FIELD OF THE INVENTION
The present invention relates to direct and computed
radiography.
5
The invention more particularly relates to a method for
preventing a sub-optimal gain map quality by detecting avoidable
disturbances present in the x-ray beam-path during system
(re) calibration.
10
BACKGROUND OF THE INVENTION
System calibration is extremely important for digital and
computed projection radiography where flat-panel detectors and xray
storage media in combination with digitizers are used to
15 acquire digital images for clinical, veterinary or industrial use.
These image acquisition devices are rather complex hybrid
(analog and digital) systems which are composed of a variety of
highly interacting mechanical, electro-optical, physico-chemical,
20 electronics, software and image-processing components and
processes each having its typical tolerances and physical
properties.
The overall image quality performance of a radiographic
25 system can also depend on the ambient temperature, the humidity,
the atmospheric pressure as well as on the x-ray exposure history
linked to the degree of system usage and the system's actual age.
In addition gradually increasing levels of system
30 contamination due to the external contact of the system components
with radiographed patients, animals, objects, operator personnel
or caused by fiber- and dust particle pollution can depend on the
equipment's application-specific usage modes and the ambient
climate conditions and can influence the properties and the
35 behaviour of the system's individual components and processes thus
leading to more frequent, beyond the periodically scheduled,
cleaning and recalibration activities.
Especially for the flat-panel-detectors where next to dense
pixel-individual light-trapping or direct x-ray detection arraycircuitry
also massive amounts of highly miniaturized pixel-, row-
, and column-specific galvanic interconnections and several block-
5 wise arranged read-out electronic circuits should cooperate
harmonically the signal-quality and the x-ray response of the
individual image-pixels, the image-rows and the image-columns will
slowly or sometimes even suddenly degrade to a level where the
required high image quality level can no longer be assured without
10 corrective actions.
Therefore the image-acquisition system needs to be cleaned
and recalibrated on a regular basis.
15 These activities are typically executed not only after the
equipment is delivered and installed as a precondition to
performing its initial acceptance test but also before each
scheduled periodic quality control test.
20 After an equipment move, after system modifications and
after preventive maintenance or repair interventions to critical
system-components an additional system recalibration is often
necessary as well to ensure a safe and effective operation of the
radiographic image-acquisition system within the predetermined
25 overall image quality range.
The system calibration process not only delivers a better
adjusted and cleaner state of the radiographic equipment but also
generates one or multiple image-wide maps at pixel resolution for
30 the reconstruction of unstable and or defective pixels, rows and
columns in addition to one or more gain maps for the software- or
hardware-based, pixel-wise sensitivity-correction of raw
diagnostic images. A gain map of a detector system is an imagewide
representation of the (relative) signal response of each
35 individual detector-pixel to x-ray dose.
Once established these freshly generated correction maps are
used to optimize the image quality of each raw image acquired from
then on and these maps will remain unchanged and in effect till
the next system recalibration process, which will produce a new
5 set of correction maps, is performed successfully.
In general several thousands of raw flat-panel-detector
images acquired are corrected using the same set of correction
maps, determined during the last system (re-)calibration.
10
(Re-)Calibration activities are, although of vital
importance for a normal system operation, rather workflowdisturbing
since the process of system-cleaning, equipment readjustments,
calibration-dedicated image set acquisition and data-
15 processing to generate the correction maps often requires the
manual intervention of an operator and can be time-consuming while
making a normal diagnostic use of the system impossible.
Without regulations or local recommendations on the minimum
20 frequency of system recalibration the argument of the inevitable
down-time due to system recalibration activities can lead to
situations where users tend to postpone the system recalibration
process as long as possible.
25 This by consequence means that even more than the normally
already high amount of raw flat-panel-detector images acquired
will be corrected based on that same set of correction-maps.
If the gain map, used for the pixel-wise sensitivity
30 correction of the raw images acquired, would slowly degrade over
time to a state where it becomes insufficiently representative, a
vast amount of corrected diagnostic images might be impacted by
various levels of image-artifacts of which some could be only
faintly noticeable after a while and this might lead to
35 deteriorated reading comfort, radiologist uncertainty and
eventually to a false diagnosis.
Even an image-retake, causing a patient or object to be reexposed
in addition to introducing a time-costly workflowdisturbance,
might in such cases still be insufficient to make up
for the locally lacking, disturbed image quality of the corrected
5 image since the same suboptimal gain map, the real cause of the
problem, will again be used to correct the new image.
