Ultrasonic diagnostic device and image processing method

Abstract

The present invention relates to an ultrasonic diagnostic apparatus and an image processing method. A target image (102) generated by multi-resolution processing is provided with a focus pixel (104) and a reference pixel group (106). In a corresponding image (100) belonging to an upper layer, pixel value patterns are compared between a corresponding focus region (116) and a plurality of corresponding reference regions (118), and a plurality of weights are calculated. A corrected focus pixel value is obtained by multiplying the plurality of reference pixel values ​​by the plurality of weights.

Description

Technical Field

[0001] The present disclosure relates to an ultrasonic diagnostic apparatus and an image processing method, and more particularly to a technique for improving the image quality of ultrasonic images. Background Art

[0002] An ultrasonic diagnostic apparatus is a device that generates and displays ultrasonic images based on data obtained by transmitting and receiving ultrasonic waves to a living body. Known ultrasonic images include tomographic images and blood flow images.

[0003] Ultrasonic images contain speckle noise, which is inherent noise. To improve the image quality of ultrasonic images, it is necessary to reduce speckle noise. Applying a typical smoothing filter to ultrasonic images to reduce speckle noise blurs the edges of the ultrasonic images.

[0004] Multi-resolution processing is a known technique for reducing speckle noise while preserving edges (see, for example, Japanese Patent Application Publication Nos. 2009-153918 and 2014-64736). In this process, the resolution of an input image is gradually reduced to generate multiple hierarchical input images corresponding to multiple hierarchies (levels). A filter is then applied to each hierarchical input image. The resolution of the filtered images is then increased to generate an output image. Conventional multi-resolution processing does not utilize pattern comparison. Summary of the Invention

[0005] The present disclosure aims to reduce speckle noise while preserving edges representing tissue morphology or structure in ultrasonic image processing. Alternatively, the present disclosure aims to implement multi-resolution processing using pattern comparison.

[0006] The ultrasonic diagnostic apparatus according to the present disclosure is characterized by comprising: a transform unit for generating a plurality of hierarchical input images provided to a plurality of hierarchical layers by gradually reducing the resolution of an input image; a plurality of filters for functioning in the plurality of hierarchical layers; and an inverse transform unit for increasing the resolution of the plurality of hierarchical output images generated in the plurality of hierarchical layers to generate an output image corresponding to the input image, wherein at least one of the plurality of filters comprises: a calculator for setting, in a corresponding image corresponding to a target image input to the filter and belonging to a layer immediately above the layer to which the filter belongs, a corresponding region of interest corresponding to a target pixel in the target image and a plurality of corresponding reference regions corresponding to a plurality of reference pixels in the target image, and calculating a plurality of weights by comparing a pixel value pattern in the corresponding region of interest with a pixel value pattern in each of the corresponding reference regions; and a corrector for correcting a pixel of interest value of the target pixel by applying the plurality of weights to a plurality of reference pixel values ​​of the plurality of reference pixels.

[0007] The image processing method disclosed herein includes the following steps: generating a plurality of hierarchical input images provided to a plurality of hierarchical layers by gradually reducing the resolution of an input image generated by transmitting and receiving ultrasonic waves; and increasing the resolution of a plurality of hierarchical output images generated in the plurality of hierarchical layers, performing filter processing in at least one of the plurality of hierarchical layers, wherein in the filter processing, a corresponding focus region corresponding to a focus pixel in the target image input to the filter processing and belonging to a layer immediately above the layer to which the filter processing belongs is set, a plurality of corresponding reference regions corresponding to a plurality of reference pixels in the target image are set, a plurality of weights are calculated by comparing a pixel value pattern in the corresponding focus region with a pixel value pattern in each corresponding reference region, and a plurality of reference pixel values ​​of the plurality of reference pixels are weightedly added based on the plurality of weights, thereby correcting the focus pixel value of the target pixel. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is a block diagram showing the configuration of an ultrasonic diagnostic apparatus according to an embodiment.

[0009] Figure 2 It is a diagram showing a first configuration example of the image processing unit.

[0010] Figure 3 It is a diagram showing weighted addition processing.

