(54) Title: METHOD OF DETERMINING THE SPATIAL RESPONSE SIGNATURE OF A DETECTOR IN COMPUTED RA o DIOGRAPHY
o (57) Abstract: Method of determining the spatial response signature of a x-ray detector comprising a photostimulable phosphor
by generating a flat field image of the detector, generating a low-pass filtered version of the flat field image and background de
modulating the flat field image by pixel-wise dividing it by means of corresponding pixel values in the low-pass filtered version.
Method of determining the spatial response signature of a detector
in computed radiography.
[DESCRIPTION]
FIELD OF THE INVENTION
The present invention relates to computed radiography. The
invention more particularly relates to a method and a system for
determining the spatial response signature of a photo-stimulable
phosphor detector used in computed radiography.
BACKGROUND OF THE INVENTION
Computed radiography (CR) performance is tightly coupled to the
overall image quality and detection capabilities of the entire image
acquisition, processing and display chain. For diagnosis or during
technical image quality testing patient- or target (phantom) -
images are created on an intermediate storage medium, called image
plate or detector. During exposure the image plate traps the locally
impinging x-rays and stores the latent shadow image until it is
scanned and converted into a digital image by a digitizer.
Physical process limitations and tolerances during image plate
manufacturing generate local sensitivity variability across the
detector surface. Storage phosphor based ( amorphic or crystal ) CR
detectors are multi-layered structures composed of a substratum, an
adhesion layer, a conversion and storage layer and a protective
sealing layer. Each of these functional layers and their interfaces
may suffer from various levels of typical imperfections, blemishes
and artifacts causing locally deviating image plate sensitivity. The
medium to high spatial frequency components of the relative
sensitivity distribution across a detector's surface reflect the
image plate structure (IPS), the detector's unique signature.
A CR-image should closely reflect the patient's or object's x-ray
shadow information. Since the detector's local sensitivity is the
multiplicative factor controlling the conversion of the latent dose
information into the image signal, the IPS is inevitably water
marked into each CR image acquired from it. Local image plate
sensitivity variability can by consequence lead to diagnostic image
quality loss because the relevant patient information is polluted by
the detector's IPS. Like dose-related quantum (photon) noise and
digitizer noise, the IPS is a detector-related, disturbing noise
source which diminishes the Detective Quantum Efficiency (DQE) of
the CR system. Excessive IPS thus reduces the radiologist's reading
comfort and confidence level since it becomes more difficult to
discern subtle but important image information.
Mammography, an image quality wise highly demanding CR market,
imposes tough requirements to the magnitude and spatial extent of
the detector's sensitivity variability distribution. Stringent IPS
control is key to preserve a sufficient visibility of tiny objects
like micro-calcifications and the sharp delineation of subtle,
medium to large structures inside the breast tissue. Image plate
artifacts, isolated sensitivity disturbances, also part of a
detector's characteristic IPS, are of major concern in diagnostic
image viewing since their distinct presence can potentially hide
pathology and hamper the reading of the surrounding image area .
Excessive detector sensitivity variability can easily generate
costly yield loss in detector manufacturing.
It is an object of the present invention to characterize an x-ray
detector's unique spatial response signature.
SUMMARY OF THE INVENTION
The above-mentioned aspect is realised by a method having the
specific features set out in claim 1 .
Specific features for preferred embodiments of the invention are set
out in the dependent claims .
A detector's signature is defined as the relative, medium to high
spatial frequency components of a computed radiography detector's
characteristic sensitivity.
With the method of the present invention, it will be possible to
extract the medium to high spatial frequency components of the
relative sensitivity distribution across the surface of a CR
detector to perform a more representative quality control (QC)
testing in CR image plate manufacturing.
A more representative QC testing will result in an improvement of
the yield in CR image plate manufacturing by having IPS removal
indirectly ( improved image quality and DQE ) weaken the need for
tough IPS acceptance criteria in detector QC .
Furthermore, it will enable the removal of an image plate's
disturbing IPS noise from diagnostic CR images to obtain an
unprecedented image quality level and increased detection
capabilities (DQE) .
The method of the present invention will further allow to
redetermine a CR image plate's IPS to compensate for detector wear
and IPS drift thus optimally assuring improved image quality and
better detection capabilities (DQE) over time.
