WO2018083073A1

WO2018083073A1 - Apparatus for noise reduction in body part imagery

Info

Publication number: WO2018083073A1
Application number: PCT/EP2017/077841
Authority: WO
Inventors: Bernhard Johannes Brendel; Frank Bergner; Thomas Koehler
Original assignee: Koninklijke Philips N.V.
Priority date: 2016-11-01
Filing date: 2017-10-31
Publication date: 2018-05-11

Abstract

The present invention relates to an apparatus (10) for noise reduction in body part imagery. It is described to provide (210) at least one image; wherein, the at least one image comprises pixel image data of a body part. At least one edge in the at least one image is detected (220), wherein at least one subset of the pixel image data is associated with the at least one edge. Noise in the at least one image is reduced (230), the reduction in noise comprising application of a regularization term to the at least one subset of the pixel image data, wherein the regularization term comprises a plurality of regularization weights. A first vector is determined (240) that is normal to a contour of the at least one edge that is associated with a first pixel of the at least one subset of the pixel image data. At least one second vector is determined (250) that extends from the first pixel to at least one second pixel of the at least one subset of the pixel image data. At least one regularization weight associated with the first pixel and the at least one second pixel is determined (260), wherein a first regularization weight associated with the first pixel and a first one of the at least one second pixel is based in part on an angle between the first vector and a second vector associated with the first one of the at least one second pixel.

Description

Apparatus for noise reduction in body part imagery

FIELD OF THE INVENTION

The present invention relates to an apparatus for noise reduction in body part imagery, to a system for noise reduction in body part imagery, and to a method for noise reduction in body part imagery, as well as to a computer program element and a computer readable medium.

BACKGROUND OF THE INVENTION

The general background of this invention relates to the field of X-ray spectral computed tomography (CT). In a CT system an X-ray source emits X-ray radiation. The emitted radiation traverses an examination region with a subject or object located within and is detected by a detector array opposite the X-ray source. The detector array detects the radiation traversing the examination region and the subject and generates projection data, e.g. raw detector data or projection images. A reconstructor processes the projection data and reconstructs a volumetric image of the subject or object. X-ray Spectral CT is an imaging modality that extends the capabilities of a conventional CT system. Dual-Energy (DE) CT, which is a specific configuration of spectral CT, utilizes two attenuation values acquired at two photon energies to solve the photoelectric and Compton contribution that consists of the mass attenuation coefficient of a material, and thus to identify an unknown material by its value of photoelectric and Compton contribution. Because the attenuation in any tissue in the human body, as well as iodine (which is often used as contrast agent in CT), can be spectrally very well approximated by a combination of photoelectric effect and Compton scatter (within the range of diagnostic x-ray energies), this is often used as a basis for material

decomposition. However, other combinations of two basis functions can also be used for material decomposition (e.g., water/bone or water/iodine), as long as they represent different combinations of photoelectric effect and Compton scatter. The basis material images provide new applications such as monochromatic image, material cancellation image, effective atomic number image and electron density image. There are several approaches to perform dual energy CT acquisition such as dual-source, fast kVp switching, and dual-layer detector configurations. In addition, quantitative imaging is one of the current major trends in the medical imaging community. Spectral CT supports this trend, as the additional spectral information improves the quantitative information that can be measured about the scanned object and its material composition.

In iterative reconstruction and denoising methods for conventional and multi- energy X-ray computer tomography (CT), regularization is included to reduce noise in the final image or images. For this, the objective function to maximize comprises in addition to the data term a regularization term, which enforces a smooth image, e.g., by penalizing the differences between neighboring image pixels. The parameter adjusting the regularization strength (i.e., the amount of smoothing) is often chosen to be constant over the whole image. Alternatively, approaches have been suggested to choose spatially varying regularization strength parameters to achieve, e.g., constant noise level in the whole image and/or isotropic noise texture.

However, there is a need to further improve de-noising in such imagery. SUMMARY OF THE INVENTION

Therefore, it would be advantageous to have an improved technology for reducing noise in conventional and CT X-ray imagery.

The object of the present invention is solved with the subject matter of the independent claims, wherein further embodiments are incorporated in the dependent claims. It should be noted that the following described aspects of the invention apply also for the apparatus for noise reduction in body part imagery, system for noise reduction in body part imagery and the method for noise reduction in body part imagery, and for the computer program element and the computer readable medium.

In a first aspect, there is provided an apparatus for noise reduction in body part imagery, comprising:

an input unit; and

a processing unit.