[0011] Figure 4 It is a diagram showing a modified example of the weighted addition process.

[0012] Figure 5It is a diagram showing a second structural example of the image processing unit.

[0013] Figure 6 It is a diagram showing a third structural example of the image processing unit. DETAILED DESCRIPTION

[0014] The following describes the embodiments based on the drawings.

[0015] (1) Overview of Implementation

[0016] An ultrasonic diagnostic apparatus according to an embodiment includes a transform unit, multiple filters, and an inverse transform unit. The transform unit generates multiple hierarchical input images for multiple layers by gradually reducing the resolution of an input image. The multiple filters function in the multiple layers. The inverse transform unit increases the resolution of the multiple hierarchical output images generated in the multiple layers to generate output images corresponding to the input image. At least one of the multiple filters includes a calculator and a corrector. The calculator sets a corresponding region of interest corresponding to a pixel of interest in the target image and multiple corresponding reference regions corresponding to multiple reference pixels in the target image, in a corresponding image corresponding to the target image input to the filter and belonging to the layer immediately above the layer to which the filter belongs. The calculator calculates multiple weights by comparing the pixel value pattern within the corresponding region of interest with the pixel value patterns within each corresponding reference region. The corrector corrects the pixel value of the target pixel by applying the multiple weights to the multiple reference pixel values ​​of the multiple reference pixels. The transform unit corresponds to the transform unit. The inverse transform unit corresponds to the inverse transform unit.

[0017] Typically, structures (biological tissues) within ultrasound images have inherent pixel value patterns (pixel value distributions), and the pixel value patterns of two locations within the same structure are similar. In contrast, randomness is observed in the pixel value patterns of speckle noise. The above configuration, based on these differences in properties, corrects the pixel value of interest by comparing pixel value patterns. This configuration effectively reduces speckle noise while preserving edges. Furthermore, in the above configuration, when calculating the multiple weights used by the filter, reference is made not to the target image input to the filter but to the corresponding image belonging to the layer immediately above the filter's layer. Because the resolution of the corresponding image is higher than that of the target image, more precise pattern comparison can be performed, thereby calculating a more appropriate set of weights. When determining a region for pattern comparison on the target image, the minimum size of the region is limited by the resolution of the target image. When determining a region for pattern comparison on the corresponding image, a smaller region can be determined.

[0018] As the filter, a non-local means filter can be used. A reference pixel group is set within a local area containing a pixel of interest on the target image. Furthermore, by setting corresponding reference areas near the corresponding pixel of interest on the corresponding image, information far from the pixel of interest can be prevented from being reflected in the pixel of interest.

[0019] The conversion unit may include multiple downsamplers or multiple other converters. The inverse conversion unit may include multiple upsamplers or multiple other converters. In an embodiment, the reference pixel group includes a pixel of interest. That is, the reference pixel group consists of the pixel of interest and multiple nearby pixels.

[0020] In an embodiment, a region in a target image containing multiple reference pixels has a first size. The corresponding region of interest and the multiple corresponding reference regions each have a second size. The second size corresponds to a third size in the filter hierarchy. The third size is smaller than the first size. This configuration reduces the size of each region used in the comparison of pixel value patterns, assuming that the comparison is performed on a high-resolution corresponding image. Multiple weights generated by comparing the corresponding region of interest and the multiple corresponding reference regions can be interpolated to calculate multiple weights for applying to multiple reference pixel values.

[0021] In the embodiment, the plurality of reference pixels is p×p pixels. The corresponding region of interest and the plurality of corresponding reference regions are each composed of q×q pixels. p and q are equal integers greater than or equal to 3. This structure simplifies calculations. In the embodiment, weights are calculated for each orientation from the perspective of the pixel of interest, and each weight is applied to the values ​​of nearby pixels in each orientation.

[0022] In the embodiment, the plurality of layers include an nth layer (where n is an integer greater than or equal to 2) as the lowest layer. The plurality of filters include an nth layer filter that functions in the nth layer. An nth layer input image is input to the nth layer filter as a target image.

