JP5595988B2 - Ultrasonic diagnostic apparatus and image quality improving method of ultrasonic diagnostic apparatus - Google Patents

Description

  The present invention relates to an ultrasound diagnostic apparatus that acquires an image by transmitting and receiving ultrasound to and from a subject, and more particularly, an ultrasound diagnostic apparatus and an ultrasound having a function of performing image quality improvement processing by image processing on an acquired image The present invention relates to a method for improving image quality of a diagnostic apparatus.

Ultrasound diagnostic apparatuses are used in examinations of many parts in the body including the abdomen and heart. Unlike X-ray examinations, ultrasonic diagnostic equipment is widely used because it has advantages such as no harm to the living body, simple operation, and real-time observation of moving images. . In the ultrasonic diagnostic apparatus, an ultrasonic wave is emitted from the ultrasonic probe toward the subject, a reflected wave from the tissue inside the subject is received by the ultrasonic probe, and displayed on the monitor. By scanning an ultrasonic wave focused in a specific direction for a plurality of locations, a two-dimensional image can be acquired in real time. The types of ultrasound images include a B-mode image in which the reflectance of the biological tissue of the subject is converted to the brightness of the pixel value, a Doppler image having movement speed information of the biological tissue, a strain amount and an elastic modulus of the biological tissue A tissue elasticity image having hue information and a synthesized image obtained by synthesizing information of these images.

However, the two-dimensional image obtained by the ultrasonic diagnostic apparatus includes speckle noise generated by interference of a plurality of reflected waves from a fine structure inside the living body. In addition, since the reception signal obtained by the ultrasonic probe is a band-limited signal, a high-frequency component that should originally be obtained at the tissue boundary cannot be obtained sufficiently, and the edge included in the image may be blunted. The generation of speckle noise and the blunting of the edge cause a deterioration in image quality and have an adverse effect on diagnosis. In order to accurately read important structures such as lesions, display of ultrasonic images subjected to speckle noise reduction and edge enhancement processing is required.

  As a technique for reducing speckle noise, a frequency compound method and a spatial compound method can be cited. The frequency compound method is a method of generating a plurality of images by radiating a plurality of ultrasonic waves having different frequencies toward the same site, and acquiring one image by an averaging process of these images.

The speckle noise pattern varies greatly depending on the frequency of the ultrasonic wave to be used. On the other hand, the reflected wave from the tissue boundary or the like has a small variation due to the frequency. Therefore, the speckle noise can be reduced by the above-described averaging process. However, in the frequency compound method, since the frequency is divided and used, there is a problem that the frequency band of the image becomes narrow and the edge becomes dull. The spatial compound method is a method of generating a plurality of images by radiating ultrasonic waves from different directions toward the same part, and acquiring one image by adding and averaging the images. In this method, speckle noise is reduced by utilizing the fact that the pattern of speckle noise varies depending on the radiation direction of ultrasonic waves. However, since the space compound method usually requires a long time to acquire one image, there is a problem that the display speed of the image decreases.

  On the other hand, as a method different from the above, there is a noise reduction method by image processing. Due to high performance and low cost of image processing processors in recent years, complex image processing that has been difficult to put into practical use in terms of processing speed can be mounted relatively easily. It is known that noise reduction methods such as smoothing filters that have been known for a long time have problems such as edge blunting and loss of important signals. On the other hand, represented by Wavelet transform and Laplacian Pyramid transform A noise removal method and an edge enhancement method using such multi-resolution analysis have been attracting attention (for example, Patent Documents 1 to 3).

US Pat. No. 5,497,777 JP 2006-116307 A Recently, as a more advanced multiresolution analysis method, Curvelet transform (Non-patent document 1), Contourlet transform (Non-patent document 2), Complex Wavelet transform (Non-patent document 3), Steerable Pyramid transform (Non-patent document 3) A method such as Patent Document 4) has been proposed.

  Also, Non-Patent Document 5 proposes application of these advanced multi-resolution decomposition type ultrasonic diagnostic apparatuses.

In the conventional Wavelet transform, the number of divisions in the edge direction is 3, and in the Laplacian Pyramid transformation, the number of divisions in the edge direction is 1. On the other hand, the above advanced multi-resolution decomposition method has a division number in the edge direction of 4 or more. It is a method that can be done. Here, the number of divisions in the edge direction is K, which means that the image is decomposed into K decomposition coefficients that react strongly to patterns having brightness changes in different K types of directions at each resolution level and each position. . In the conventional Wavelet transform, an image is represented by three edge direction decomposition coefficients: vertical (0 degrees), horizontal (90 degrees), and diagonal (45 degrees and 135 degrees). An edge in the degree direction cannot be identified. In order to perform high-performance image quality improvement processing, it is important to use a multi-resolution decomposition method having at least four divisions in the edge direction.

  In the image quality improvement method based on multi-resolution decomposition, the intensity conversion of the decomposition coefficient is usually performed based on the estimated value of the amount of noise included in each decomposition coefficient. That is, the strength of the decomposition coefficient estimated to contain a lot of signal components is preserved or enhanced, and conversely, the strength of the decomposition coefficient estimated to contain a lot of noise components is reduced, and then the image is reconstructed from the decomposition coefficients. By performing the processing, an image subjected to noise reduction and edge enhancement can be obtained. Therefore, estimation of the noise amount and intensity conversion of the decomposition coefficient are important processes.

