JP2008206968A - Image processing device - Google Patents

Abstract

In a noise reduction process due to the nature of a digital image for medical diagnosis, a relatively uniform noise reduction process is realized for an image whose S / N ratio varies locally, or the pixel size is compared. To remove artifacts caused by noise reduction processing on rough images.
An image processing apparatus includes a storage unit that stores data of a digital image, a rotation processing unit that generates a plurality of rotated digital images having different rotation angles from the digital image, and a plurality of rotated images. Image processing units 19 and 21 that generate a plurality of image processed digital images from a digital image, and a rewind processing unit 15 that generates a plurality of rewinded digital images from a plurality of image processed digital images. And a synthesis processing unit 15 for synthesizing a plurality of unwound digital images into one digital image.
[Selection] Figure 1

Description

  The present invention applies to digital images. In particular, an image processing apparatus that is applied to noise reduction processing of an image in which a ratio (S / N ratio) of a signal component and a noise component changes locally, typically a medical diagnosis image such as a nuclear medicine image, CT image, or MRI image About.

  Noise reduction in digital images has been implemented by blocking (removing) high-frequency components with a high-frequency blocking filter such as a Butterworth filter or a Gauss filter. However, since the same high-frequency component is blocked over the entire image, there is no problem with general digital images (such as landscape images with a digital camera). There was a problem of degrading.

  The first cause of this problem is that the ratio (S / N ratio) of the signal component and the noise component differs for each local area (minimum unit is for each pixel) in the medical diagnostic image. This is because in the medical diagnostic image, the S / N ratio changes as statistical noise depending on the collection count obtained for each pixel. For this reason, if the same high frequency component is cut off over the entire image, overcorrection and undercorrection are mixed depending on the location. As a result, there were areas where information (positional resolution, contrast, etc.) dropped (overcorrection) and areas where noise removal was insufficient (undercorrection).

The second cause is that the medical diagnostic image has a coarse (large) pixel size. For example, the position resolution of a nuclear medicine image is about 10 mm, but the pixel size of the image representing it is several mm. When the filter process is performed by Fourier transform, the pixel size is coarse, so that sufficient sampling may not be performed and an artifact may occur. See Patent Document 1.
JP 2001-59872 A

  An object of the present invention is to realize a relatively uniform noise reduction process for an image in which the S / N ratio varies locally in a noise reduction process due to the nature of a digital image for medical diagnosis, or a pixel An object of the present invention is to realize removal of artifacts caused by noise reduction processing on an image having a relatively coarse size.

  One aspect of the present invention includes a storage unit that stores digital image data, a rotation processing unit that generates a plurality of rotated digital images having different rotation angles from the digital image, and a plurality of rotated digital images. An image processing unit for generating a plurality of image processed digital images, a rewinding processing unit for generating a plurality of unwound digital images from the plurality of image processed digital images, and a plurality of rewinds An image processing apparatus is provided that includes a synthesis processing unit that synthesizes a digital image into one digital image.

  According to the present invention, in noise reduction processing due to the properties of a digital image for medical diagnosis, it is possible to realize relatively uniform noise reduction processing for an image whose S / N ratio varies locally, or pixels It is possible to realize removal of artifacts caused by the noise reduction processing on an image having a relatively coarse size.

Embodiments of an image processing apparatus according to the present invention will be described below with reference to the drawings.
The image processing apparatus according to the present embodiment stores or generates digital medical image data such as PACS, X-ray computed tomography apparatus (CT), magnetic resonance imaging apparatus (MRI), and X-ray diagnostic apparatus via the interface 10. Connected to an external device. An image storage unit 13 is provided for storing digital image data to be image processed received from these external devices via the interface 10. In the interface 10 and the image storage unit 13, a control unit 11 that controls operation of the entire apparatus via the data / control bus 12 includes an image processing unit 15, a cutoff frequency calculation unit 17, a wavelet transform processing unit 19, and filter processing. It is connected together with the part 21.