The gain map, which is determined as one of the outputs of
the (re)calibration process, is often calculated from a set of
l o non-x-ray exposed, raw dark-images in combination with a set of
dedicated, homogeneously exposed raw flat field images.
Once the tubers focal spot, the source-to-image distance,
the level of beam-collimation, the beam-filtration, the tube-
15 voltage, the tube current and the exposure-time are set up and
also the flat-panel-detector is geometrically positioned these
dedicated image sets, required for the gain map determination
during calibration, can easily be acquired sequentially from the
operatorrs control cabinet without the need for further manual
20 interventions to the system itself.
Sometimes though, visible objects that are forgotten to be
removed (e.g. a dosimeter, a cleaning cloth, tools, all other
kinds of objects commonly used by operators and service personnel
25 during maintenance, repair and recalibration) as well as invisible
objects (falsely positioned collimator blades, a badly mounted
grid or automatic exposure control chamber) can still disturb the
x-ray beam path during the actual acquisition of the calibrationdedicated
images. These beam-path disturbance problems will not
30 likely be seen from that remote operator location given that these
individual images are acquired semi-automatically without any
visual inspection performed on them at all.
These disturbing objects can be size-wise difficult to
35 detect and can partially absorb the x-ray radiation towards the
flat-panel detector-surface thus locally impacting the homogeneous
character of the exposure field, required for the successful
acquisition of the set of representative flat field images for the
calculation of the gain map.
Disturbing objects can have various dimensions from very big
5 (e.g. a screw-driver, a pull-over) to very small (e.g. a screw, a
washer, a lost staple).
Some objects with a mixed material composition can locally
introduce strongly fluctuating x-ray attenuation (e.g. a
10 dosimeter) whereas others have hardly noticeable, fuzzy and noisy
object-borders (e-g. a cleaning-cloth). Even an unexpected visit
of an insect accidentally interfering with the x-ray beam path
during system recalibration can't be totally excluded.
In addition non-corrected raw flat-panel detector images,
15 also in scope for this inspection, can exhibit a significant level
of streakiness and strip-wise signal-variation due to the multiple
ASICs-based electronic circuits used for the parallel image readout
of the detector array.
An inspection concept relying on the signal difference
20 between an object and its smooth background or on signal-gradientbased
edge-analysis will be insufficiently effective to look for
all these kinds of image-disturbances given the presence of the
other non-object related noise sources inherent to raw, noncorrected
images.
25
If no disturbing object inspection on each of the raw flat
field exposed images is performed to upfront determine whether or
not that image is sufficiently disturbance free and can be allowed
to act as a valid and representative input image for the
30 calculation of the gain map, sub-optimal gain maps for the
correction of many thousands of future raw diagnostic images can
potentially be generated without notice.
Best case such a sub-optimal gain map will generate a
35 corrected verification flat field image, often acquired as a final
step in the calibration procedure, which is showing a sufficient
level of image disturbance to be regarded as a system calibration
problem by the operator who is performing the (re)calibration
activities.
In that case the calibration must be regarded as failed and
5 a new calibration process with the acquisition of new flat field
images is required.
Re-performing the syst.em-calibration requires additional
work and can significantly increase the system's down-time.
10 The problem originating from a disturbed gain map passes the
calibration process undetected in case no final flat field
verification, using the new but object-disturbed gain map for
correction of the raw flat field image, is performed.
It will then depend on the radiologists experience and on
15 the level of disturbance visibility whether or not that suboptimal
calibration state, generating disturbed diagnostic images
after raw image correction, can be either detected and mitigated
quickly by a system-recalibration or will .inevitably lead to a
reduced overall image quality level which might hamper the proper
20 diagnoses of many corrected images to follow.
Although the fact that the acquisition of gain correction
data should be done in the absence of any object in the X-ray beam
irradiating the detector is well-known from prior art references
25 EP 2 050 395; WO 2007/043974 and DE 10 2005 017491, none of the
prior art references addresses the non-trivial problem with
disturbing objects which may lead to imperceptible acquisition of
corrupted gain correction data.
30 It is thus an aspect of this invention to provide a method
to avoid acquisition of unacceptable gain correction data.
A similar problem with beam-path disturbing objects or
surface-contamination arises for the highly demanding computed
35 radiography applications too where gain maps to correct for the
detector-specific, two-dimensional sensitivity-distributions of
the x-ray storage detector-media are determined in manufacturing
according to European patent application EP 2407016 A2 entitled
"Method of determining the spatial response signature of a
5 detector in computed radiographyr' and in co-pending European
patent application 11151202.6 entitled "Method of removing the
spatial response signature from a computed radiography image."