Further advantages and embodiments of the present invention will
become apparent from the following description and drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. 1 is a flow chart illustrating the different steps of the image
plate structure determination method of the present invention,
Fig. 2 is an illustration of the method steps performed to select a
reference image for spatial warping o f the other images of the image
set ,
Fig. 3 shows the spatial correlation results for a 200 pixels square
virtual marker defined in a reference image and its spatially associated
220 pixels search region in a different image acquired from that same
detector .
DETAILED DESCRIPTION O F THE INVENTION
Below a specific embodiment of the process o f determining an image
plate's (also called CR detector') spatial response signature is
described.
The image plate used in computed radiography typically comprises a
photo-stimulable phosphor.
Examples of suitable detectors comprising a photo-stimulable
phosphor are described e.g. in European patent application 1 818 943
and European patent application 1 526 552.
Flat field image generation
The process in its most general formulation comprises the steps o f
generating a flat field image by homogeneously exposing a wellcleaned
detector to radiation such as X-rays and scanning,
preferably line-wise scanning, the homogeneously exposed detector by
means of light, e.g. laser light and by digitizing the scanned
image .
Next, a low-pass filtered version of the flat field image is
generated and the flat field image is background demodulated by
means o f corresponding pixel values in said low-pass filtered
version .
Examples o f a scanning and digitizing method and apparatus (also
called digitizer) are well known in the art. The apparatus
generally comprises means for line-wise scanning (main scan
direction) a computed radiography detector that has been exposed to
penetrating irradiation (e.g. x-rays) with stimulating light (e.g.
according to the flying spot scanning principle) and means for
transporting the detector in a second direction substantially
perpendicular to the main scanning direction (sub-scan or slow scan
transport direction) to obtain a two-dimensional scan. Upon
stimulation the radiography detector emits image-wise modulated
light. Means (such as a photomultiplier ) are provided to detect
this image-wise modulated light and convert it into an electric
image signal. The electric image signal is next digitized by an
analog- to-digital convertor.
Several steps (some of which are optional) of the present invention
are described hereinafter. It will be clear to the man skilled in
the art that the numerical values which are disclosed are only given
for illustrative purposes and do not limit the present invention.
Dose- linear signal conversion
The digitizer's characteristic dose response curve is used to
convert image signals from the native format obtained by scanning
and digitizing into dose-linearized signals because the intended
removal of an image plate's structure (IPS) from CR (computed
radiography) images requires a multiplicative demodulation (if the
native format is not dose-linear) .
This first step in image preparation is performed for each of the
available CR-detector images, part of the image set (see below) .
Off Image Plate signal reconstruction
In one embodiment various views (images obtained by homogeneously
irradiating a well-cleaned photostimulable phosphor plate and
digitizing the image read out from these homogenous irradiated
phosphor plates) are acquired from the same CR detector. The images
constitute a so-called image set.
These views are preferably created by exposing and scanning a
slightly wider physical region to be able to capture and
characterize the image plate's entire screen structure up to its
borders .
The on detector pixels carry flatfield signals showing slowly
varying signal gradients due to the exposure heel effect and to the
source to image distance variation between the X-ray tube's focus
and the various locations on the image plate surface.
The off detector pixels, between the image plate borders and the
edges of the image have signals near to zero.
It is important to replace the off detector signals by a signal
level that could be expected based on the signal gradients as
reflected in the on plate signals to minimally disturb the upcoming
background signal normalization process for the on detector pixels
near the image plate's edges.
The isotropic gradient based edge-line is calculated to localize the
border-pixels of the detector.
Then the pixel -locations of the inner border line at about 1 mm
distance from the image plate's edges are calculated.
These pixels still carry normal signals since they aren't affected
by a too close proximity of the detector's border yet.
Next the 19 pixels square average signal is calculated per inner
border line pixel. The value '19' is given for illustrative
purposes and does not limit the present invention.
The off image plate pixel signal is reconstructed as follows .
First the nearest inner border line pixel position and its point
symmetrical location are determined.
Then the 19 pixels square average signal are calculated for both
locations .
Finally the off image plate signal is reconstructed b y subtracting
the difference between the point symmetry average signal and the
inner border average signal from that last one and by assigning that
value to it .
This linear extrapolation o f the average signal level and signal
gradients for the pixels beyond the detector's inner border line
ensures that the signal levels o f the inner border pixels, the
farthest pixels for which the image plate's characteristic IPS is
calculated, match with their reconstructed average background
signals .