  The image processing unit 15 has a function of rotating the digital image to be processed according to the rotation angle instructed from the control unit 11 and adds a plurality of digital images filtered by the filter processing unit 21 in accordance with the instruction from the control unit 11 On average it has the function of generating the final filtered digital image. As will be described later, the cutoff frequency calculation unit 17 calculates the cutoff frequency for each local region of the digital image based on the standard deviation of each local region.

  The wavelet transform processing unit 19 performs wavelet transform processing on the digital image data to be image processed. The wavelet transform process is a process for expressing a digital image in a frequency space while retaining the spatial information of the original digital image (original digital image). The filter processing unit 21 blocks high frequency components exceeding the cutoff frequency calculated for each local region by the cutoff frequency calculating unit 17 for each local region with respect to the digital image subjected to the wavelet transform process. The filtered digital image is returned to the original real space region by inverse wavelet transform processing in the wavelet transform processing unit 19. A plurality of digital images that have been rotated at different angles and passed through the filter are added and averaged by the image processing unit 15.

  The wavelet transform process can be replaced with a Fourier transform process in the same frequency analysis process category.

  In the above description, it is specified that the image processing includes rotation processing and addition averaging processing. However, the image processing is not limited to this, and the image processing includes processing such as rotation processing, addition averaging processing, frequency analysis processing, and filter processing. It has a broad meaning including at least one.

First, the outline of the wavelet transform process will be described. The principle of wavelet transform is defined by the following equation as is well known.

  In the present embodiment, as shown in FIG. 3, the wavelet transform is a two-dimensional process because the object is an image. As is well known, the Fourier transform converts everything into frequency components, so that spatial information is lost. However, wavelet transform can be expressed in frequency space while maintaining spatial information. For example, by two-time two-dimensional wavelet transform, each component is divided into a low frequency vertical component, a low frequency horizontal component, a low frequency diagonal component, a high frequency vertical component, a high frequency horizontal component, and a high frequency diagonal component. Is displayed.

  FIG. 2 shows a filter processing procedure according to the present embodiment. First, in step S 10, the “cutoff frequency for each local region” used for noise reduction processing (filter processing) on the digital image after wavelet transform is performed on the input data (original digital image data) by the cutoff frequency calculation unit 17. Calculated.

  5A and 5B show the cutoff frequency calculation process. In the conventional wavelet transform, noise reduction has no idea of locality, and one cut-off frequency is applied to the entire image. A typical example is Donoho's method, which determines the cutoff frequency from the standard deviation of the entire image. In the present embodiment, first, a local region is determined in order to obtain local information, and an index indicating the S / N ratio in the local region is considered. Here, a square with a length of several pixels is adopted as the local region. The local cutoff frequency is determined from the coefficient of variation C.V. value in this region. By performing the processing in the local region throughout the entire image, the cut-off frequency for each local region in the entire image can be determined.

Specifically, a plurality of local regions are set for the original digital image. The cutoff frequency Z is calculated individually for each of the plurality of local regions based on the standard deviation SD with each local region as the target range. More specifically, it is represented by the following formula.
Z = f (CV) × Coef
CV: coefficient of variation
Coef; coefficient
f (CV); SD × (2 × ln (n))
n: Number of pixels in the local region Returning to FIG. 3, in step S11, the image processing unit 15 rotates the original digital image to be processed around the center of the image. A process of rotating the specified angle is performed, wavelet transform, filter process (noise reduction process), and wavelet inverse transform are performed. Finally, the image is reversely rotated by the same amount and the image angle is restored. This process is performed every integer multiple of the angle specified between 0 degrees and 360 degrees. Since the target image is a square matrix, a rotation of 90 degrees can be equivalent to a rotation of 360 degrees. An averaged image is created using all images processed at each angle. By this processing, artifacts generated by noise reduction processing in an image with a coarse pixel size can be removed.

