d into
ion as
cterior
found
cations
d.
ent is
patient
e that
Nt | L———À ÁN
these
cation
scribed
jatient
Camera Calibration
System cameras must be calibrated at regular
intervals, after changing camera positions for
different treatment positions or after accidental
disturbance.
When a calibration is needed, the chair is lowered
below the floor and the control frame is placed in
position. All three camera images are then
thresholded and automatically searched for control
targets. The target positions are stored and
displayed by circles on the video monitors for
verification by the operator. The targets are then
automatically identified by comparing their observed
image coordinates with a list of expected image
positions of all the control points visible from that
station.
A least squares DLT solution provides the
transformation parameters, which are stored for
subsequent target position determinations.
System Check
In a pre-treatment check the camera/chair system is
tested. The test entails the coordinate determination
of a set of check points situated on the chair. Here,
as in all other stages, target coordinates are
evaluated by means of least Squares Space
intersections. Results within preset tolerance levels
confirm that the transformation parameters still
reflect the true camera parameters and that the chair
system is in adjustment. Failure of this test
necessitates a full re-calibration of all three
cameras and re-initialisation of the chair.
In the interest of patient safety, system protocol
prohibits entry into the patient positioning stage
until this check is passed.
Patient Positioning
Now the crucial stage of the procedure, *he
positioning of the patient into the proton beam, is
initiated. The software is structured to execute this
in three steps:
X coordinating the reference targets on
the patient’s head
2 calculating the translations and
rotations necessary to position the
patient into the beam line
3 instructing the chair to move the
patient accordingly
To realise this process the patient is seated and
provisionally aligned with the beam by means of the
manual chair controller.
After image capture, thresholding and target
detection and centring, the operator interactively
identifies reference targets on the patient's head to
correlate with the target numbers allocated in the
scanning stage. The reference target coordinates are
then calculated in the beam system.
The scan coordinates of beam entry point and lesion
are transformed (iterative least squares model) into
the beam coordinate system via the reference targets,
now known in both systems. These transformed
positions are then used to evaluate the necessary
translations and rotations for aligning the
lesion/entry point vector into the beam line.
The alignment information is sent by communication
port to the chair computer, which converts this into
mechanical translations and rotations for the chair.
The chair is automatically moved to place the patient
into the treatment position. Finally, before beam
activation, the patient position is redetermined by
PPPS as a check.
Patient Monitoring
Throughout the treatment, the patient, exposed to the
active beam is closely monitored for possible
movement. It is here that the highest computing
Speed is needed and real-time capability is most
essential. A modified processing approach,
characterised by the following, is thus implemented:
1 The interactive thresholding and target
identification stages are eliminated as
the relevant information is assumed to
remain practically unchanged.
2 The target detection stage is omitted
and the target centring routine occurs
in predetermined search windows centred
around the expected image coordinates of
the reference points.
3 As any substantial movement is likely to
be discovered on coordinating the first
reference target, point " by point
processing (centring and intersection)
is employed to provide a fast
intermediate check on any unwanted
patient movement.
4 As a main check a non-iterative
transformation is used to compute the
positions of lesion and beam entry
point. If one of these positions is
found to have moved beyond a preset
tolerance the beam is immediately
deactivated.
If no patient movement occurs, monitoring continues
at a high frequency until the required dose is
received. As a precautionary measure, a manual
overdrive can at all times deactivate the beam.
SYSTEM TESTS
Laboratory tests proved entirely satisfactory for the
intended application. Sub-millimetre accuracy of
target positions was achieved in simulations,
compared to expected CT scan accuracies of + 1.5
millimetres. Tests with the chair showed sub-
millimetre agreement between chair movements
evaluated with PPPS and as recorded by chair
decoders. Monitoring speeds of + 0.3 seconds for a
complete check loop were registered for three images
with 8 reference targets and 9x9 pixel search
windows.
CONCLUSIONS
Digital close range photogrammetry appears ideally
suited for the placing and monitoring of patients
undergoing proton beam therapy. The system tests have
resulted in satisfactory precisions acceptable within
the NAC parameters. At the time of preparing this
paper, the PPPS system has been installed at the NAC
in Faure and is undergoing tests. Beyond the
application discussed in this paper a wide range of
other positioning problems could be solved using
slightly modified versions of the PPPS concept.
REFERENCES
Adams, L.P., 1989. Report on Patient Support System
for Proton Therapy Stereophotogrammetric Positioning
System, University of Cape Town Report, 5 pages.
Adams, L.P., 1990. Report on obtaining 3 dimensional
Coordinates from CT Scans, NAC project report, 13
pages.
Adams, L.P. and H.Rüther, 1989. A Stereo-
photogrammetric System Using Multiple Digital Cameras
for the Accurate Placement of a Proton Beam, Optical
3-D Measurement Techniques, Editors: A.Gruen &
H.Kahmen, Wichmann, Karlsruhe, pp.164-172.
Brown, D.C., 1982. STARS, A Turnkey System for Close
Range Photogrammetry, Proceedings International