The input unit is configured to provide the processing unit with at least one image. The at least one image comprises pixel image data of a body part. The at least one image comprises a photoelectric effect image and a Compton scatter image. The processing unit is configured to detect at least one edge in the at least one image, wherein at least one subset of the pixel image data is associated with the at least one edge. The processing unit is configured to also determine the at least one subset of the pixel image data on the basis of pixel values in the photoelectric effect image and the Compton scattering image. The processing unit is also configured to reduce noise in the at least one image. The reduction in noise comprises application of a regularization term to the at least one subset of the pixel image data, wherein the regularization term comprises a plurality of regularization weights. For a first pixel of the at least one subset of the pixel image data, the processing unit is configured to determine a first vector that is normal to a contour of the at least one edge that is associated with the first pixel. For at least one second pixel of the at least one subset of the pixel image data, the processing unit is configured to determine at least one second vector that extends from the first pixel to the at least one second pixel. The processing unit is configured to determine at least one regularization weight associated with the first pixel and the at least one second pixel. A first regularization weight associated with the first pixel and a first one of the at least one second pixel is based in part on an angle between the first vector and a second vector associated with the first one of the at least one second pixel.

In other words, edges are detected in one or more images. Normal vectors to edges are determined, where a vector is determined that is perpendicular to an edge at the point of intersection with an edge. A regularization process is then applied to one or more images in order to de-noise the image(s), and regularization strengths for an edge pixel can be determined on the basis of the angle the edge pixel makes with neighboring pixels and a normal to the edge.

To put this another way, in an example two images can be de-noised simultaneously, but the reduction of regularization can be applied across an edge in only one of the two images.

In this manner, different regularization strengths can be applied across one or more image to de-noise the image(s), whilst the regularization strength can be reduced at edges in the image(s), whilst the regularization strength can be maintained at higher strengths at other parts of image(s).

This provides for improved de-noising within one or more images, including that associated with edges in the image(s).

To put this another way, de-noising in an image is provided, whilst avoiding smoothing of the edges, because the regularization strength can take account of (and hence be reduced) for certain directions associated with an edge, and crosstalk artefacts that can occur when several images are de-noised simultaneously is mitigated.

In the first aspect, a second regularization weight associated with the first pixel and a second one of the at least one second pixel is smaller than the first regularization weight if the angle between the first vector and a second vector associated with the second one of the at least one second pixel is smaller than the angle between the first vector and a second vector associated with the first one of the at least one second pixel.

In other words, the regularization strength can be reduced with respect to pixels that are tending towards being orthogonal to an edge, for which a strong gradient of the pixel values is expected.

To put this another way, the regularization strength is reduced only for certain neighboring pixels, that are tending towards being orthogonal to edges in the imagery.

In this way, noise across the image can be reduced, while at the same time the integrity of edges maintained.

In an example, the first regularization weight associated with the first pixel and a first one of the at least one second pixel is calculated based in part on a cosine of the angle between the first vector and the second vector associated with the first one of the at least one second pixel.

In this way, a simple and easily implemented way is provided to reduce the regularization weight or strength for pixels depending upon the angle to an edge.

In an example, the processing unit is configured to determine the at least one regularization weight associated with the first pixel and the at least one second pixel based in part on the distance of the first pixel to the detected edge it is associated with.

In other words, the distance a pixel is away from an edge can be taken into account when apply a regularization term to an image. In an example, the regularization weight is reduced as the distance between a pixel and an edge increases.

In an example, the processing unit is configured to determine the at least one subset of the pixel image data.

In an example, the determination comprises determining at least one intensity gradient between at least two pixels.

In other words, the directional noise reduction processing can be applied to areas of the edges in imagery, where a strong gradient of pixel values is apparent. This reduces artefacts at edges, whilst keeping a certain amount of de-noising.

In an example, the regularization term comprises a plurality of intensity factors, wherein at least one intensity factor associated with the first pixel of the at least one subset and the at least one second pixel of the at least one subset is calculated on the basis of a value of the first pixel and at least one value of the at least one second pixel of the at least one subset. In an example, a first intensity factor associated with the first pixel and the first one of the at least one second pixel is based in part on the difference in the value of the first pixel and a value of the at least one second pixel.

In an example, prior to detection of the at least one edge, at least one duplicate of the at least one image has been subjected to de-noising processing, and an edge detection algorithm is applied to the at least one duplicate image to detect at least one position of at least one edge, and wherein information relating to the position of the at least one edge in the at least one duplicate image is useable to detect the at least one edge in the at least one image. In this manner, a more robust edge detection is provided on the basis of a mirror image that has been de-noised. Then noise reduction provided on the basis of directional regularization can be applied to the raw original data.