As the rotation processing, as shown in FIG. 7A, the image may be rotated while fixing the coordinate system, or as shown in FIG. 7B, the coordinate axis used when performing the wavelet transform may be rotated. The rotation pitch is typically set to 5 degrees as shown in FIG. The rotation pitch R is set to an angle arbitrarily selected from the range of 0 <R ≦ 45 °. Preferably, the rotation pitch R is set to around 5 ° (3 ≦ R ≦ 10 °).
Further, the reference angle θ of the rotation pitch R is defined as follows. The rotation pitch R is preferably set to n × θ. As shown in FIG. 11, n is a positive integer and is equal to or less than ½ of the number M of pixels parallel to the X axis. n should typically be set to 3 or 5 from the balance between the processing amount and the processing effect in a trade-off relationship.
tanθ = L2 / L1
L1: Distance from the image center to the edge of the endmost pixel on the Y-axis
L2: Side length of a single pixel Note that the rotation angle is zero degrees at the first time, that is, the digital image is not rotated. The wavelet transformation processing unit 19 applies wavelet transformation processing to the rotated digital image (step S12).

  Each of the plurality of local regions of the digital image subjected to the wavelet transform process is filtered by the filter processing unit 21 using the shielding frequency calculated for each local region in step S10 (step S13). By this filtering process, high frequency components exceeding the shielding frequency calculated individually for each local region are removed.

  In practice, as shown in FIGS. 4 and 7, the filter processing unit 21 uses a wavelet image (three-component wavelet image (high-frequency vertical component, high-frequency horizontal component, high-frequency diagonal component) obtained by wavelet transform processing). (ω images) are added, and each of a plurality of local regions in the added image is subjected to filter processing by the filter processing unit 21 using the shielding frequency calculated for each local region in step S10. The three-component wavelet image is multiplied by the ratio of the pixel values of the added image before and after the filtering process and rearranged at the original position.

  Returning to FIG. 3, the digital image subjected to the filter processing is subjected to wavelet inverse transform processing in the wavelet transform processing unit 19 (step S <b> 14), and returned to the original real space region. In step S15, the digital image is rotated in the opposite direction at the same angle as in step S11, and returned to the initial angle.

  The loop of steps S11 to S15 is repeated a predetermined number of times while the rotation angle is increased by 5 degrees, and a plurality of filtered digital images having different rotation angles by 5 degrees are generated. The digital image is subjected to an averaging process in the image processing unit 15 (step S16).

  FIG. 8A shows an example of the phantom image as the input data of FIG. 2, and the phantom image subjected to the conventional filter processing shown in FIG. 8B and the filter processing of the present embodiment shown in FIG. 8C. Shown in comparison with phantom images. FIG. 9A shows an example of a SPECT (nuclear medicine) clinical image subjected to the conventional filter processing, and FIG. 9B shows an example of the clinical image subjected to the filter processing of the present embodiment. 10A and 10B show a phantom image by CT, FIG. 10D shows an example of a CT phantom image subjected to the filter processing of the present embodiment, and FIG. 10C shows a CT phantom image subjected to conventional filter processing. An example is shown.

  As can be seen by comparing these image examples, according to the present embodiment, it is possible to realize a relatively uniform noise reduction process for an image with a locally varying S / N ratio, and for an image with a relatively coarse pixel size. Artifacts caused by the noise reduction processing can be removed.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