10 SuMM7UtY OF THE INVENTION
The above-mentioned aspects are obviated by a method having
the specific steps set out in claim 1. Specific steps and
features for preferred embodiments of the invention are set out in
the dependent claims.
15
This invention proposes a fast and effective method to
prevent the above discussed problems caused by a disturbed gain
map correction resulting from an insufficiently homogeneous x-ray
beam state due to the presence of avoidable objects or surface
20 contamination present anywhere in the beam-path during the system-
(re)calibration activities.
An automatic method, relying on the statistical analysis of
a multitude of adjacent or partially overlapping, potentially
25 obj ect-disturbed regions-of -interest, can act on the raw (noncorrected),
dedicated flat field images, acquired for the purpose
of gain map calculation to optimize the overall image quality
performance of direct and computed radiography image-acquisition
systems.
30
As a result a sub-optimal image condition can be detected
and. mitigated already early on in the process of gain map
determination while performing an initial, periodic or additional
system- (re) calibration.
35
Costly time-loss and unnecessary tube-wear associated with
the operator-assisted acquisition of a set of beam-path disturbed,
useless flat field images can thus be avoided and a higher
certainty-level regarding the validity of the calculated gain map,
an output of the (re)calibration process, can be achieved.
5 Further advantages and embodiments of the present invention
will become apparent from the following description and drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. 1 illustrates the decision making logic based on the
lo set of local shoulder significance and unbalance results analyzed
in the image as compared to a set of predetermined threshold
levels,
Fig. 2 represents a local region-of-interest as cropped from
15 an object-disturbed image-tile and its statistical translation
into spatially distributed shoulder-pixel patterns,
Fig. 3 represents the valid pixel histogram of the objectdisturbed
local region-of-interest as analyzed and split
20 statistically into distinctly gap-separated lower and higher
shoulder bins,
Fig. 4 represents both local shoulder fractions and their
shoulder ratio relative to their significance and unbalance
25 threshold levels.
DETAILED DESCRIPTION OF THE INVENTION
Upon an initial calibration after installation or while
performing a periodic or an additional re-calibration of a flat
30 panel detector direct radiography system, the use of this
equipment for diagnostic imaging purposes is temporarily
suspended.
Time and personnel are freed up to acquire new calibration-
35 dedicated image sets, composed of non-exposed as well as
homogeneously exposed raw flat panel detector images, and to be
able to calculate the various updated defective pixel maps and
updated gain maps, necessary to convert the many thousands of raw
diagnostic images to follow into optimally corrected images.
Below a specific embodiment of the method of detecting a
5 sub-optimal gain-correction-map condition, caused by the
acquisition of beam-path disturbed quasi homogeneously exposed
images during system-calibration, is described for a flat panel
detector radiography system where individual x-ray images,
acquired to compose the set of homogeneously exposed images are
l o individually subjected to a fast, automatic, disturbing object
inspection method prior to being accepted as a valid input image
for the calculation of an updated gain map.
Fig. 1 explains a specific embodiment of the method of this
15 invention by means of a process flow-chart acting on an individual
image, acquired for the purpose of gain map determination, which
might contain a disturbed, local region-of-interest.
The avoidable, disturbed x-ray beam path condition is in
20 this exemplary embodiment caused by an elastic band which was left
behind by accident on the detector-surface before starting with
calibration-dedicated image acquisitions. This is depicted in
Fig. 2.
25 A disturbed image-tile, represented with high contrastmagnification
for improved visibility of this low x-ray absorption
object, shows how this 'forgotten to remove' object partially
intersects with an arbitrarily chosen region-of-interest ROIij and
how it locally influences the normally expected smooth background
30 noise pattern which is typical for a homogeneously exposed
detector image.
The process of automatic detection of disturbing objects in
homogeneously exposed flat field images starts by dividing the
35 image, subjected to this inspection, in a plurality of much
smaller, local regions-of-interest.
Although exactly adjacent local inspection ROIs minimize the
inspection work, a partial overlap of neighbouring ROIs will be
more effective for the detection of small, disturbing beam-path
objects.
5
If these occur at the edge or in the corner of a local ROI
the small size objects are sub-divided into even smaller imagedisturbances
scattered across the adjacent ROIs.