All the image pixels beyond the image plate's defined inner border
are replaced accordingly and this region-specific processing is
performed for each o f the available detector views, part o f the
image se .
Background normalization
The IPS represents the relative, medium to high spatial frequency
components o f a CR detector's characteristic sensitivity. Low
spatial frequency image signal components, resulting from the uneven
exposure distribution across the image plate's surface, are
preferably removed upfront .
A pixel -centered 1 square cm (200 pixels square) background average
kernel low pass filters the entire image.
Exposure shading background demodulation is achieved b y dividing the
pixel signals by their background average signals. The result can
b e scaled with a fixed factor to obtain a desired background
normalized signal level.
This normalization step is performed for each of the available
detector views, part of the image set.
Bidirectional de- streaking
The images are obtained by scanning and digitizing the flat-field
exposed detectors in a digitizer. In a specific embodiment the
scanner is a flying spot scanner. Two-dimensional scanning is
obtained by line-wise scanning the detector by means of deflected
laser light in a first direction (fast scan direction) and by
transporting the detector in a second direction substantially
perpendicular to the first direction (slow scan direction) .
Sub-optimal shading compensation in the digitizer's fast scan
direction and residual speed fluctuation in the image plate
transport system introduce residual streaking in both main image
directions .
Since these streak-artifacts are not resulting from a CR-detector 's
IPS they must also be removed upfront. A statistics based filter
process, described in published European patent application 1 935
340, effectively removes these digitizer artifacts.
Each of the available detector images, part of the views set, is
streak-filtered accordingly.
High Stop Filtered Background normalization
A pixel -centered x 5 pixels square background average kernel high
stop filters the entire bidirectionally destreaked image and the low
to medium-low spatial frequency components of the detector's
relative sensitivity spectrum are demodulated by dividing the pixel
signals by their background average signals and by scaling that
result with a fixed factor to obtain the desired normalized signal
level .
The 5x5 pixels kernel size is given for illustrative purposes and
does not limit the present invention.
This normalization step is performed for each of the destreaked
detector views, part of the preprocessed image set.
Delta Clipping
Virtual marker correlation acts on two sets of neighbouring pixel
clusters, each located in a different, preprocessed view of the same
image-plate .
Polluting surface particles in one of these views can generate high
signal contrasts and these can seriously diminish the accuracy of
the register vector detection.
Signal delta clipping limits the relative maximum deviation of the
local pixel signal to +/- 1% of its local background to prevent
this. This +/- 1% clip level is given for illustrative purposes and
does not limit the present invention.
Signal delta clipping is performed for each of the high stop
filtered background normalized detector views.
Reference Image determination ( fig. 2 )
Due to tolerances in the digitizer's image plate alignment system,
the fluctuations in the slow scan detector transport mechanism and
the limited flying spot repeatability in the fast scan stimulation
and detection systems, the set of n pre-processed detector views is
inevitably affected by translation, rotation, resizing and
distortion .
It is important to restrict the amount of inter- image rotation to a
minimum before trying to calculate the spatial register of the
detector's IPS for a pair of images, part of the set.
Selecting the right reference image for spatial registration from
the image-set ensures that the accuracy o f the spatial register
calculation will b e maximally preserved by minimizing the largest,
absolute angular difference between two detector views.
It is impossible to physically equip the image plate surface with
easily discernable, physical landmarks since these would potentially
hide valuable image -information and hamper the easy reading o f the
surrounding region due to their disturbing shape and contrast.
The IP structure pattern, representing the spatial distribution o f
the image plate's relative dose response, is available a s a faint
watermark everywhere on the detector's surface though.
Pixel clusters can b e sampled from that image plate structure to act
a s flexible, virtual landmarks since they are unique and spatially
relate to only one physical region on the image plate surface.
The ability to detect these soft-markers in every image from the set
is a big asset since it enables accurate, spatial distortion
measurements between the various image plate views .
In a specific embodiment o f the present invention, two pixel -
clusters are defined a s virtual markers and centered about the
positions A and B in Image 1 in figure 2 , acting a s the spatial
reference view initially.
Both markers are defined at two different regions on the detector's
surface a t a sufficient distance. Searching through each o f the
other image plate views, the two corresponding image -locations a t
which the detector's structure matches best with these markers are
detected.
This is done by sampling all the integer pixel shifted similarly
sized clusters instances from the associated marker's larger search
region and b y looking for the pixel -position where a maximum spatial
correlation result between the virtual marker (Image 1 ) and the
cluster instance (other image) is obtained.
