Comparison of Cone Beam Computed Tomography With MV Imaging for Paraspinal Radiosurgery Patient Alig

1
Comparison of Cone Beam Computed Tomography With
MV Imaging for Paraspinal Radiosurgery Patient
Alignment and Position VerificationS.
Kriminski, D.M. Lovelock, H. Amols, I. Ali, Y.
YamadaMemorial Sloan-Kettering Cancer Center,
New York, NYASTRO 2006
2

• Introduction
• Patients with oligo-metastatic paraspinal disease
  are treated at MSKCC to a high dose (18 24 Gy)
  image guided single fraction treatment
• Patients are setup to within 2 mm in an
  immobilization cradle
• A pair of MV localization images has
  traditionally been used for setup.

3

• MV images for patient alignment
• Quality of the MV images for clinically
  acceptable doses of few MU can be poor
• Only in 1/3 of all cases registration is
  possible based on patient bony anatomy
• It is difficult to say in advance whether bony
  registration would be feasible for a given
  patient
• All patients were required to undergo
  implantation of 6 gold fiducial markers into the
  spine.

4

• kV Cone Beam Computed Tomography
• MSKCC recently acquired linac with gantry mounted
  kV source and 2D flat panel detector
• Permits the acquisition of a volumetric image
  (cone beam computed tomography CBCT) at the
  time of treatment
• Improved bone contrast is expected to permit
  precise registration of the cone beam image with
  the planning CT scan without the use of implanted
  fiducial markers

5

• Motivation
• Disadvantages of MV localization
• Due to poor quality of images patient alignment
  using MV localization images requires
  implantation of gold markers into the spine
• Automatic registration of MV images is not
  feasible
• Advantages of kV CBCT
• Improved bone contrast (due to use of kV X-ray)
  together with 3D imaging makes registration more
  robust
• Automatic 3D registration is feasible
• 3D image gives clear indication of possible
  rotational setup errors

6

• Objective
• To investigate whether the setup corrections from
  kV CBCT are consistent with that from MV
  localization images
• If corrections are consistent
• The need for implantation of markers would be
  eliminated

7

• Materials and Methods
• Patients
• 16 patients with paraspinal disease eligible for
  single fraction treatment to spinal metastasis
  were accrued to an IRB approved protocol.
• Each patient underwent implantation of 6 gold
  fiducial markers, generally into the left and
  right lamina of three different vertebrae
• Three MV image pairs (AP LL or LAO) were
  acquired for each of 16 patients
• Initial, to determine couch position correction
• Verification, after corrections, prior to
  treatment
• Final, after treatment
• Each MV pair was accompanied by kV CBCT scan

8

• Positioning with MV images
• Digitally composited radiographs (DCRs) created
  using AcqSim software (Philips Medical Systems)
  were used as reference
• Similar to DRRs but voxels with HU corresponding
  to implanted metal have their opacity increased
• Results in images in which the fiducial markers
  are sharply demarcated
• Every MV image pair was manually registered to
  DCRs from plan CT to determine setup corrections
• All patient setup corrections were based on the
  corrections from the MV technique.

9

• Examples of MV images and DCR
• Outlines (yellow) are drawn manually on an
  overlay layer DCR using vendors software (Varian
  Medical Systems)
• The MV image is processed to enhance markers
• The overlay layer with the yellow marker outlines
  is manually registered with the markers on MV
  image

10

• Systematic Uncertainties in localization by MV
  images
• Anatomic features seen in the MV images are
  referenced to a grid projected onto all images
• The grid is projected by a physical device
  inserted into the head of the Linac
• Monthly quality assurance procedures keep the
  grid positioning error to less than 1 mm

11

• Positioning with kV CBCT
• Three 3D CBCT scans were acquired for each
  patient following MV images using gantry-mounted
  kV on-board imaging system (Varian Medical
  Systems, Palo Alto, CA)

• 3D image registration of CBCT to plan CT was
  performed automatically by maximizing mutual
  information using a region of interest excluding
  markers

• Corrections from MV images were used for
  comparison with that from CBCT

12

• 3D Registration of CBCT to plan CT
• Cone beam scan (cyan) is overlaid with the
  planning CT scan (red)
• In the superior/inferior direction, the ROI
  extends 2 or 3 vertebrae

Selection of ROI excluding implanted markers
prior to automatic registration
Result of maximizing mutual information
13