[0023] In an embodiment, the plurality of layers further include an n-1th layer. The plurality of filters further include an n-1th layer filter functioning in the n-1th layer. The n-1th layer filter receives as input an n-1th layer intermediate image generated in the n-1th layer as a target image.

[0024] In the embodiment, the n-1th layer intermediate image is an image generated by adding the n-1th layer high-frequency component image included in the n-1th layer input image and the n-1th layer low-frequency component image generated based on the n-1th layer output image.

[0025] In this embodiment, the n-2nd-level intermediate image generated on the n-2nd level is input to the n-1th-level filter as the corresponding image. The n-2nd-level intermediate image is generated by adding the n-2nd-level high-frequency component image contained in the n-2nd-level input image and the n-2nd-level low-frequency component image generated from the n-1th-level intermediate image.

[0026] In the embodiment, the n-2th layer low-frequency component image generated from the n-1th layer intermediate image is input to the n-1th layer filter as the corresponding image. This configuration allows for easy generation of the corresponding image.

[0027] The image processing method according to the embodiment includes a transformation step and an inverse transformation step. In the transformation step, the resolution of an input image generated by transmitting and receiving ultrasound waves is gradually reduced to generate multiple hierarchical input images for multiple hierarchical layers. In the inverse transformation step, the resolution of the multiple hierarchical output images generated in the multiple hierarchical layers is increased. Filtering is performed on at least one of the multiple hierarchical layers. In the filtering, a corresponding region of interest corresponding to a pixel of interest in the target image and multiple corresponding reference regions corresponding to multiple reference pixels in the target image are set in a corresponding image belonging to a layer immediately above the layer to which the filtering is applied, corresponding to the target image input to the filtering. Multiple weights are calculated by comparing a pixel value pattern within the corresponding region of interest with a pixel value pattern within each corresponding reference region. Based on the multiple weights, multiple reference pixel values ​​of the multiple reference pixels are weighted and added together to correct the pixel value of the target pixel of interest.

[0028] The image processing method described above can be implemented through software functionality. In this case, the program executing the method is installed onto an information processing device via a network or removable storage medium. The term "information processing device" encompasses ultrasound diagnostic equipment, ultrasound diagnostic systems, and computers. The program is stored on a non-transitory storage medium within the information processing device.

[0029] (2) Details of implementation methods

[0030] Figure 1 The configuration of an ultrasonic diagnostic apparatus according to an embodiment is shown in . The ultrasonic diagnostic apparatus is installed in a medical institution or the like and is a medical device that generates and displays ultrasonic images based on data obtained by transmitting and receiving ultrasonic waves to a subject (living body).

[0031] exist Figure 1In the present invention, the probe 10 is a movable wave transceiver that contacts the surface of the object. A vibrating element array composed of a plurality of vibrating elements (transducers) is provided in the probe 10. An ultrasonic beam 12 is formed by the vibrating element array, and the ultrasonic beam 12 is electronically scanned. As a result, a beam scanning surface 14 is formed inside the object. The electronic scanning of the ultrasonic beam 12 is repeated, thereby repeatedly forming the beam scanning surface 14. The beam scanning surface 14 is a two-dimensional data acquisition area. As electronic scanning methods, electronic linear scanning methods, electronic sector scanning methods, etc. are known. A two-dimensional vibrating element array can also be provided in the probe 10 to obtain volume data from the three-dimensional space in the biological body.

[0032] The transmitter 16 is a transmission circuit that functions as a transmit beamformer. The receiver 18 is a receive circuit that functions as a receive beamformer. During transmission, the transmitter 16 outputs multiple transmit signals in parallel to the transducer array, thereby forming a transmit beam. During reception, when the transducer array receives reflected waves from within the body, multiple receive signals are output in parallel from the transducer array to the receiver 18.

[0033] The receiving unit 18 generates beam data by applying phase alignment and summing (delay and summing) to multiple received signals. The receiving unit 18 outputs a received frame data string as the ultrasonic beam 12 repeats electronic scanning. Each electronic scan of the ultrasonic beam 12 outputs one received frame data string. One received frame data string consists of multiple beam data strings arranged in the electronic scanning direction. One received beam data string consists of multiple echo data strings arranged in the depth direction. A beam data processing unit, not shown in the figure, is provided at the downstream stage of the receiving unit 18 to process each beam data string. The beam data processing unit includes an envelope detector, a logarithmic converter, and other components.

