Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Optimization-Based Geometry Correction Of Blood Flow CFD Simulations Using 4D-Flow Data

Hegardt, Fritiof LU (2022) MVKM01 20202
Department of Energy Sciences
Abstract
4D-flow is a powerful tool capable of capturing 3-dimensional,
time-resolved flow measurements of blood flow in the body. Their current
use is limited by resolution and scan times. A proposed solution is to use
Simulation Based Imaging (SBI). This combines lower resolution 4D-flow scans
with CFD simulations to improve resolution and reduce scan times. Previous
work has focused on inlet and outlet conditions. This thesis explores the
possibility of adding geometry correction to an optimization-based framework
for SBI.
Three different ways of deforming the mesh were implemented
to explore geometry optimization. First finding the correct the diameter of a
simple channel, second finding an initial rotation and translation error, and
... (More)
4D-flow is a powerful tool capable of capturing 3-dimensional,
time-resolved flow measurements of blood flow in the body. Their current
use is limited by resolution and scan times. A proposed solution is to use
Simulation Based Imaging (SBI). This combines lower resolution 4D-flow scans
with CFD simulations to improve resolution and reduce scan times. Previous
work has focused on inlet and outlet conditions. This thesis explores the
possibility of adding geometry correction to an optimization-based framework
for SBI.
Three different ways of deforming the mesh were implemented
to explore geometry optimization. First finding the correct the diameter of a
simple channel, second finding an initial rotation and translation error, and
finally small boundary perturbations using Radial Basis Functions. The CFD
simulations were performed using higher order finite element discretization. To
compare the CFD simulations the MR images a forward function was used and
the optimization was performed using gradients calculated using the adjoint
method as well as finite differences.
All three cases managed to correct the geometry errors both with
and without noise in the MR-image, but the errors increased with increased
noise levels.
The results shows that all three approaches worked in 2D
but was sensitive to noise and flow conditions.
4 (Less)
Popular Abstract (Swedish)
Kan man kombinera Magnetisk Resonanstomografi, mer k ̈ant som MRI, Magnetic
Resonance Imaging, med geometrioptimering av fl ̈odessimuleringar f ̈or att f ̊a fram
mer detaljrika bilder av blodfl ̈odet? I denna studiehar vi arbetat med att vidareutveckla
ett verktyg f ̈or att ̈oka uppl ̈osningen p ̊a tredimensionella bilder av blodk ̈arl i kroppen.
Ett syfte med den h ̈ar tekniken ̈ar att ge l ̈akare tillg ̊ang till b ̈attre information kring
utseendet och funktion av patientens blodk ̈arl vid till exempel hj ̈art- och k ̈arlsjukdomar.
Med hj ̈alp utav Magnetisk Resonanstomografi kan man f ̊a fram tidsuppl ̈osta,
tredimensionella bilder ̈over blodfl ̈odet i kroppen - 4D-fl ̈odesbilder. Detta ̈ar ett
potentiellt v ̈aldigt kraftfullt... (More)
Kan man kombinera Magnetisk Resonanstomografi, mer k ̈ant som MRI, Magnetic
Resonance Imaging, med geometrioptimering av fl ̈odessimuleringar f ̈or att f ̊a fram
mer detaljrika bilder av blodfl ̈odet? I denna studiehar vi arbetat med att vidareutveckla
ett verktyg f ̈or att ̈oka uppl ̈osningen p ̊a tredimensionella bilder av blodk ̈arl i kroppen.
Ett syfte med den h ̈ar tekniken ̈ar att ge l ̈akare tillg ̊ang till b ̈attre information kring
utseendet och funktion av patientens blodk ̈arl vid till exempel hj ̈art- och k ̈arlsjukdomar.
Med hj ̈alp utav Magnetisk Resonanstomografi kan man f ̊a fram tidsuppl ̈osta,
tredimensionella bilder ̈over blodfl ̈odet i kroppen - 4D-fl ̈odesbilder. Detta ̈ar ett
potentiellt v ̈aldigt kraftfullt verktyg f ̈or att hj ̈alpa l ̈akare att diagnostisera hj ̈art- och
k ̈arlsjukdomar men eftersom mer detaljerade bilder kr ̈aver l ̈angre tid f ̈or skannin-
gen s ̊a ̈ar i dagsl ̈aget ̈ar anv ̈andningsomr ̊adet begr ̈ansat. L ̊anga skanningtider ̈ar inte
̈onskv ̈art eftersom patienten m ̊aste ligga helt still i MR-maskinen under hela f ̈orloppet
f ̈or att f ̊a bra bilder. En l ̈osning p ̊a problemen ̈ar att anv ̈anda SBI, Simulation Based
Imaging. Den h ̈ar tekniken anv ̈ander sig av datan fr ̊an 4D-flodesbilderna f ̈or att
g ̈ora datorsimuleringar som kan ha h ̈ogre uppl ̈osning ̈an den ursprungliga bilden.