In an example, application of the regularization term comprises a summation of the at least one regularization weight associated with the first pixel.

In other words, all the neighboring pixels are taken into account, regularization strength orthogonal to edges is reduced such that the integrity of the edge is maintained whilst providing for a higher degree of noise reduction elsewhere in the imagery.

In an example, the at least one image comprises a first photo image and a second scatter image.

In other words, a pair of images comprising a photoelectric effect image and Compton scattering image pair are simultaneously processed. Directional regularization noise processing on the basis of regularization weighting that takes into account edges is provided, and crosstalk artefacts that can appear when several images are de-noised simultaneously are mitigated. Thus, the apparatus has utility in the processing of imagery from multi-energy computed tomography systems. The apparatus also has utility for imagery derived from conventional computed tomography systems.

In a second aspect, there is provided a system for noise reduction in body part imagery, comprising:

an image acquisition unit;

an apparatus for noise reduction in body part imagery according to the first aspect; and

an output unit.

The image acquisition unit is configured to acquire the at least one image. The processing unit is configured to generated a processed at least one image on the basis of the noise reduced in the at least one image. The output unit is configured to output data representative of the processed at least one image.

In a third aspect, there is provided a method for noise reduction in body part imagery, the method comprising:

a) providing at least one image; wherein, the at least one image comprises pixel image data of a body part;

c) detecting at least one edge in the at least one image, wherein at least one subset of the pixel image data is associated with the at least one edge;

d) reducing noise in the at least one image, the reduction in noise comprising application of a regularization term to the at least one subset of the pixel image data, wherein the regularization term comprises a plurality of regularization weights;

e) determining a first vector that is normal to a contour of the at least one edge that is associated with a first pixel of the at least one subset of the pixel image data;

f) determining at least one second vector that extends from the first pixel to at least one second pixel of the at least one subset of the pixel image data; and

g) determining at least one regularization weight associated with the first pixel and the at least one second pixel, wherein a first regularization weight associated with the first pixel and a first one of the at least one second pixel is based in part on an angle between the first vector and a second vector associated with the first one of the at least one second pixel.

In an example, the method comprises iterating steps e-g for the pixels of the at least one subset of the pixel image data.

In other words, in a first iteration, a pixel, for example pixel 1 , is considered as the first pixel, and other pixels, 2, 3, 4, ...n etc. are the at least one second pixel, and steps d-f are carried out. Thus, regularization weights are determined with respect to pixels pairs, 1-2, 1-3, 1-4, 1-n etc. Then, in a second iteration pixel 2 is the first pixel, and pixels 1, 3, 4, ... n are the at least one second pixel. Thus, regularization weights are determined with respect to pixels pairs, 2-1, 2-3, 2-4, 2-n etc. Then in a further iteration the first pixel is pixel 3 etc., and in the next iteration the first pixel is pixel 4 etc., until the first pixel is pixel n, and all the pixels of the subset have been considered.

In this manner, iterative reconstruction and de-noising of one or more images is provided on the basis of directional regularization, that preserves edge based information through an appropriate reduction of regularization weights associated with edges, whilst enabling other regions of the imagery to be subjected to higher degrees of de-noising. According to another aspect, there is provided a computer program element, which when executed by a processing unit, is adapted to perform the method steps as previously described.

According to another aspect, there is provided a computer program element controlling apparatus as previously described which, when the computer program element is executed by a processing unit, is adapted to perform the method steps as previously described.

According to another example, there is provided a computer readable medium having stored a computer element as previously described.

Advantageously, the benefits provided by any of the above aspects and examples equally apply to all of the other aspects and examples and vice versa.

The above aspects and examples will become apparent from and be elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

Exemplary embodiments will be described in the following with reference to the following drawings:

Fig. 1 shows a schematic representation of an example of an apparatus for noise reduction in body part imagery;

Fig. 2 shows a schematic representation of an example of a system for noise reduction in body part imagery;

Fig. 3 shows an example of a method for noise reduction in body part imagery;

Fig. 4 shows imagery associated with edge detection processing for an example of an apparatus for noise reduction in body part imagery;

Fig. 5 shows imagery associated with the calculation of regularization weights for an example of an apparatus for noise reduction in body part imagery; and

Fig. 6 shows imagery associated with conventional de-noising, imagery associated with an example of an apparatus for noise reduction in body part imagery, and idealized "Ground Truth" imagery.