[0034] The image forming unit 20 generates a display frame data string based on the received frame data string. The image forming unit 20 includes a DSC (digital scan converter). The DSC functions as a coordinate conversion unit. Specifically, the DSC converts data conforming to the wave transmission and reception coordinate system into data conforming to the display coordinate system. The display frame data constituting the display frame data string may be, for example, B-mode tomographic image data. Other ultrasonic image data may also be generated by the image forming unit 20.

[0035] The image processing unit 24 applies multi-resolution processing for speckle noise reduction to each display frame data constituting the display frame data sequence. Figure 2The processed display frame data sequence is output from the image processing unit 24 and input to the display processing unit 26. Multi-resolution processing for speckle noise reduction may be applied to the received frame data sequence.

[0036] The display processing unit 26 has functions such as image synthesis and color calculation. The display processing unit 26 generates images for display on the display 28. In the embodiment, these images include tomographic images, which are dynamic images. According to the embodiment, speckle noise is reduced in the image processing unit 24, allowing high-quality tomographic images to be displayed on the display 28. The display 28 is composed of an LCD, an organic EL device, or the like.

[0037] The control unit 30 is composed of a CPU that executes a program. Figure 1 The operation of each structure shown in the figure. The operation panel 32 is connected to the control unit 30. The operation panel 32 is an input device including a plurality of buttons, a plurality of knobs, a trackball, a keyboard, and the like.

[0038] The image forming unit 20, the image processing unit 24, and the display processing unit 26 are each composed of a processor that executes a program. The functions performed by these units can be realized by the CPU.

[0039] Figure 2 1 shows a first configuration example of the image processing unit 24. In the first configuration example, the first to n-th layers are shown. Here, n is usually an integer greater than or equal to 2. In the example shown in the figure, n=2. The n-th layer is the lowest layer. Figure 2 Each structure shown can be realized by the function of software.

[0040] Reference numeral 34 denotes the 0th layer (level 0) to which the input image 40 belongs. Reference numeral 36 denotes the 1st layer (level 1) which is one level lower than the 0th layer 34. Reference numeral 38 denotes the 2nd layer (level 2) which is one level lower than the 1st layer 36. By gradually reducing the resolution of the input image 40, a plurality of low-resolution images provided to the plurality of layers 36 and 38 are generated.

[0041] Each tomographic image (each received frame data) constitutes an input image 40. Image processing for speckle noise reduction is applied to each tomographic image. Ultrasonic images other than tomographic images may also be processed in the image processing unit 24. The conversion unit 200, comprised of multiple downsamplers (DSs) 42 and 46, gradually reduces the resolution of the input image 40.

[0042] The inverse transform unit 202 increases the resolution of the output images 68 and 86 generated in the plurality of hierarchical layers 36 and 38 and is composed of a plurality of upsamplers (US) 70 and 88. In addition to these US 70 and 88, a plurality of US 48, 54, and 78 are also provided.

[0043] DS42 downsamples the input image to generate a first-level input image 44. During downsampling, every other pixel (pixel value) aligned in the x-direction (horizontally) is thinned out, and every other pixel (pixel value) aligned in the y-direction (vertically) is thinned out. The x-direction and y-direction thinning-out rates are both 1 / 2. First-level input image 44 is equivalent to an image reduced to 1 / 4 of input image 40.

[0044] DS46 performs the same function as DS42, and in DS46, the first-level input image 44 is downsampled to generate the second-level input image 45. The second-level input image 45 corresponds to an image reduced to 1 / 16 of the input image 40.

[0045] DS42 and DS46 actually perform Gaussian processing (convolution of Gaussian weight coefficient matrix) on the image to be processed before downsampling. On the other hand, US48, 54, 70, 78, and 88 perform Gaussian processing on the image to be processed after upsampling.

