A 2D-3D Registration Framework for Freehand TRUS-Guided Prostate
Biopsy
A 2D-3D Registration Framework for Freehand TRUS-Guided Prostate
BiopsySiavash Khallaghi, C. Antonio Snchez, Saman Nouranian, Samira
Sojoudi, Silvia Chang, Hamidreza Abdi, Lindsay Machan, Alison
Harris, Peter Black, Martin Gleave, Larry Goldenberg, S. Sidney
Fels, and Purang Abolmaesumi
Robotics and Control Laboratory,University of British Columbia,
Vancouver, Canada
1
Roger S Kirby and Manish I Patel. Fast facts: Prostate cancer.
Health Press, 2012Most common cancer in men(excluding non-melanoma
skin cancer)Prostate CancerAffects 1 in 8 during lifetimeScreening
and Diagnosis:Digital rectal examHigh PSACore needle biopsyRoger S
Kirby and Manish I Patel. Fast facts: Prostate cancer. Health
Press, 2012
Prostate
Prostate Cancer (PCa) is the most common non-cutaneous cancer in
North American men,affecting 1 in 8 men in their lifetime.
(http://www.crs-src.ca/page.aspx?pid=1325&gclid=CNv34dvhwcUCFRSUfgod-ZQAkg,
Canadian Cancer Statistics 2014)
There are several methods of screening, but the gold standard
for diagnosis is the core needle biopsy2
NeedleBiopsy GunNeedle GuideEndocavity Ultrasound Probe
The biopsy is most often guided using a 2D endocavity ultrasound
probe, like this one, with an attached biopsy gun3
4Tissue samples tested in labProstate Biopsy30% false
negativesRoger S Kirby and Manish I Patel. Fast facts: Prostate
cancer. Health Press, 2012Frequent need to re-biopsy
The probe is inserted in through the rectum, and a needle is
fired to collect a tissue sample. This sample is sent for
histopathological analysis.Unfortunately, the current method has a
high false negative rateSo patients are frequently asked to repeat
the uncomfortable procedure4
5
3D GuidancePlan:Tracked TRUS probeRoger S Kirby and Manish I
Patel. Fast facts: Prostate cancer. Health Press, 2012
TargetPrevious Core
3D guidance is needed to Improve targeting accuracyAnd for
re-biopsy patients, to integrate the previous core locations
This is done using spatial tracking of the probe and the 2D
image to target and record biopsy cores5
6Roger S Kirby and Manish I Patel. Fast facts: Prostate cancer.
Health Press, 2012Challenge:2D 3D registration required Prostate
moves and deformsProbe tracking not sufficient
Unfortunately, during the freehand biopsy procedure, the
prostate moves and deformsSo, spatial tracking of the probe is not
enough:We need to use the 2D image information to maintain
alignment with the biopsy plan63D Freehand Guidance Framework
Reference VolumeSlice-to-Volume Registration
Rigid registration for bulk motionNon-rigid registration for
deformationsPre-procedure axial sweepVolume reconstructionTarget
planning
Target registration error: 3.15 0.81 mm
We present a 2D-3D registration framework for use in freehand
prostate biopsies.
First, we generate a 3D ultrasound from a tracked axial sweep.
This reference volume used for target planning
Next, we perform the slice-to-volume registration, consisting of
2 stepsA rigid registration to account for the bulk of the prostate
motion;A non-rigid registration to account for deformations.
Currently, we achieve a mean target registration error of around
3mm for the combined method7Trajectory-based
constraint:Pre-procedure axial sweep traces rectal wall Live 2D
slice should also fall on rectal wall Constrain probe tip to
trajectoryImage-based metricSum-of-squared distances (SSD)Local
minima due to:Limited spatial informationLow SNR of TRUS
Rigid Registration
For the rigid alignment step, we minimize the sum-of-squared
distances between image intensitiesThis method alone is highly
susceptible to local minima due tothe limited spatial
informationAnd the low signal-to-noise ratio of ultrasound
So, we introduce a constraint:We know that the pre-procedure
sweep traces the rectal wallDuring the procedure, the live 2D slice
is also somewhere on the rectal wallThus, we constrain the probe to
the pre-procedure trajectory,This accounts for any large spatial
offset, ensuring the rectal walls coincide8Finite element-based
method:Incorporates physical properties of the tissueSimultaneously
minimize strain and image SSD
Non-Rigid Registration
For residual motion and for deformations, we use a FEM-based
methodThis allows us to incorporate material properties of the
prostate and surrounding tissue, letting it deform physicallyThe
FEM is driven by image forces, as we simultaneously minimize the
total strain and the SSD metric.9
Example Registration
This is the algorithm in action
The 2D slice slides along the trajectory from the 3D volume,
finding the best match based on image intensitiesThe finite element
method then takes over, moving and bending the volume to achieve a
better fit10
2D Target SliceInitializationExample Registration
Here we see the 2D target slice, and the initial guess within
the volume, based solely on spatial tracking. As you can see, the
boundaries of the prostate appear quite different between the
images.11
2D Target SliceRigid RegistrationExample Registration
After rigid registration, the prostate boundaries are much more
similar, and there is a calcification visible at the same location
in both images12
2D Target SliceDeformable RegistrationExample Registration
Following finite-element correction, the calcification is
brighter in the middle. The volume has also deformed to better
match the boundary features.13Registration Results
TargetInitialRigidFEM-Based
Both stages of the registration method play significant rolesThe
trajectory-based rigid registration helps avoid local minima,
bringing us to a good starting point for the FEM-based registration
to take over
Our method is shown to robust, locating some of the finer
details present in the slice14
C. Antonio SnchezDept. of Elec. & Comp. Eng.University of
British ColumbiaVancouver, BC, [email protected]
KhallaghiDept. of Elec. & Comp. Eng.University of British
ColumbiaVancouver, BC, [email protected]
If youd like to find out more about our work, please come and
find us. We would be delighted to discuss the details of the
method, and answer any of your questions.
Thank-you.15