The bigger the ROI-overlap, the smaller the effect of
l o spatial scattering across the neighbour ROIs will be.
The beam path inspection concept described in this
embodiment uses 64x64 pixels ( 8 rnrn square ) inspection-R0Is with
a 1/4th ROI-size overlap in both image directions.
By consequence the entire image-impact caused by disturbing
15 objects with lengths below 4 mun will thus always have its full
effect either locally or in one of the eight partially
overlapping, neighbour ROIs which are all subjected to this
inspection too.
20 Once the image's inspection area is divided into a grid of
partially overlapping inspection ROIs each of these local ROIs is
analyzed as shown by the loop-structure in the Fig.1 inspection
method flowchart.
25 For direct radiography flat-panel detector imagery a
predetermined so-called defect map which flags the detectorarray's
unreliable pixels, rows and colums may be available for
the purpose of image-reconstruction using neighbouring, reliable
pixel data.
30 Using that defect map (optionally), the inspection-R0Irs
valid pixels subset is derived from which either the local median
(in a preferred embodiment) or the average or modus signal value
is calculated.
That value represents the central signal value "C" for the
further operations performed on the histogram of valid pixel
values as depicted in Fig. 3.
Since all the image-pixels are valid in digitizer- / media-
5 based computed radiography (CR) a defect map is neither available
nor required for the detection of beam-path disturbing objects
during the inspection of the homogeneously exposed images made for
the calculation of the CR-cassette's gain-image.
10 The histogram of undisturbed image-noise, as present in the
smooth background signal of Fig.2, is quantum noise dominated and
is depicted as the dashed Gaussian-like distribution centred about
the central value C in Fig. 3.
15 The image impact of a local object disturbance in the
inspection ROI introduces x-ray absorption which reduces the
signal values of the affected pixels in a way that the histogram
becomes asymmetrical with respect to the vertical dash-dot line,
representing the central signal value without the presence of that
20 disturbing object.
The higher the object's absorption level, the lower the
average signal value of the outward bulking lower signal shoulder.
25 The higher the amount of ROI pixels affected by the object,
the bigger that secondary maximum of the local histogram.
Let's consider the lower and the higher signal values: Cdelta
and C+delta, symmetrically centered about the central signal
30 value.
Assuming +/- l%C signal-value gaps, relative to the central
signal value, a lower and a higher signal shoulder can be defined
inside the local histogram as the collections of the lower and the
higher valid pixel shoulder fractions composed of the spatially
35 distributed pixels having signal-values below C-delta or above
C+delta.
These shoulders can be expressed as fractions of the local
ROIfs valid pixels subset.
From the Shoulder Pixels Spatial Distribution view in fig.
2, it can be seen that (for this image and with this +/-l%C delta
5 ) nearly equal amounts of about spatially evenly distributed
pixels in the smooth background noise still belong to the low and
the high histogram shoulders.
Seen from the pixel positions where the beam-path disturbing
elastic band object is intersecting with the local inspection ROI
10 however it is clear that these disturbed valid pixels all belong
to the low histogram shoulder.
The histogram-gap, determined by the signal delta, that
separates the low and the high shoulders from the central value
1s can be predetermined as either a fraction of the central value
itself or as a predetermined factor times the standard deviation
of the smooth background noise, calculated as the median noise
deviation estimate from a limited set of spatially distributed
image ROIs.
20 This way the C-centered signal-gap separating both
histogram-shoulders can automatically adapt itself to the amount
of smooth background-noise present in the image under inspection.
Many experiments conducted on homogeneously exposed images
with various types of disturbing objects in the beam path have
25 shown that the combination of a sufficient level of unbalance
between the local histogram's low and high shoulder fractions with
a sufficient level of significance of at least one of both
shoulder f-:actions is a sensitive, reliable and fast indicator for
the presence of a disturbing object condition in the image under
30 inspection.
The ROI's low and high shoulder fractions L and H are
calculated as the amount of valid pixels with signals below the Cdelta
or above the C+delta value devided by the total amount of
valid pixels locally present.
The shoulder ratio is calculated by dividing the biggest of
both shoulder fractions by the smallest. Undisturbed image noise
will typically return a near equity shoulder ratio.
The results of these calculations are shown in Fig. 4.
5 As soon as an object will generate an unbalance of the
shoulder fractions, the shoulder ratio will increase.
Once the L and H shoulder fractions and their ratio R are
locally calculated, each of them is compared to its predetermined
10 threshold level.