[0046] The filter 60 applies filtering to the second-level input image 45. Specifically, each pixel constituting the second-level input image 45 is designated as a pixel of interest, and a reference pixel group is determined centered around the pixel of interest. In the embodiment, the reference pixel group also includes the pixel of interest. A corrected pixel of interest value for the pixel of interest is determined by weighted addition of the pixel values ​​of the multiple reference pixels constituting the reference pixel group. The filter 60 includes a calculator 62 that calculates a weight set and a corrector 64 that performs weighted addition using the weight set.

[0047] In the embodiment, the calculator 62 calculates the weight set (see reference numeral 66 ) based on the first-layer input image 44 belonging to the first layer 36 , which is the layer above the second layer, rather than the second-layer input image 45 belonging to the second layer 38 .

[0048] From the perspective of filter 60, the second-level input image 45 is the target image, and the first-level input image 44 is the corresponding image corresponding to the target image. Each pixel constituting the target image is designated as a pixel of interest, and a reference pixel group is set for the target image, centered around the pixel of interest. Furthermore, a corresponding region of interest is set in the corresponding image, centered around the corresponding pixel of interest. Furthermore, multiple corresponding reference regions are set, centered around multiple corresponding reference pixels corresponding to multiple reference pixels. A pixel value pattern comparison is performed between the corresponding region of interest and the multiple corresponding reference regions, and multiple weights are determined based on the multiple similarities obtained. Filter 60 outputs the second-level output image 68 as the processed image.

[0049] The US54 generates a first-level low-frequency component image by upsampling the second-level input image 45. The subtractor 56 generates a first-level high-frequency component image 58 by subtracting the first-level low-frequency component image from the first-level input image 44. Meanwhile, the US70 generates a processed first-level low-frequency component image 72 by upsampling the second-level output image 68. The adder 74 generates a first-level intermediate image 76 by adding the first-level high-frequency component image 58 and the processed first-level low-frequency component image 72.

[0050] Filter 80 belonging to first layer 36 has the same structure as filter 60 described above and performs the same function as filter 60 described above. Filter 80 includes a calculator 82 and a corrector 84. The target image processed by filter 80 is first-layer intermediate image 76. Calculator 82 generates a weight set, referring to a corresponding image corresponding to the target image.

[0051] The corresponding image is a high-resolution image (see reference numeral 85 ) belonging to layer 0. The corrector 84 calculates a corrected pixel value of interest for the pixel of interest by performing weighted addition using the weight set calculated by the calculator 82 .

[0052] US48 generates a 0th-level low-frequency component image by upsampling the 1st-level input image 44. A subtractor 50 generates a 0th-level high-frequency component image 52 by subtracting the 0th-level low-frequency component image from the input image (0th-level input image) 40. Meanwhile, US78 generates a processed 0th-level low-frequency component image by upsampling the 1st-level intermediate image 76. An adder 79 generates a 0th-level intermediate image by adding the processed 0th-level low-frequency component image to the 0th-level high-frequency component image 52. This 0th-level intermediate image is used as the corresponding image (see reference numeral 85).

[0053] As already explained, a corresponding region of interest and multiple corresponding reference regions are set for the corresponding image, and a weight set is calculated by comparing pixel value patterns between them. In the first configuration example, the corresponding image provided to filter 80 is an image generated in the previous layer and is the same as the target image provided to filter 60.

[0054] Filter 80 outputs a first-level output image 86 as a processed image. US 88 applies upsampling to first-level output image 86. Adder 92 adds input image 40 and upsampled image 90 to generate output image 94. Adder 92 performs weighted addition. The two weights assigned to the two images can be fixed or adaptively set.

[0055] Figure 3 The processing performed by the filter 60 is schematically shown. In the second layer 38, each pixel of the target image 102 is set as a focus pixel 104. A reference pixel group 106 consisting of p×p reference pixels is set with the focus pixel 104 as the center. In the example shown in the figure, p=3. p can be set to an integer greater than 4 (usually an odd number). The reference pixel group 106 consists of the focus pixel 104 and 8 nearby pixels. By adding multiple multiplication values ​​obtained by multiplying multiple reference pixels by multiple weights, a corrected pixel value for the focus pixel 104 is obtained.