DETAILED DESCRIPTION OF EMBODIMENTS

Fig. 1 shows an example of an apparatus 10 for noise reduction in body part imagery. The apparatus 10 comprises an input unit 20, and a processing unit 30. The input unit 20 is configured to provide the processing unit 30 with at least one image. The at least one image comprises pixel image data of a body part. The processing unit 30 is configured to detect at least one edge in the at least one image, wherein at least one subset of the pixel image data is associated with the at least one edge. The processing unit 30 is also configured to reduce noise in the at least one image. The reduction in noise comprises application of a regularization term to the at least one subset of the pixel image data, wherein the

regularization term comprises a plurality of regularization weights. For a first pixel of the at least one subset of the pixel image data, the processing unit 30 is configured to determine a first vector that is normal to a contour of the at least one edge that is associated with the first pixel. For at least one second pixel of the at least one subset of the pixel image data, the processing unit 30 is configured to determine at least one second vector that extends from the first pixel to the at least one second pixel. The processing unit 30 is also configured to determine at least one regularization weight associated with the first pixel and the at least one second pixel. A first regularization weight associated with the first pixel and a first one of the at least one second pixel is based in part on an angle between the first vector and a second vector associated with the first one of the at least one second pixel.

In an example, the at least one image is provided as pairs of identical images. In this example, one of the pairs of images is de-noised and an edge is detected in this de- noised image. The other image can then be considered to be a raw image, and has not been de-noised. However, localization of the edge in the de-noised image then enables the edge to be detected in the raw image that has not be de-noised. In this way, a de-noised image can be used to detect an edge, and then noise reduction in the original raw image data can be performed by application of a regularization term to the raw data. In other words, the reduction in noise in one or more images can progress as follows: Initial denoising of original data. Edge detection on initially demised data; and using this edge information for denoising with reduced regularization of edges on the original data. In this way, a regularization strength from edge pixel can be determined primarily only on the basis of certain neighboring pixels, because the regularization strength associated with other pixels can be reduced to the extent that those other pixels are not taken into account in the process. This means that a general reduction of regularization strength at an edge is not applied, which would lead to a significant increase of the noise levels at the edges, rather a targeted reduction in

regularization strength is undertaken taking into account the angulation of pixels to an edge, and in this way edge features can be maintained and noise levels optimized at the same time.

In an example, the at least one subset of the pixel image data is all the pixels of the at least one image. According to an example, a second regularization weight associated with the first pixel and a second one of the at least one second pixel is smaller than the first regularization weight if the angle between the first vector and a second vector associated with the second one of the at least one second pixel is smaller than the angle between the first vector and a second vector associated with the first one of the at least one second pixel.

In an example, the absolute value of the angle is utilized. For example, the regularization weight can be proportional to the angle (that is always positive), and tend to zero as the angle tends to zero. In an example, the regularization weight can be proportional to the absolute value of the sine of the angle, under such tend towards the absolute angle for small angles. In an example, the realization weight can be based on 1- minus the absolute value of the cosine of the angle, or 1 minus the squared value of the cosine of the angle.

According to an example, the first regularization weight associated with the first pixel and a first one of the at least one second pixel is calculated based in part on a cosine of the angle between the first vector and the second vector associated with the first one of the at least one second pixel.

According to an example, the processing unit is configured to determine the at least one regularization weight associated with the first pixel and the at least one second pixel based in part on the distance of the first pixel to the detected edge it is associated with.

In an example, the processing unit is configured to determine the at least one regularization weight associated with the first pixel and the at least one second pixel based in part on the length of the second vector. In other words, based on the distance between the first pixel and the second pixel(s).

In an example, the first regularization weight associated with the first pixel and a first one of the at least one second pixel is calculated based on an absolute value of a cosine of the angle between the first vector and the second vector associated with the first one of the at least one second pixel being subtracted from the number one.

In other words, the regularization weight is proportional to 1- abs(cos(9)).

In this way, as the angle tends to zero the regularization weight also tends to zero, thus directional regularization is provided at edges in imagery providing for the reduction in regularization weight normal to the edge, enabling edges not to be smoothed out through de-noising whilst providing for de-noising in the wider imagery.

According to an example, the processing unit 30 is configured to determine the at least one subset of the pixel image data. In an example, the at least one subset of the pixel image data is determined on the basis of there being expected artefacts at edges.

In an example, the at least one image comprises a photoelectric effect image and a Compton scattering image, and the processing unit is configured to determine the at least one subset of the pixel image data on the basis of pixel values in the photoelectric image and the Compton scattering image. In this way, the values in a photoelectric image and the Compton scattering image at the same location in the images can be used to determine the anticipated or expected material at that location, and at regions in the image where edges exists, or in the regions of those edges, the edge regions to be processed can be determined on the basis of the material at that location, determined from the photoelectric image values. In an example, the ratio between the pixel values at a location in the photoelectric image and the pixel values at a corresponding location in the Compton scattering image are used to determine a material. This is because the value of this ratio can be indicative of a certain material.