FIG. 1 is a diagram showing a configuration of an image processing apparatus according to an embodiment of the present invention. FIG. 2 is a flowchart showing a procedure of noise reduction processing according to the present embodiment. FIG. 3 is a conceptual diagram of wavelet transform. FIG. 4 is an explanatory diagram of a local cutoff frequency calculation process of FIG. FIG. 5A is a diagram showing a preceding stage of the cut-off frequency calculation processing of FIG. FIG. 5B is a diagram illustrating a latter stage of the cutoff frequency calculation process of FIG. 2. FIG. 6 is a diagram schematically illustrating the procedure of noise reduction processing according to the present embodiment. FIG. 7A is an explanatory diagram of the rotated image processing of FIG. FIG. 7B is an explanatory diagram of another rotated image process of FIG. FIG. 8A is a diagram showing an example of a phantom image as input data in FIG. FIG. 8B is a diagram illustrating an example of a phantom image that has undergone a conventional filter process. FIG. 8C is a diagram illustrating an example of a phantom image that has undergone the filter processing of the present embodiment. FIG. 9A is a diagram illustrating an example of a SPECT clinical image that has undergone conventional filtering. FIG. 9B is a diagram illustrating an example of a SPECT clinical image that has undergone the filter processing of the present embodiment. FIG. 10A is a diagram illustrating an example of a phantom image by CT as input data of FIG. FIG. 10B is a diagram illustrating an example of another phantom image by CT as input data in FIG. 2. FIG. 10C is a diagram illustrating an example of a phantom image by CT that has been subjected to conventional filter processing. FIG. 10D is a diagram illustrating an example of a phantom image by CT that has undergone the filter processing of the present embodiment. FIG. 11 is a diagram illustrating the reference angle of the rotation pitch according to the present embodiment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Interface, 11 ... Control part, 12 ... Data / control bus, 13 ... Image storage part, 15 ... Image processing part, 17 ... Cutoff frequency calculation part, 19 ... Wavelet transformation processing part, 21 ... Filter processing part

Claims (13)

A storage unit for storing digital image data;
A rotation processing unit for generating a plurality of rotated digital images having different rotation angles from the digital image;
An image processor for generating a plurality of image processed digital images from the plurality of rotated digital images;
A rewind processing unit for generating a plurality of rewinded digital images from the plurality of image-processed digital images;
An image processing apparatus comprising: a combining processing unit configured to combine the plurality of rewound digital images into one digital image. The image processing unit
A wavelet transform processor for subjecting the plurality of rotated digital images to a wavelet transform;
A filter processing unit that filters the digital image subjected to the wavelet transform with a cut-off frequency that is different for each region;
The image processing apparatus according to claim 1, further comprising: an inverse wavelet transform processing unit that subjects the filtered digital image to an inverse wavelet transform. The image processing unit further includes a cut-off frequency calculating unit that calculates a cut-off frequency for each local area with respect to the rotated digital image,
The image processing apparatus according to claim 2, wherein the filter processing unit filters the digital image subjected to the wavelet transform according to the calculated cutoff frequency for each local area. 4. The image processing apparatus according to claim 3, wherein the cutoff frequency calculation unit calculates a cutoff frequency based on vertical component data, horizontal component data, and diagonal component data of the digital image subjected to the wavelet transform. The cut-off frequency calculation unit calculates the cut-off frequency based on a result of addition-average weighting addition of vertical component data, horizontal component data, and diagonal component data of the digital image subjected to the wavelet transform. The image processing apparatus according to claim 3. The image processing apparatus according to claim 1, wherein the image processing unit filters the plurality of rotated digital images. The image processing apparatus according to claim 1, wherein the synthesis processing unit performs addition averaging or weighted addition on the plurality of rewinded digital images. The image processing apparatus according to claim 1, wherein the plurality of rotated digital images have different rotation angles by a predetermined angle. The image processing apparatus according to claim 8, wherein the predetermined angle is selectively set from a range of more than 0 ° and not more than 45 °. The image processing apparatus according to claim 8, wherein the predetermined angle is 3 to 10 °. The predetermined angle is set to n × θ.
n is a positive integer and is ½ or less of the number M of pixels parallel to the X axis.
tanθ = L2 / L1
L1: Distance from the image center to the edge of the endmost pixel on the Y-axis
9. The image processing apparatus according to claim 8, wherein L2 is a side length of a single pixel. 12. The image processing apparatus according to claim 11, wherein n is 3 or 5. A storage unit for storing digital image data;
A rotation processing unit that performs rotation processing on the digital image;
A wavelet transform processing unit for subjecting the digital image subjected to the rotation processing to a wavelet transform;
A filter processing unit for filtering the digital image subjected to the wavelet transform;
An inverse wavelet transform processor for subjecting the filtered digital image to an inverse wavelet transform;
An image processing apparatus comprising: an addition average processing unit that adds and averages a plurality of digital images subjected to the inverse wavelet transform with different rotation angles in the rotation processing.

Images