[0056] When calculating the multiple weights, corresponding image 100 belonging to first layer 36, which is the layer immediately above second layer 38, is used. Specifically, in corresponding image 100, corresponding pixel of interest 112 corresponding to the pixel of interest is used as a base point, and corresponding region of interest 116 for pattern comparison is determined with this base point as the center. Meanwhile, in corresponding image 100, multiple corresponding reference pixels 118A corresponding to multiple reference pixels are determined, and multiple corresponding reference regions 118 are defined with these base points as the centers. Multiple corresponding reference regions 118 are set within a predetermined region 114 centered on corresponding pixel of interest 112. Regions 114, 116, and 118 are also referred to as kernels, respectively.

[0057] In an embodiment, the corresponding region of interest 116 and the corresponding reference regions 118 are each composed of q×q pixels. In an embodiment, q=3, that is, q=p.

[0058] When the size of the region where the reference pixel group 106 exists is represented as a first size, and the sizes of the corresponding region of interest 116 and the corresponding reference regions 118 are represented as a second size, the second size corresponds to a third size in the second layer 38, and the third size is smaller than the first size. In the embodiment, by referencing an image belonging to the previous layer, a small kernel that cannot be set in the target image 102 can be set.

[0059] Pixel value patterns are compared between the corresponding focus region 116 and the 9 corresponding reference regions, thereby obtaining 9 similarities. The 9 similarities can be used as 9 weights without any change, or the 9 weights can be calculated based on the 9 similarities. Figure 3 Indicated by reference numeral 122 in the figure. For example, when reference pixel 106A and corresponding reference pixel 118A are in a corresponding relationship, the pixel value (usually the luminance value) of reference pixel 106A is multiplied by weight 108 calculated between corresponding focus area 116 centered on corresponding focus pixel 112 and corresponding reference area 118 centered on corresponding reference pixel 118A. The weighted pixel value is reflected in the corrected focus pixel value (see reference numeral 110) for focus pixel 104. This process is applied to all reference pixels, and the corrected focus pixel value is obtained as the weighted addition result. The above process is applied to each pixel constituting target image 102. In this way, a second-level output image is generated.

[0060] In ultrasound images, structures within the body typically have unique pixel value patterns. Meanwhile, speckle noise exhibits a less consistent pixel value pattern. The above processing effectively reduces speckle noise while preserving edges. Furthermore, since the image belonging to the previous layer is referenced when calculating the weight set, pixel value patterns can be compared more precisely.

[0061] In real space, the position of the pixel of interest 104 is substantially identical to, or approximately equal to, the position of the corresponding pixel of interest 112. While the positions of the reference pixels differ from the positions of the corresponding reference pixels, the orientation of the reference pixels from the perspective of the pixel of interest is identical to the orientation of the corresponding pixel of interest from the perspective of the corresponding pixel of interest. The method described in this embodiment can be considered equivalent to calculating weights for each orientation.

[0062] During the filter processing, the weight w(i, j) is calculated according to the following equations (1-1) and (1-2), for example.

[0063]

[0064]

[0065] Here, i is the coordinate of the pixel of interest, and j is the coordinate of the reference pixel. x(i) is the brightness value of the pixel of interest, and w(i, j) is the weight multiplied by x(j). Z(i) is the normalization coefficient. N represents the overall size of the reference area when calculating the weight (where N is an odd number), and M represents the size of the corresponding region of interest and each corresponding reference area (where M is an odd number). h is a smoothing parameter. Using the weights (i, j) calculated as above, the brightness value y(i) of the pixel of interest after filtering is calculated according to the following formula (2).

[0066]

[0067] The weight w(i, j) may be calculated using the following equations (3-1) and (3-2) instead of the above equations (1-1) and (1-2).

[0068]

[0069]

[0070] The calculation formula shown above is only an example, and other calculation formulas may be used to calculate the weights and corrected brightness values.

[0071] Figure 4 FIG. 6 shows a modified example of the processing performed by the filter 60. Figure 4 In, with Figure 3 The same elements as those shown are denoted by the same reference numerals, and their description is omitted.