Med hj ̈alp utav SBI finns b ̊ade potential att korta ner skannings tiderna genom att
ta 4D-fl ̈odesbilder med l ̈agre uppl ̈osning och skala upp uppl ̈osningen med hj ̈alp av
simuleringarn.
Hur bra resultat man f ̊ar med SBI ̈ar avh ̈angigt p ̊a hur n ̈ara simuleringarna kan
komma verkligheten. Tidigare studier har visat att det ̈ar m ̈ojligt att det ̈ar m ̈ojligt
att anv ̈anda sig av optimering av simuleringarnas randvillkor s ̊a som blodfl ̈odets
hastighet vid in- och utlopp f ̈or att g ̈ora simuleringarna mer verklighetstrogna. Det
h ̈ar arbetet utforskar m ̈ojligheterna kring att addera geometrianpassning till det op-
timeringsbaserade ramverket SBI.
F ̈or att underl ̈atta arbetet och f ̊a snabbare simuleringar har vi i denna studie en-
bart jobbat i tv ̊a dimensioner. F ̈or att f ̊a 4D-fl ̈odesbilder anv ̈andes en algoritm som
imiterar fysiken i en MRI f ̈or att omvandlade 2D simuleringar till tv ̊adimesionella 4D-
fl ̈odesbilder. P ̊a s ̊a vis kunde vi testa optimeringens prestanda vid olika uppl ̈osningar.
Ett artificiellt brus kunde ocks ̊a l ̈aggas p ̊a f ̈or att testa metodens robusthet om det ̈ar
brus i 4D-fl ̈odesbilden, n ̊agot som kan uppst ̊a vid kortare skanningstider. Optimering
av geometrin pr ̈ovades i tre olika fall genom att deformera geometrin.
I det f ̈orsta fallet testades att hitta den r ̈atta diametern p ̊a en enkel kanal. I
det andra fallet testades att ta bort initiala rotations- och f ̈orskjutningsfel mellan
simuleringen och orginalgeomtrin som skulle kunnan uppst ̊a om till exempel patienten
r ̈or p ̊a sig under skanningen. I det sista fallet testades att korrigera sm ̊a deformationer
p ̊a k ̈arlets v ̈aggar. F ̈or alla tre fallen lyckades vi r ̈atta till felen i geometrin b ̊ade med
och utan brus i MR-bilden, dock ̈okade felen med ̈okat brus i bilden. Resultatet
visade att alla tre tillv ̈agag ̊angss ̈atten fungerade i 2D men fortsatt utveckling kr ̈avs
f ̈or att utvidga metoden till att ̈aven fungera i tre dimensioner. S ̈arskilt det sista
fallet var k ̈ansligt f ̈or omr ̊aden med l ̊angsamma fl ̈oden d ̈ar signalen blir svagare och
d ̈arf ̈or sv ̊arare att optimera. (Less)
Please use this url to cite or link to this publication:
author
Hegardt, Fritiof LU
supervisor
organization
course
MVKM01 20202
year
type
H2 - Master's Degree (Two Years)
subject
report number
LUTMDN/TMHP-21/5472-SE
ISSN
0282-1990
language
English
id
9091776
date added to LUP
2022-06-23 08:37:10
date last changed
2022-06-23 08:37:10
@misc{9091776,
  abstract     = {{4D-flow is a powerful tool capable of capturing 3-dimensional,
time-resolved flow measurements of blood flow in the body. Their current
use is limited by resolution and scan times. A proposed solution is to use
Simulation Based Imaging (SBI). This combines lower resolution 4D-flow scans
with CFD simulations to improve resolution and reduce scan times. Previous
work has focused on inlet and outlet conditions. This thesis explores the
possibility of adding geometry correction to an optimization-based framework
for SBI.
Three different ways of deforming the mesh were implemented
to explore geometry optimization. First finding the correct the diameter of a
simple channel, second finding an initial rotation and translation error, and
finally small boundary perturbations using Radial Basis Functions. The CFD
simulations were performed using higher order finite element discretization. To
compare the CFD simulations the MR images a forward function was used and
the optimization was performed using gradients calculated using the adjoint
method as well as finite differences.
All three cases managed to correct the geometry errors both with
and without noise in the MR-image, but the errors increased with increased
noise levels.
The results shows that all three approaches worked in 2D
but was sensitive to noise and flow conditions.
4}},
  author       = {{Hegardt, Fritiof}},
  issn         = {{0282-1990}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Optimization-Based Geometry Correction Of Blood Flow CFD Simulations Using 4D-Flow Data}},
  year         = {{2022}},
}