vendor workflow documentation

QA testing?Data flowSoftware adaptations

MRI RTT image processing unit

Radiationoncologist

Planning Position verification

trainingWork instructions

protocollogistics

4Diffusion of

MR-only RT

Time

MR-only so far…

1976

MR

in-vivo

1998

First

Investigation

Beavis AW, et al.

Radiotherapy treatment planning of brain tumours using MRI alone.

1998 Br J Radiol.;71:544–8

MR-only not considered due to hardware constraints:

Limits on geometric accuracy

Time

MR-only so far…

1976

MR

in-vivoInitial

Investigations

sCT

generationFeasibility

Edmund, JM & Nyholm, T.

2017, Radiat Oncol, 12(1):28

Pu

bli

cati

on

s RT

MR/PET

Both

1998 2004 2006 2007 2008 2010 2011 2012 2013 2014 2015

Time

MR-only so far…

1976

MR in

in-vivoInitial

Investigations

2014

sCT

generationFeasibility

1998 2007 2012

Korhonen, J. et al.

Clinical Experiences of Treating Prostate Cancer Patients With Magnetic Resonance Imaging–Only Based Radiation

Therapy Treatment Planning Workflow

2016 Int J Radiat Oncol Bio Phys 96(2):S225

First Site

MR-only

Patients

treated with

MR-only

WIP towards

clinic

2016

2016 2018

20172014

Perspectives5

Outlook

• So far mostly brain & prostate, let’s extend to

– General pelvis, Head & neck

– Towards torso/abdomen!

The big challenge for MRI

Motion

prostate

H&N

Rectum/Gyn

esophagus

low highmedium

brain

swallowing

Bowel

motion

Respiratory

motion

Cardiac

motion

tem

po

ral fr

eq

uen

cyh

igh

low

pancreas

2. Planning 3. Treatment1. Imaging

Outlook: the additional gain

Image

fusion

Reference

Images

Plan &

DoseDelineation

Linac

Position

Verification

Repeated by the nr

of fractions

“Much to be gained by

addressing

position/motion

uncertainties during

dose delivery”Baumann et al, Nature. Rev. 2016

• MR-guided RT promises to

address position/motion

uncertainties

• Integration of MRI and linac

MR-linac, an artistic view before it became reality

MR-guided Radiotherapy?

JJW Lagendijk et al. Semin Radiat Oncol 24:207-209 2014 https://doi.org/10.1016/j.semradonc.2014.02.009

Outlook: MR-only for MR-guided RT

• largely unexplored area

– sCT with dose calculation in a magnetic field

γ

γ

e-

e-

γ

e-

B = 0

γ

γ

e-

e-

γ

e-

B = 1.5 T

Outlook: MR-only for MR-guided RT

• largely unexplored area

– sCT with dose calculation in a magnetic field

– Face the motion challenge

• Motion-robust sCT generation

– Non-cartesian MRI

• Fast sCT generation: acceleration techniques

– Parallel imaging

– Undersampling

Summary: bring home message

• MR-only Radiotherapy is feasible and a clinical possibility

• Position verification within MR-only can be accurately

performed

• General pelvis and brain are the most largely investigated area

• Commercial solutions are available and can facilitate clinical

implementation

• Each institute currently requires a reasonable effort to

streamline the workflow

Acknowledgements

• Nico van den Berg

• Peter Seevinck

• Anna Dinkla

• Mark Savenije

• Marielle Philippens

• Rob Tijssen

• Bjorn Stemkens

• Christopher Kurz

• Marcus Tyyger

• Benjamin Douglas

• Ellart Alberts

• Gerald Schubert

• Marijn Kruiskamp

• Aleksi Halkola

• Matti Lindstrum

• Martin Deppe

Thank you very much

for your attention!

[email protected]

UMCU Clinical Introduction team MR-only prostate

2. Planning

Plan &

Dose

3. Treatment1. Imaging

Buon MR-only!

Reference

Images

Delineation

Linac

Position

Verification

sCT

generation

Dosimetric

evaluation

MR

simulation

MR-only

delineations

MR-based

Ref Images

sCT

[email protected]

Graphical Work: Matteo Maspero