[0072] As already described, the target pixel 104 and the reference pixel group 106 are set for the target image 102. The reference pixel group 106 is composed of p×p reference pixels, where p is 3 in this example.

[0073] On the other hand, a corresponding focus region consisting of q×q pixels is set for corresponding image 100, centered around corresponding focus pixel 112, where p < q < 2×p, for example, q = 5. Region 128, which is determined as the sum of multiple corresponding reference regions 132, consists of 9×9 pixels, but may also consist of 7×7 pixels.

[0074] Weights are calculated between the corresponding region of interest 130 and each corresponding reference region 132 (see reference numeral 134). Interpolation processing based on the multiple weights (weight set) calculated between the corresponding region of interest 130 and the multiple corresponding reference regions 132 is used to calculate an interpolated value 108A for each reference pixel as a weight (see reference numeral 136). This interpolated value 108A is multiplied by the reference pixel value. Furthermore, when calculating the weight set between the corresponding region of interest 130 and the multiple corresponding reference regions 132, calculation of weights not referenced in the interpolation process can be omitted. Filter processing conditions, including the value of p and the value of q, can be set by the user or automatically according to the characteristics of the input image and the processing purpose.

[0075] Figure 5 The second structural example of the image processing unit is shown. Figure 5 In, with Figure 2 The same structures as shown are marked with the same reference numerals and their descriptions are omitted. Figure 6 The same applies to the third structural example shown.

[0076] exist Figure 5 In the second structural example shown, the input image of the filter 80A is the first-level intermediate image 76, which is no different from the first structural example. On the other hand, in the second structural example, the upsampled image output from US78 is input as is to the filter 80A belonging to the first level 36 (refer to the reference numeral 85A). This image is an image belonging to the 0th level from the point of view of resolution, which is the same as the first structural example. According to the second structural example, the corresponding image referenced by the filter 80A can be easily generated. The structure of this part can be simplified. In addition, when adopting Figure 2 In the case of the first structural example shown, pixel value pattern comparison can be performed with better accuracy, that is, there is an advantage in that the possibility of calculating a more appropriate weight set can be increased.

[0077] Figure 6The third configuration example of the image processing unit is shown. The first-layer intermediate image 76 is input as the target image to the filter 80B belonging to the first layer, which is the same as the first and second configuration examples. Meanwhile, in the third configuration example, the corresponding image referenced by the filter 80B is also the first-layer intermediate image 76 (see reference numeral 85B). The third configuration example further simplifies the structure of the first layer 36. Similarly to the first and second configuration examples, the filter 60 belonging to the second layer 38 references the first-layer input image belonging to the previous layer as the corresponding image. This allows the filter processing in the second layer 38 to achieve the advantages described above. The third configuration example can be used when it is desired to simplify the structure while improving the accuracy of the filter processing for low-frequency components compared to high-frequency components.

[0078] Since speckle noise generally has diverse frequency components, multi-resolution processing is effective for this purpose. In this context, when performing weighted addition by comparing pixel value patterns during filter processing within each hierarchy, referencing the image of the previous hierarchy allows calculation of a more appropriate set of weights. Furthermore, by limiting the reference range to a local area, the reflection of information located far from the pixel of interest in the pixel of interest can be avoided or minimized.

[0079] Resolution conversion may also be performed using techniques other than downsampling and upsampling. For example, wavelet transform may be used. Multiple filter processes may also be applied to the target image at each level.