J.L.Starck, E.J.Candes, et al .: IEEE Trans. Image Processing, 11, 6, pp.670-684 (2002) M.N.Do, M.Vetterli: IEEE Trans. Image Processing, 14, 12, pp.2091-2106 (2005) N.G.Kingsbury: Proceedings of European Signal Processing Conference, pp.319-322 (1998) E.P.Simoncelli, W.T.Freeman: Proceedings of IEEE International Conference on Image Processing, 3, pp.444-447 (1995) E.H.O.Ng: Applied science in electrical ana computer engineering, University of Waterloo (Master thesis), pp.1-112 (2005)

  However, in the image quality improvement methods proposed in Non-Patent Documents 1 to 5, a sufficient improvement effect cannot be obtained for low S / N images typified by biological ultrasound images as listed below. Exists. The present invention provides a means for obtaining a sufficiently improved image quality improvement result even in such a case.

  (1) The amount of speckle noise included in the ultrasonic image is determined by the nature of imaging conditions (type of ultrasonic probe, magnification, frequency used, presence / absence of compound method, scanning pitch, etc.) and type of image (B Mode image, Doppler image, tissue elasticity image, composite image, etc.) and the imaging target, and changes depending on the resolution level, edge direction, and position. In addition, there are cases where it is difficult to distinguish a signal from noise at a resolution level, edge direction, and position that contain a relatively large amount of signal components compared to noise components. Therefore, there is a problem that it is difficult to estimate a stable noise amount with respect to the imaging conditions, the types of images, and the imaging targets. For example, there is a conventional method for performing noise estimation based on the assumption that the decomposition coefficient in which the noise component is dominant occupies the most, but the above assumption does not always hold.

(2) Generally, it is difficult to separate a noise component and a signal component. When the S / N is high to some extent, the resolution coefficient whose signal component is dominant often has a large amplitude. Therefore, even if only information relating to the amplitude of each resolution coefficient is used as in Patent Document 1, the performance is sufficiently high. Separation is possible. However, in the case of a low S / N image, it is difficult to discriminate between a signal and noise only with information on individual decomposition coefficients, and even in such a case, it is a problem to obtain sufficient separation performance.

(3) Processing time and image quality improvement performance in image quality improvement processing are usually in a trade-off relationship. On the other hand, the required processing time differs depending on the examination application. For example, when it is desired to observe a moving tissue such as a heart at a high frame rate, processing with a small amount of calculation is required. However, with simple processing with a small amount of calculation, sufficient performance cannot be obtained, for example, when it is desired to inspect in detail a site where the temporal variation is slow.
The problems shown in the above (1) to (3) lead to failure of intensity conversion of the decomposition coefficient and a sufficient image quality improvement result.

  The present invention solves the above problems by an ultrasonic image quality improvement method including the following processing and an ultrasonic diagnostic apparatus equipped with the image quality improvement method.

  (1) In order to estimate the noise amount with high accuracy, an estimated value of the noise amount is calculated for each resolution level. However, as described above, some of the noise amount estimation accuracy may be low, so that the reliability for noise amount estimation is given for each estimated noise amount, and the noise amount estimation is performed according to the reliability. By correcting the value, a highly accurate correction value of the noise amount estimation value is calculated. The reliability may be prepared in advance as a constant value inside the system, for example, or may be a distance between the estimated noise amount calculated by an estimation method different from the noise amount estimation method and the estimated noise amount. Can also be given.

(2) As the classification of the noise component and the signal component, not only information regarding individual decomposition coefficients for each position, resolution level, and edge direction but also distribution information of the decomposition coefficients obtained from a plurality of decomposition coefficients is used. The signal component may form a characteristic distribution in the decomposition coefficient distribution due to its tissue shape, regularity, and the like. For example, a high-amplitude decomposition coefficient is likely to be present in the vicinity of the peak of the decomposition coefficient distribution formed by signal components. (A vicinity of a certain decomposition coefficient is a decomposition coefficient close to the position, resolution level, edge direction, etc.) And the target decomposition factor). The information of the decomposition coefficient distribution is also regarded as a feature amount, and the discrimination performance can be improved by determining the noise component and the signal component based on the feature amount. Further, in the case of a moving image, information of continuous time-series image frames is used. Since signal components are often regular and moderate in temporal variation compared to noise components, it is possible to improve the classification performance by extracting regular features at corresponding positions in time-series image frames. .

  (3) It is characterized in that processing parameters are set so as to obtain an appropriate image display speed and image quality effect according to the imaging condition, the type of image, and the imaging target. For example, a processing parameter is set such that an image quality improvement result with the highest performance can be obtained without reducing the display speed with respect to the image display speed determined by the imaging condition, the type of image, and the imaging target. The processing parameters related to the trade-off between image quality and processing time include, for example, the number of resolution level divisions and the number of edge direction divisions in multi-resolution decomposition, and each resolution level, edge direction, and position subject to noise amount estimation. Whether or not to perform the determination, the number of decomposition coefficients used for separating the noise component and the signal component for each decomposition coefficient, whether or not to use information of the time-series image frame, and the like.