The threshold for the shoulder fraction determines if the
measured value is sufficiently significant to be fiagged as one of
the prerequisites for object detection.
The threshold for the shoulder ratio determines if the
1s measured value is sufficiently unbalanced to be flagged as one of
the prerequisites for object detection.
The decision logic for the detection of a ROI-disturbance is
such that a local ROI is regarded as object-disturbed if a
sufficient level of shoulder unbalance is present and if at the
20 same time at least one shoulder fraction is sufficiently
significant.
The result of that local ROI decision can be stored in an
image-wide disturbance memory for further decision making
regarding the image-disturbance at a higher level.
25 In case the detection of a single ROI-disturbance would be
sufficient to regard the entire image as disturbed the further
looping through the other inspection ROIs can be skipped.
Once all the local ROIs have been investigated and their
30 logical disturbance status is known the decision about the object
disturbance of the entire image is made by comparing the
predetermined criteria with the content of the image-wide
disturbance memory.
An image-wide criterion could be that a very limited amount
of solitary disturbed ROIs can still be accepted if these isolated
disturbances all occur in ROI's adjacent to the image borders.
If these criteria aren't fulfilled the investigation of the
5 image is halted and the inspected image is accepted and added to
the set of input-images for the purpose of system (re-
)calibration.
If the image-wide disturbance criteria are met the beam path
is regarded as object disturbed. In that case additional
l o inspection, cleaning and or a correction ( e.g. the removal of the
disturbing object ) of the x-ray beam path might be necessary
before an image-retake can be performed.

WE CLAIM:
1. A method for deciding on the acceptance of gain correction data
for a direct radiography or computed radiography image by
5 evaluating whether a source of x-ray beam inhomogeneity was
present during acquisition of said gain correction data for an
x-ray detector comprising the steps of
- acquiring gain correction data by exposing said x-ray
detector to a substantially uniform x-ray beam to generate an
10 x-ray image and converting said x-ray image into a digital
image representation,
- performing a statistical analysis on pixel values of
at least one region of interest in said x-ray image,
- deciding on the acceptance of said gain correction
15 data by comparing the results of the statistical analysis with
least one predetermined acceptance criterion.
2. A method according to claim 1 wherein at least two partially
overlapping regions-of-interest are subjected to said
20 statistical analysis.
3. A method according to claim 1 wherein a single region of
interest is subjected to said statistical analysis.
25 4. A methad according to claim 1 wherein said criterion is an
image-wide criterion.
5. A method according to claim 1 wherein said statistical analysis
is only performed on non-defective pixels in a region of
30 interest.
6. A method according to claim 1 wherein said statistical analysis
comprises an evaluation of low and high histogram shoulders of
the histogram of said pixel values in said region of interest
35 determined relative to a center value in said histogram.
7. A method according to claim 6 wherein said center value is the
median, the average or the modus value of pixel values of said
region of interest.
5 8 . A method according to claim 7 wherein a decision upon the
detection of a cause of beam in-homogeneity is obtained as a
result of an evaluation of at least one significance level
being determined by thresholding low and high shoulder
fractions, said low and high shoulder fractions being
10 determined as the amount of ROI-pixels with values below a low
shoulder value divided by the total amount of POI-pixels and
the amount of ROI-pixels with values above a high shoulder
value divided by the total amount of ROI-pixels.
15 9 . A method according to claim 7 wherein a decision upon the
detection of a cause of beam in-homogeneity is obtained as a
result of an evaluation of an unbalance level of a shoulder
ratio relative to a predetermined ratio-threshold level, said
shoulder ratio being determined as a largest shoulder fraction
20 divided by a smallest shoulder fraction, wherein said low and
high shoulder fractions being the amount of ROI-pixels with
values below a low shoulder value and the amount of ROI-pixels
with values above a high shoulder value.
25 10. A method according to claim 8 or 9 whereby said low and high
shoulder values are determined as said center-value or average
or modus value minus and plus a predetermined fraction of
- said center-value, or
- the median or the average of the standard deviations
30 calculated for a subset of spatially distributed regions-ofinterest
in the image.
11. A method according to claim 8 or 9 wherein a decision upon the
detection of a cause of beam in-homogeneity is obtained as a
35 result of a logical operation performed on at least one
significance level and an unbalance level.
12. A method wherein either a non-corrected detector-image, an
offset-map corrected image, a defective-pixel-map corrected
image or a combination of both corrections is subjected to the
method of claim 1.