Claims

1. An ultrasonic diagnostic device, characterized in that Include: A conversion unit (200) generates a plurality of hierarchical input images (44, 45) provided to a plurality of hierarchical layers (36, 38) by reducing the resolution of the input image in stages. a plurality of filters (60, 80) functioning in the plurality of layers (36, 38); An inverse transformation unit (202) increases the resolution of the plurality of hierarchical output images (68, 86) generated in the plurality of hierarchical layers (36, 38) in order to generate an output image (94) corresponding to the input image (40). At least one of the plurality of filters (60, 80) comprises: A calculator (62, 82) sets a corresponding focus region (116) corresponding to a focus pixel (104) in the object image (102) and a plurality of corresponding reference regions (118) corresponding to a plurality of reference pixels (106) in the object image (102) on a corresponding image (100) corresponding to the object image (102) input to the filter and belonging to a layer one layer above the layer to which the filter belongs, and calculates a plurality of weights by comparing a pixel value pattern in the corresponding focus region (116) with a pixel value pattern in each of the corresponding reference regions (118); and The corrector (64, 84) corrects the pixel value of interest of the pixel of interest (104) by causing the plurality of weights to act on the plurality of reference pixel values ​​of the plurality of reference pixels (106).

2. The ultrasonic diagnostic apparatus according to claim 1, wherein On the object image (102), the region where the plurality of reference pixels (106) exist has a first size, The corresponding region of interest (116) and the corresponding reference regions (118) each have a second size, The second size corresponds to the third size in the hierarchy to which the filter belongs. The third size is smaller than the first size.

3. The ultrasonic diagnostic apparatus according to claim 1, wherein The plurality of reference pixels (106) is p×p pixels, The corresponding focus region (116) and the corresponding reference regions (118) are each composed of q×q pixels, The p and the q are the same integer greater than or equal to 3.

4. The ultrasonic diagnostic apparatus according to claim 1, wherein The plurality of layers (36, 38) include an nth layer (38) as the lowest layer, wherein n is an integer greater than or equal to 2, The plurality of filters (60, 80) include an n-th layer filter (60) functioning in the n-th layer (38), An n-th layer input image (45) is input to the n-th layer filter (60) as the target image (102).

5. The ultrasonic diagnostic apparatus according to claim 4, wherein The plurality of layers (36, 38) further include an n-1th layer (36), The plurality of filters (60, 80) further include an n-1th layer filter (80) functioning in the n-1th layer (36), The n-1th layer filter (80) is input with the n-1th layer intermediate image (76) generated in the n-1th layer (26) as the target image (102).

6. The ultrasonic diagnostic apparatus according to claim 5, wherein The n-1th layer intermediate image (76) is an image generated by adding the n-1th layer high-frequency component (58) contained in the n-1th layer input image (44) and the n-1th layer low-frequency component (72) generated based on the n-1th layer output image (68).

7. The ultrasonic diagnostic apparatus according to claim 5, wherein The n-2nd layer intermediate image (85) generated in the n-2nd layer is input to the n-1th layer filter (80) as the corresponding image (100). The n-2th layer intermediate image (85) is an image generated by adding the n-2th layer high-frequency component (52) contained in the n-2th layer input image and the n-2th layer low-frequency component generated based on the n-1th layer intermediate image (76).

8. The ultrasonic diagnostic apparatus according to claim 5, wherein The n-2nd layer low-frequency component image (85A) generated based on the n-1th layer intermediate image (76) is input to the n-1th layer filter (80A) as the corresponding image (100).

9. An image processing method, characterized in that: Include: A step (200) of generating a plurality of hierarchical input images (44, 45) provided to a plurality of hierarchical layers (36, 38) by gradually reducing the resolution of an input image (40) generated by transmitting and receiving ultrasonic waves; and Step (202) of increasing the resolution of the plurality of hierarchical output images (68, 86) generated in the plurality of hierarchical layers (36, 38), performing a filter process (60, 80) in at least one of the plurality of layers (36, 38), In the filter processing (60, 80), On a corresponding image (100) corresponding to a target image (102) input to the filter processing and belonging to a layer one layer above the layer to which the filter processing belongs, a corresponding focus region (116) corresponding to a focus pixel (104) in the target image (102) and a plurality of corresponding reference regions (118) corresponding to a plurality of reference pixels (106) in the target image (102) are set; Calculating a plurality of weights by comparing the pixel value pattern in the corresponding region of interest (116) with the pixel value pattern in each corresponding reference region (118), The plurality of reference pixel values ​​of the plurality of reference pixels (106) are weightedly added based on the plurality of weights, thereby correcting the pixel value of interest of the pixel of interest (104).

Images