According to an example, the determination of the at least one subset of the pixel image data comprises determining at least one intensity gradient between at least two pixels.

In an example, when the at least one image comprises a pair of images (e.g., a photoelectric effect image - relating to attenuation due to the photoelectric effect and a Compton scattering image - relating to attenuation due to Compton scattering), then the structure in the image pair can be anti-correlated. Then a pair of pixels in one image that have an intensity gradient in one direction in one image will have an intensity gradient in the opposite direction in the other image, and this can be used to determine regions of edges to select in order to have the regularization weighting processing applied.

In an example, the at least one subset of the pixel imaged data is determined on the basis of the at least one intensity gradient between at least two pixels being greater than a threshold level.

In this way, simple determination can be made as to which parts of the imagery should be processed.

According to an example, the regularization term comprises a plurality of intensity factors. At least one intensity factor associated with the first pixel of the at least one subset and the at least one second pixel of the at least one subset is calculated on the basis of a value of the first pixel and at least one value of the at least one second pixel of the at least one subset. According to an example, a first intensity factor associated with the first pixel and the first one of the at least one second pixel is based in part on the difference in the value of the first pixel and a value of the first one of the at least one second pixel.

In an example, detection of the at least one edge comprises utilization of a canny edge detector. Other edge detection algorithms can be utilized.

According to an example, prior to detection of the at least one edge, at least one duplicate of the at least one image has been subjected to de-noising processing. An edge detection algorithm is applied to the at least one duplicate image to detect at least one position of at least one edge, and wherein information relating to the position of the at least one edge in the at least one duplicate image is useable to detect the at least one edge in the at least one image.

According to an example, application of the regularization term comprises a summation of the at least one regularization weight associated with the first pixel.

According to an example, the at least one image comprises a first photo image and a second scatter image.

Fig. 2 shows an example of a system 100 for noise reduction in body part imagery. The system 100 comprises an image acquisition unit 110, an apparatus 10 for noise reduction in body part imagery as described with reference to Fig. 1, and an output unit 120. The image acquisition unit 110 is configured to acquire the at least one image. The processing unit 30 is configured to generated a processed at least one image on the basis of the noise reduced in the at least one image. The output unit 120 is configured to output data representative of the processed at least one image.

In an example, the acquisition unit is an X-ray scanner such as a C-arm scanner. In an example, the acquisition unit is a tomography X-ray scanner.

In an example, the output unit is configured to display image data, for example displaying images on one or more monitors such as one or more visual display units VDUs. In an example, the input unit 20 is the image acquisition unit 110.

Fig. 3 shows a method 200 for noise reduction in body part imagery in its basic steps. The method 200 comprises:

a providing step 210, also referred to as step a), providing at least one image; wherein, the at least one image comprises pixel image data of a body part;

a detecting step 220, also referred to as step c), detecting at least one edge in the at least one image, wherein at least one subset of the pixel image data is associated with the at least one edge; a noise reducing step 230, also referred to as step d), reducing noise in the at least one image, the reduction in noise comprising application of a regularization term to the at least one subset of the pixel image data, wherein the regularization term comprises a plurality of regularization weights;

a determining step 240, also referred to as step e), determining a first vector that is normal to a contour of the at least one edge that is associated with a first pixel of the at least one subset of the pixel image data;

a determining step 250, also referred to as step f), determining at least one second vector that extends from the first pixel to at least one second pixel of the at least one subset of the pixel image data; and

a determining step 260, also referred to as step g), determining at least one regularization weight associated with the first pixel and the at least one second pixel, wherein a first regularization weight associated with the first pixel and a first one of the at least one second pixel is based in part on an angle between the first vector and a second vector associated with the first one of the at least one second pixel.

In an example, in step g), a second regularization weight associated with the first pixel and a second one of the at least one second pixel is determined to be smaller than the first regularization weight if the angle between the first vector and a second vector associated with the second one of the at least one second pixel is smaller than the angle between the first vector and a second vector associated with the first one of the at least one second pixel.

In an example, in step g) the first regularization weight associated with the first pixel and a first one of the at least one second pixel is calculated based on a cosine of the angle between the first vector and the second vector associated with the first one of the at least one second pixel.

In an example, in step g) the first regularization weight associated with the first pixel and a first one of the at least one second pixel is calculated based on a cosine of the angle between the first vector and the second vector associated with the first one of the at least one second pixel being subtracted from the number one.