Fig. 3 shows the spatial correlation results for 1 cm 2 , a 200 pixels
square, virtual marker defined in a reference image and its
spatially associated, slightly larger, 220 pixels search region in a
different image acquired from that same detector.
Bidirectional interpolation, executed at each 0.1 pixel pitch
spatial instance within a maximum-centered 3 x 3 pixels correlation
result matrix followed by peak localization returns the subpixel
estimated location of the virtual marker spatial register point.
A spatial register vector RVi , starting at pixel position A in Image
i and pointing to the sub-pixel spatial register location, which
corresponds best with the detector structure as present in Virtual
Marker A (in Image 1), is defined.
Repeating that virtual marker based spatial registration process for
Virtual Marker B , the second spatial register vector RV is
established.
Having both spatial register vectors available per image in the set,
the relative angular difference, between the Image i and the Image 1
views, can be calculated and the best Reference Image can be
selected as follows:
V i , j , n 1 < i , j < n :
angle i = ATAN [ ( RV B - Rvi ) / ( B + RV - ( A + RV A ) ) ]
angle mid = ( MAX [ angle i ] + MIN [ angle i ] ) / 2
IF
I angle mid - angle i | < | angle id - angle -j |
THEN
Image i becomes the Reference Image for spatial registration with
minimal rotation impact.
Virtual marker frame definition
Once the rotation-wise best image is selected from the set to act as
the reference image for spatial registration, a virtual marker grid
(or mesh) spanning the majority of the detector's surface is
defined .
Using a bidirectional grid-pitch of 100 pixels at a 50 micron pixelsize
a 5 mm maze-size grid of invisible but accurately detectable
landmarks is generated in the Reference Image.
This creates a 57 x 45 (2565) virtual marker array for a 30 x 24 cm
CR cassette format (a cassette carrying a CR detector) and enables a
tight local control of each view's spatial distortions.
The numerical data are given for illustrative purposes and are not
limitative for the present invention.
Spatial register vector calculation
Virtual marker pixel -clusters , centered about the mesh-points within
the spatial reference image are individually sub-pixel correlated
with their corresponding, similarly sized, pixel clusters arranged
within their slightly larger, corresponding register vector search
regions in each of the other image plate views .
This way local register vectors are detected with a sufficient
surface resolution by sub-pixel interpolated maximum correlation or
by correlation-maximum centered, (thresholded) , gravity-center
determination based on the detector's hidden IPS between each
individual image plate view and the reference image. The local
register vectors found are arranged in a map per image plate view
and by concept the register vector map of the reference image would
be filled with zero vectors.
Register vector map verification and corrections
The register vector map containing the in sub-pixel spatial register
information for each of the available pictures with the reference
image is cross-checked for local unexpected virtual marker
correlation abnormalities. This is done by calculating the
interpolated or extrapolated ( virtual marker grid borders and
corners ) average register vector based on the immediately
surrounding register vectors. The locally calculated register vector
is replaced by its surrounding vector average if its vectordifference
exceeds a certain sub-pixel distance. Each of the
available image register vector maps is subjected to this
verification and correction process.
Warp-vector map generation
Once the virtual marker register vector map has been checked and
possibly modified, the local warp vectors, relating each of the
reference image pixels to their sub-pixel spatially associated
points in the other images, are created. Interpolation and or
extrapolation of the available register vector map data generates
this many thousand times bigger, image-wide map of sub-pixel
accurate warp vectors at pixel resolution. Each of the available
maps passes this map widening step.
In reference -image register warping
Based on the sub-pixel accuracy warp vector available for each
reference image pixel, the in spatial register signal reconstruction
is performed by using the pixel signals from the correlated image.
Warp vector steered interpolation of the correlated image's
surrounding pixel signal computes the in register signal . This way
each detector-view, part of the image set, is replaced by its in
spatial register (with the reference image) computed image. The
result is that the pixel signals of that warped image have been
calculated at the same physical position at the surface of the
detector as the signals from their corresponding pixels in the
reference image .
