Transcript Slide 1
Correlation Evaluation of a
Tumor Tracking System
Using Multiple External
Markers
Hui Yan, Fang-Fang Yin, et al
(Duke University Med. Ctr.)
Overview
Patient set-up & tumor localization difficult sites
with frequent organ movement, like lungs
CTV >
PTV by considerable margin to account for target
displacement
Actual dose differ from intended dose distribution to tumor
Causes of internal target displacement:
Position-related target shift,
Interfractional organ motion;
Intrafractional organ motion (esp. respiration related motion);
Overview
Several
breath-holding techniques developed to minimize
respiratory-related organ motion
Reduce CTV margins, reduce motion, BUT cannot eliminate
Direct
tumor tracking system have been employed using
implanted metal seeds and markers with x-ray imaging
Continuous imaging causes radiation to be significant
Indirect tumor
tracking systems:
Spirometer & strain gauge ;
External markers/sensors (Infrared LEDs)
Investigation
This study: multiple external marker tracking system
was investigated.
Infrared
cameras and a clinical simulator were used to
acquire the motion of an internal and multiple external
markers simultaneously.
Correlation between internal and external motion signals
were analyzed using a cross-covariance method.
Composite signals for each comparison were generated
with multiple external signals using linear regression.
Experiment/Data Acquisition
7 patients undergoing
radiotherapy for lung
cancer (all with Karnofsky >/= 70)
3 to 5 IR reflective
external markers were
placed on patients’ chest
wall
Experiment/Data Acquisition
With each patient:
6
sessions with 3 identical
sessions in each imaging
direction
2
S1: Free breathing for 40s (FFB)
S2: Free breathing for 10s, hold for
5s, resume free breathing 10s (BH)
S3: Free breathing for 40s (SFB)
IR cameras collect data, 10Hz
3D marker location time index
saved
Fluoroscopic
images, 15Hz
Experiment/Data Acquisition
The mean displacement and σ
of the tumor center for each
patient: Table II
Mean
deviation ~2 pixels, avg.
peak-to-peak displacement was
up to 30 pixels
Mean deviation to relatively
small
All the data was normalized to
the range of [0,1] for ease of
analyzing and comparison
Correlation Analysis Method
The cross-covariance (XCOV in Matlab) function
was used
Same
as traditional correlation coefficients, but also
provided additional information about the phase shift
XCOV func. φxy(m) is the cross-correlation of 2 meanremoved time series xn and yn:
Finite-length time series, XCOV becomes:
Correlation Analysis Method
After
index conversion from [-N,N] to [1,2N-1] and
normalization:
The
phase shift between 2 input series can found from
XCOV sequence. If no shift, max XCOV sequence value
will occur at index N.
where δ is the phase shift;
Correlation Analysis Method
Correlation Analysis between external and
internal signals
XCOV function used between
all pairs of external & internal
signals to gather mean, min,
and max of the phase shifts
(Table III).
Max phase shift = 0.81s
Mean varies from 0.12s - 0.52s
Correlation coeff. [0,0.98]
After the correction for phase
shift, the average correlation
coeffcient value increased
significantly and the
corresponding deviation
decreased.
Correlation coefficients
grouped by breathing patterns
(Table IV).
Correlation Analysis between composite
and internal signals
Different composite signals were generated using different
combinations of external signals.
To see the effect of the number of external markers, the
combination formula was used: Cmn= n!/[(n-m)!m!]
# different cmbinations of m external markers from n markers.
Correlation errors of the 3 composites for a combination were
averaged; mean, max & min were tabulated
Effect of the number of external markers
Most
cases, a decrease in mean correlation error was observed
when more external signals were taken into account
But minimum values of correlation error do not decrease as the
number of external markers increased.
Effect of dimensional components of
internal and external signal
Composite signals generated from external signals in a specified
dimension or directions (grouped by breathing pattern):
Lateral, Longitudinal, Vertical, Lateral-Longitudinal, LongitudinalVertical, Lateral-Vertical, and Lateral-Longitudinal-Vertical;
Effect of dimensional components of
internal and external signal
Minimal
correlation error was achieved by the composite
signal consisting of external markers in ALL three
dimensions;
Effect of dimensional components of
internal and external signal
With
external marker dimensional components fixed,
composite signals were generated and compared to the
internal signal in the same dimension.
Effect of dimensional components of
internal and external signal
Correlation errors
were lower when more components
external signals were included in the composite signal.
Relatively, the largest correlation errors were found in
internal signals in the lateral direction of AP imaging.
Effect of the breathing pattern
The
two free breathing sessions (FFB & SFB) exhibited a
similar level of correlation errors (mean, min & max) in all
patients.
Patients 1,2,4,5,7 had similar correlation errors for all 3
breathing patterns.
Patients 3 & 6: Visible differences in the correlation errors
of BH and free-breathing sessions.
Effect of the breathing pattern
The bars represent the min &
max values of the correlation
errors
Most of the points follow an
approximately linear
relationship
This linear relationship
indicates that the correlation
error between the composite
and internal signals is affected
inversely by the quality of
correlation coefficient
between external and internal
signals.
Effect of the phase shift
Table
IX tabulates correlation errors caused by the external
composite signals before and after the correction for the
phase shift
Significant decrease in correlation error
Patients 1, 2, 4, 6, 7 similar levels of correlation errors
before & after correction
Patients 3 & 5 mean and max values were decreased by up
to 20%
Effect of the phase shift
In
addition to the decrease in mean value of correlation
errors, consistent decreases of the maximum and minimum
values of correlation errors were also observed in most of
patients.
Questions?
GO GATORS!