In an example, step c) comprises determining the at least one subset of the pixel image data, the determination comprising determining at least one intensity gradient between at least two pixels. In an example, in step c) the at least one subset of the pixel imaged data is determined on the basis of the at least one intensity gradient between at least two pixels being greater than a threshold level.

In an example, regarding step d) the regularization term comprises a plurality of intensity factors, wherein at least one intensity factor associated with the first pixel of the at least one subset and the at least one second pixel of the at least one subset is calculated on the basis of a value of the first pixel and at least one value of the at least one second pixel of the at least one subset.

In an example, regarding step d) a first intensity factor associated with the first pixel and the first one of the at least one second pixel is based on the difference in the value of the first pixel and a value of the first one of the at least one second pixel.

In an example, the method comprises step b), subjecting 270 the at least one image to de-noising processing.

In an example, step d) comprises summing of the at least one regularization weight associated with the first pixel.

The apparatus, system and method for noise reduction in body part imagery as described with respect to Figs 1-3, is now described in more detail with respect to Figs 4-6.

In iterative reconstruction and de-noising methods for conventional and multi- energy CT a regularization term is applied to the image to achieve noise reduction.

Regularization terms are commonly parameterized with a global regularization strength parameter (leading to a constant regularization strength in the whole image) or with a spatial resolved regularization strength parameter (to consider effects as e.g. increasing noise levels in the center of the patient).

Since the regularization term cannot distinguish between noise and object edges the application of a regularization term for denoising leads to certain artifacts in the demised images:

edges are smoothed. This can be reduced to a certain degree by using LI -norm based regularizers, but such smoothing cannot be completely avoided;

the contrast between neighboring object regions is reduced; and if multiple images with different image content are de-noised simultaneously

(e.g., photo and scatter images in multi-energy CT), artifacts occur at some edges leading to a smoothing of the edge in one image and at the same time a sharpening of the same edge in another image. The presently described apparatus, system and method for noise reduction in body part imagery with respect to Figs 1-6 address these issues.

It is to be noted that a general reduction of the regularization strength at edges could only help in addressing the above issues if the regularization strength would be substantially reduced. However, this has been found to be disadvantageous because it leads to a significant increase in the noise level at the edges. Therefore, in contrast to a general reduction of the regularization strength at edges, i.e., reducing for an edge pixel the regularization strength for all neighboring pixels considered in the regularization term, a different approach is taken in the apparatus, system and method for noise reduction in body part imagery. In this approach the regularization strength is reduced only for certain neighboring pixels, i.e., neighboring pixels orthogonal to the edge, for which a strong gradient of the pixel values is expected. This reduces artifacts at edges, while keeping a certain amount of de-noising.

The following describes, in combination with the imagery in Figs 4-6, a detailed example of the described apparatus, system and method for noise reduction in body part imagery.

A description of a regularization term to be applied to imagery is given by:

The regularization term runs over all pixels j in the image. For each pixel j, certain neighboring pixels k are chosen. The difference between the values of the neighboring pixels (jUj^k) is taken (or acquired), and transformed with the potential function ψ and weighted with the weighting factor Wkj. This weighting can be done for each neighbor of each pixel individually, allowing the regularization strength to be reduced only for certain directions with respect to the "central" pixel j. Here a framework is presented to exploit this to reduce regularization artifacts at object edges.

Thus, in a first step, object edges are calculated (or detected), e.g. with a Canny edge detector. There are many other approaches for edge detection available which could also be used. To improve the robustness of edge detection, pre-de-noised images can be used as input, however this is not required. Artifacts at edges may occur in such pre- denoising if it is applied, but are not a drawback for this step as long as the edge is still detectable. The edge detection processing is illustrated in Fig. 4, based on a pair of a photo and scatter images, which should be de-noised simultaneously. In Fig. 4 the input pair of photo and scatter images are de-noised (as shown in the left column). For both images, gradients in pixel intensities are calculated (as shown in the second column from the left). The gradients are combined by summing up the absolute values of the gradients (as shown in the third column from the left). Then edge detection processing is carried out, which in this example was carried out with a conventional Canny edge detector, leading to the resulting image as shown on the right.