• Systematic Uncertainties in localization by kV
  CBCT
• Systematic errors between the radiation and CBCT
  isocenters were measured frequently over a 18
  month period and found to be very stable
• Position of radiation isocenter (for HFS
  coordinates)
• These offsets were corrected for in all
  registrations
• Residual systematic errors are estimated to be
  0.3mm

14

• Results
• Random component of MV localization
• Histograms of the difference between SI setup
  corrections determined from AP and LAO or LL MV
  localization images

• Mean and std of absolute difference are given
• If and AP LAO are used for alignment, LR and AP
  random errors have additional factors of 1.7 and
  1.4

0.50.4 mm
15

• Is this precision acceptable?
• Goal of image guidance is to reduce setup errors
  to lt 2 mm
• With MV systematic error lt 1 mm, this precision
  is sufficient for SRS
• This leaves out the issue that MV registration
  requires implantation of seeds
• Also, leaves out the question whether seeds
  represent tumor position correctly

16

• An example of CBCT to plan CT registration

• Red plan CT
• Cyan - CBCT
• Contour - PTV
• Rectangle ROI for registration
• It is seen that with good bony registration seeds
  are slightly misaligned

• Registrations based on bony structures and based
  on BB are slightly different

17

• Comparison of MV and CBCT setup corrections
• Histograms of the differences between setup
  corrections determined from the CBCT scans and
  corresponding MV localization image
• Mean and std of absolute difference are given

1.00.7 mm 1.00.6 mm
0.80.6 mm
18

• Are these differences acceptable?
• Goal of image guidance is to reduce setup errors
  to lt 2 mm
• kV CBCT provides sufficient accuracy and
  precision for high dose single fraction treatment

19

• Source of MV and CBCT differences
• Differences between CBCT and MV corrections,
  which are not due to CBCT errors, are caused by
• Uncertainty in MV / DCR registration
• Systematic errors associated with the field
  localization in MV images
• Patient motion within the immobilization cradle
• Couch re-positioning errors between the MV images
  and CBCT scans
• The random component in these sources can be
  estimated by comparing the MV corrections before
  and after treatment

20

• Pre-post change in patient position from MV
• Histograms of the differences between setup
  corrections determined from the MV localization
  image pairs before and after treatment
• Mean and std of absolute of the difference are
  given

1.11.0 mm 1.20.8 mm
1.11.0 mm
21

• Are the differences between CBCT and MV
  corrections reasonable?
• Differences between setup corrections determined
  from MV images before and after treatment
  1.11.0, 1.20.8, and 1.11.0 mm (meanstd) for
  LR, AP and SI
• Differences between the CBCT and MV setup
  corrections are 1.00.7, 1.10.6, and 0.80.6 mm
• These differences are similar
• We speculate that the contribution to these
  differences by errors in setup correction from
  CBCT is comparable or smaller than by that from MV

22

• Shifts during treatment
• Histograms of the differences between the setup
  corrections determined from kV CBCT before and
  after treatment
• Mean and std of absolute of the difference are
  given

0.50.5 mm 0.70.6 mm
1.00.9 mm
23

• Are shifts during treatment significant?
• Differences between setup corrections determined
  from kV CBCT before and after treatment are
  0.50.5, 0.70.6, and 1.00.9 mm (meanstd) for
  LR, AP and SI
• Differences between setup corrections determined
  from MV image pairs before and after treatment
  are 1.11.0, 1.20.8, and 1.11.0 mm (meanstd)
• Both MV imaging and kV CBCT indicate that patient
  displacements during treatment are small

24

• Shifts from CBCT vs. shift from MV
• Differences between the setup corrections
  determined from kV CBCT before and after
  treatment vs differences between the setup
  corrections determined from MV image pair before
  and after treatment
• Lines show ideal correlations between the
  differences

25

• Some shifts are not correlated why?
• Patient displacements marked by red circles are
  inconsistent between CBCT and MV
• For one outlier a change in posture (bending of
  the spinal cord) was observed, as clearly seen on
  CBCT
• Because MV and CBCT use different ROIs/BBs for
  patient registrations, change in the shape of
  spinal cord leads to differences in patient
  alignments
• Other discrepancies shown may be attributed to
  patient displacements between CBCT scans and MV
  imaging

26

• Conclusions
• Setup corrections determined using CBCT without
  the use of implanted markers were consistent with
  that from marker registration in MV localization
  images and DCRs
• This and other advantages of CBCT such as robust
  automatic 3D registration and clear indication of
  possible rotational setup errors have led our
  clinic to adopt CBCT for the setup of all single
  fraction paraspinal patients
• Both MV imaging and kV CBCT indicate that patient
  displacements during treatment are small