Sub-pixel phases controlled MTF reconstruction
The various images have been warped according to their image-wide
warp vector maps based on their locally surrounding original pixel
data and their bidirectional sub-pixel phases corresponding with the
actual interpolation ( resampling ) point, indicated by the spatial
register vector. A certain amount of sharpness loss is inherent to
this image resampling ( warping ) process and the resulting,
bidirectional image-blur depends on the interpolation point's
bidirectional, sub-pixel phase's magnitudes. The closer the
interpolation point is located to the nearest original image pixel
in a certain image direction, the sharper the warped image in that
direction will be. The sharpness loss is at maximum when the
interpolation point is located in the center of the four surrounding
original data pixels at identical 0,5 pixel phases in both main
image directions. Modulation Transfer Function ( sharpness )
reconstruction extracts the bi-directionally decoupled, sub-pixel
phases from the verified and corrected register vector map and
distills an anisotropic convolution filter kernel from it to re
establish the sharpness of the warped image at the level before
warping .
The frequency domain gains of this bidirectional sharpness
reconstruction filter process are near unity for the low spatial
frequencies and gradually increase towards the higher spatial
frequencies according to the levels of upsharping required for blur
removal .
Statistical filtering
Finally the IPS is calculated by statistical filtering, preferably
median averaging, the in sub-pixel spatial register interpolated and
MTF reconstructed image signals, calculated for each of the
available images, including the reference image signal. This
statistical filter process acting on the set of images reduces the
non IP related photon noise component significantly because the
photon noise, related to the limited amount of dose in a single
image, is proportional to the root of the number of images
participating during averaging. In addition polluting loose
particles or cleanable stains, occasionally present at the surface
of the image plate during scan in a minority of the images
available, are effectively removed by this statistical filtering
process since they invoke lower image signals, at the extremities
the image pixel signal histogram, due to their light absorption
during read-out.
The image plate signature can be stored as a file. It can be
encrypted prior to file export.
[CLAIMS ]
1 . Method of determining the spatial response signature of a two
dimensional x-ray detector comprising a photostimulable phosphor by
- generating a flat field image by homogeneously exposing said
detector to radiation and scanning the homogeneously exposed
detector and by digitizing the scanned image,
- generating a low-pass filtered version of said flat field image,
- demodulating said flat field image by means of corresponding pixel
values in said low-pass filtered version characterized in that
- said signature is obtained by processing multiple flat field
images generated for the same detector and
- said multiple flat field images are spatially registered by
applying spatial warping whereby a reference flat field image for
said spatial warping is selected such that the maximum angular
difference between the pixel matrices obtained by scanning said
reference flat field image and any other of said multiple flat field
images, is minimal.
2 . A method according to claim 1 wherein a signal value in a flat
field image is clipped if said signal value deviates more than a
preset threshold percentage from the pixel value of a corresponding
pixel in said demodulated flat field image.
3 . A method according to claim 1 wherein a register vector map is
calculated by cross-reg istering a multitude of spatially distributed
markers arranged in the detector surface and present in each flatfield
image.
. A method according to claim 3 wherein said markers are virtual
markers which consist of a neighbouring pixel-cluster defined within
said reference flat-field image.
5 . A method according to claim 4 wherein for a marker an in-register
location is calculated by cross-correlating its pixel-cluster data
in said reference flat-field image with a multitude of neighbouring,
bi-directionally pixel-shifted, pixel-cluster-data sets in the other
flat-field images so as to obtain cross-correlation results.
6 . A method according to claim 5 where a marker's in-register
location is determined with sub-pixel accuracy by interpolating said
cross-correlation results or by a gravity-center determination
centered around the correlation maximum of said cross-correlation
results .
7 . A method according to claim 5 wherein a bidirectional marker-grid
is defined in the reference flat-field image.
8 . A method according to claim 7 where interpolation or
extrapolation of the data in said register vector map is used to
compose a warp vector map which links each pixel of said reference
flat-field image to its physically associated location in the other
flat-field images.
9 . A method according to claim 8 where each flat-field image is
warped to correspond pixel-wise with the reference flat-field image
by interpolating or extrapolating the data at the positions
indicated by the corresponding warp vector map such that pixel
signals of the warped image have been calculated at the same
position on the detector surface as the signals of their
corresponding pixels in the reference image.
10. A method according to claim 9 wherein bi-directionally
decoupled, sub-pixel phases from said register vector map are
extracted and an anisotropic convolution filter kernel is derived
from said phases to re-establish the sharpness of the warped image
before warping.
11. A method according to claim 1 wherein a statistically filtered
value of all corresponding pixels from the warped flat-field images
is selected as signature pixel.