The output of the edge detector is used to calculate the regularization weights Wjk, which is illustrated in Fig. 5. This is done in this example by using one minus the absolute value of the cosine of the angle between the normal vector of the detected edge and the vector between the central pixel and the neighboring pixel. Other approaches can be utilized that lead to a minimization of regularization weight as a pixel tends towards being orthogonal with an edge, such as using the absolute angle value, the sine of the angle etc. In Fig. 5 images relating to the calculation of regularization weights Wjk from the detected object edges (as shown in the left image, which corresponds to the right hand image shown in fig 4). Color coding of weights is black for a value of one and white for a value of zero. In the middle image of Fig. 5, regularization weights are shown exemplarily for neighboring pixels to the bottom right of the central pixel. As can be seen weights are zero (white) if the neighboring pixel on the bottom right is expected to be positioned orthogonal to an edge with respect to the central pixel. The right image shown in Fig. 5 shows another example, where neighboring pixels are to the top right of the central pixel.

Using these weights in a simultaneous denoising of both images leads to the results shown in Fig. 6. Simultaneous de-noising of a pair of images is shown, were de- nosing of a scatter image is shown in the top row and de-noising of a photo image is shown in the bottom row. In the left hand column of Fig. 6 the results of conventional de-noising is shown. In the middle column of Fig. 6 de-noising using the present method for noise reduction is shown, and in the right hand column "Ground truth" or perfect de-noising is shown. Referring to the middle column of Fig. 6, it can be seen that the edges within the five large circular inhomogeneities are much better preserved than that achievable with conventional de-noising. Also, quantitative values within the half-circles are much better reproduced using the presently described method of noise reduction.

In addition to simultaneously de-noising a pair of images, applicable for example to multi-energy CT imaging, application to one image or indeed more than two images can be undertaken. Only a subset of the edges can be processed, these being those that are likely to show artifacts in the de-noised image. Thus, the apparatus, system and method for noise reduction in body part imagery reduces the regularization strength for certain directions (i.e., orthogonal to edges) to reduce smoothing of edges and crosstalk artifacts, which can be exacerbated when several images are de-noised simultaneously, for example multi-energy CT.

In another exemplary embodiment, a computer program or computer program element is provided that is characterized by being configured to execute the method steps of the method according to one of the preceding embodiments, on an appropriate system.

The computer program element might therefore be stored on a computer unit, which might also be part of an embodiment. This computing unit may be configured to perform or induce performing of the steps of the method described above. Moreover, it may be configured to operate the components of the above described apparatus. The computing unit can be configured to operate automatically and/or to execute the orders of a user. A computer program may be loaded into a working memory of a data processor. The data processor may thus be equipped to carry out the method according to one of the preceding embodiments.

This exemplary embodiment of the invention covers both, a computer program that right from the beginning uses the invention and computer program that by means of an update turns an existing program into a program that uses the invention.

Further on, the computer program element might be able to provide all necessary steps to fulfill the procedure of an exemplary embodiment of the method as described above.

According to a further exemplary embodiment of the present invention, a computer readable medium, such as a CD-ROM, is presented wherein the computer readable medium has a computer program element stored on it which computer program element is described by the preceding section.

A computer program may be stored and/or distributed on a suitable medium, such as an optical storage medium or a solid state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the internet or other wired or wireless telecommunication systems.

However, the computer program may also be presented over a network like the

World Wide Web and can be downloaded into the working memory of a data processor from such a network. According to a further exemplary embodiment of the present invention, a medium for making a computer program element available for downloading is provided, which computer program element is arranged to perform a method according to one of the previously described embodiments of the invention.

It has to be noted that embodiments of the invention are described with reference to different subject matters. In particular, some embodiments are described with reference to method type claims whereas other embodiments are described with reference to the device type claims. However, a person skilled in the art will gather from the above and the following description that, unless otherwise notified, in addition to any combination of features belonging to one type of subject matter also any combination between features relating to different subject matters is considered to be disclosed with this application.

However, all features can be combined providing synergetic effects that are more than the simple summation of the features.

While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing a claimed invention, from a study of the drawings, the disclosure, and the dependent claims.

In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items re-cited in the claims. The mere fact that certain measures are re-cited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting the scope.

Claims

CLAIMS:

1. An apparatus (10) for noise reduction in body part imagery, comprising:

an input unit (20); and

a processing unit (30);

wherein, the input unit is configured to provide the processing unit with at least one image;

wherein, the at least one image comprises pixel image data of a body part, and wherein the at least one image comprises a photoelectric effect image and a Compton scatter image;

wherein, the processing unit is configured to detect at least one edge in the at least one image, wherein at least one subset of the pixel image data is associated with the at least one edge, and wherein the processing unit is configured to determine the at least one subset of the pixel image data on the basis of pixel values in the photoelectric effect image and the Compton scattering image;

wherein, the processing unit is configured to reduce noise in the at least one image, the reduction in noise comprising application of a regularization term to the at least one subset of the pixel image data, wherein the regularization term comprises a plurality of regularization weights;

wherein, for a first pixel of the at least one subset of the pixel image data, the processing unit is configured to determine a first vector that is normal to a contour of the at least one edge that is associated with the first pixel;

wherein, for at least one second pixel of the at least one subset of the pixel image data, the processing unit is configured to determine at least one second vector that extends from the first pixel to the at least one second pixel; and

wherein, the processing unit is configured to determine at least one regularization weight associated with the first pixel and the at least one second pixel, wherein a first regularization weight associated with the first pixel and a first one of the at least one second pixel is based in part on an angle between the first vector and a second vector associated with the first one of the at least one second pixel, and wherein a second regularization weight associated with the first pixel and a second one of the at least one second pixel is smaller than the first regularization weight if the angle between the first vector and a second vector associated with the second one of the at least one second pixel is smaller than the angle between the first vector and a second vector associated with the first one of the at least one second pixel.

2. Apparatus according to claim 1, wherein the first regularization weight associated with the first pixel and a first one of the at least one second pixel is calculated based in part on a cosine of the angle between the first vector and the second vector associated with the first one of the at least one second pixel.

3. Apparatus according to any of claims 1-2, wherein, the processing unit is configured to determine the at least one regularization weight associated with the first pixel and the at least one second pixel based in part on the distance of the first pixel to the detected edge it is associated with.

4. Apparatus according to claim 1, wherein, the determination comprises determining at least one intensity gradient between at least two pixels.

5. Apparatus according to any of claims 1-4, wherein the regularization term comprises a plurality of intensity factors, wherein at least one intensity factor associated with the first pixel of the at least one subset and the at least one second pixel of the at least one subset is calculated on the basis of a value of the first pixel and at least one value of the at least one second pixel of the at least one subset.

6. Apparatus according to claim 5, wherein a first intensity factor associated with the first pixel and the first one of the at least one second pixel is based in part on the difference in the value of the first pixel and a value of the first one of the at least one second pixel.

7. Apparatus according to any of claims 1-6, wherein prior to detection of the at least one edge, at least one duplicate of the at least one image has been subjected to de- noising processing, and an edge detection algorithm is applied to the at least one duplicate image to detect the position of the at least one edge, and wherein information relating to the position of the at least one edge in the at least one duplicate image is useable to detect the at least one edge in the at least one image.

8. Apparatus according to any of claims 1-7, wherein application of the regularization term comprises a summation of the at least one regularization weight associated with the first pixel.

9. A system (100) for noise reduction in body part imagery, comprising:

an image acquisition unit (110);

an apparatus (10) for noise reduction in body part imagery according to any preceding claim; and

an output unit (120);

wherein, the image acquisition unit is configured to acquire the at least one image;

wherein, the processing unit (30) is configured to generated a processed at least one image on the basis of the noise reduced in the at least one image; and

wherein, the output unit is configured to output data representative of the processed at least one image.

10. A method (200) for noise reduction in body part imagery, comprising:

a) providing (210) at least one image; wherein, the at least one image comprises pixel image data of a body part, and wherein the at least one image comprises a photoelectric effect image and a Compton scatter image;

c) detecting (220) at least one edge in the at least one image, wherein at least one subset of the pixel image data is associated with the at least one edge, and wherein the at least one subset of the pixel image data is determined on the basis of pixel values in the photoelectric effect image and the Compton scattering image;

d) reducing (230) noise in the at least one image, the reduction in noise comprising application of a regularization term to the at least one subset of the pixel image data, wherein the regularization term comprises a plurality of regularization weights;

e) determining (240) a first vector that is normal to a contour of the at least one edge that is associated with a first pixel of the at least one subset of the pixel image data; f) determining (250) at least one second vector that extends from the first pixel to at least one second pixel of the at least one subset of the pixel image data; and

g) determining (260) at least one regularization weight associated with the first pixel and the at least one second pixel, wherein a first regularization weight associated with the first pixel and a first one of the at least one second pixel is based in part on an angle between the first vector and a second vector associated with the first one of the at least one second pixel, and wherein a second regularization weight associated with the first pixel and a second one of the at least one second pixel is determined to be smaller than the first regularization weight if the angle between the first vector and a second vector associated with the second one of the at least one second pixel is smaller than the angle between the first vector and a second vector associated with the first one of the at least one second pixel .

11. Method according to claim 11 , wherein the method comprises iterating steps e-g for the pixels of the at least one subset of the pixel image data.

12. A computer program element, which when executed by a processor is configured to carry out the method of any one of claims 10 to 11.