  As a specific advantage of changing each processing parameter, for example, as the number of resolution level divisions in multi-resolution decomposition is increased, the signal and noise can be separated in a lower frequency region, and more noise is generated at a relatively low frequency. Effective when ingredients are included. Further, as the number of divisions in the edge direction in the multiresolution decomposition is increased, the shape of the tissue can be stored with higher accuracy, and the tissue having a complicated shape can be inspected in detail. In addition, multiple algorithms are prepared in the multi-resolution decomposition method, noise amount estimation method, noise amount correction method, noise component and signal component discrimination method, and higher performance results are obtained by switching these algorithms. be able to.

According to the present invention, in the ultrasonic diagnostic apparatus and the image quality improvement method of the ultrasonic diagnostic apparatus, the multi-resolution decomposition method is adopted, noise amount estimation and correction using multiple resolution level decomposition coefficients, multiple decomposition coefficients, and By performing intensity conversion of the decomposition coefficient based on the corrected noise amount, it is possible to obtain an ultrasonic image with reduced noise components, which can improve the visibility of the tissue structure and lesion site compared to conventional devices. I can do it.
Furthermore, by switching the processing parameters according to the imaging conditions, the type of image, and the imaging target, it is possible to reduce both the processing time and improve the image quality according to the application.

It is a figure which shows the flow of the image quality improvement process in the Example of this invention. It is a block diagram showing the structure of the ultrasonic diagnosing device based on the Example of this invention. It is a block diagram showing a multi-resolution decomposition method. It is explanatory drawing of the frequency division | segmentation method in multi-resolution decomposition. It is a figure which shows the flow showing the noise amount estimation method. It is explanatory drawing of correction noise amount calculation processing. It is a figure which shows the flow of an intensity | strength conversion process. It is explanatory drawing of the amplitude conversion process of a decomposition coefficient. It is explanatory drawing of the correction process of the decomposition coefficient based on a preservation degree. It is explanatory drawing of the coefficient strength conversion process based on a some decomposition coefficient. It is a figure which shows the flow showing a preservation | save degree calculation process. It is explanatory drawing of intensity | strength conversion based on the weighting addition process of the some decomposition coefficient in the position of a vicinity. It is explanatory drawing of the coefficient strength conversion process of the decomposition coefficient based on the decomposition coefficient in the some flame | frame of the vicinity. It is a figure showing a flow showing preservation degree calculation processing using a decomposition coefficient in a plurality of neighborhood frames. It is a block diagram showing the reconstruction method. It is a figure which shows the flow of the image quality improvement process which can change a process parameter according to imaging conditions, the kind of image, and the imaging target. It is a figure which shows the table showing the process parameter corresponding to each imaging condition, the kind of image, and the imaging target.

  Embodiments according to the present invention will be described with reference to the drawings.

  The present invention is a process and apparatus for performing image processing using multi-resolution decomposition in order to improve the image quality of a captured image acquired by performing transmission and reception of ultrasonic waves.

  An embodiment of the present invention will be described with reference to FIGS.

FIG. 1 is a diagram showing an example of the flow of image quality improvement processing in the present invention. First, in the multiresolution decomposition process 101, a decomposition coefficient w is obtained by performing multiresolution decomposition on the processed input image x. The processing input image x is a vector having a scalar value x [m, n] for each position (m, n). The decomposition coefficient w is a vector having a scalar value w _{j, o} [m, n] for each position (m, n), each resolution level j, and each edge direction o.

The decomposition coefficient w generally takes a real or complex value for each resolution level, edge direction, and position. Next, an estimated noise amount z ′ is obtained by the noise amount estimating process 102, and the corrected noise amount z is calculated by correcting the estimated noise amount by the corrected noise amount calculating process 103. When it is not necessary to correct the estimated noise amount z ′, the corrected noise amount calculation process 103 may output the same value as the estimated noise amount z ′ as the corrected noise amount z. Thereafter, the intensity conversion processing 104 performs intensity conversion of the decomposition coefficient w based on the correction noise amount z. The decomposition coefficient after the intensity conversion is called an image quality improvement decomposition coefficient w ′. Finally, an image after image quality improvement processing (hereinafter, image quality improved image) y is obtained by the reconstruction processing 105. The image quality improved image y is similar to the processing input image x at each position (m,
A vector having a scalar value y [m, n] for n). In addition, the estimated noise amount z ′, the corrected noise amount z, and the image quality improvement decomposition coefficient w ′ are scalars for each position (m, n), each resolution level j, and each edge direction o, similarly to the decomposition coefficient w. It is a vector with values z ′ _{j, o} [m, n], z _{j, o} [m, n], w ′ _{j, o} [m, n].

Next, the configuration of the ultrasonic diagnostic apparatus according to the present invention will be described with reference to FIG. FIG. 2A is a diagram illustrating an example of the configuration of the ultrasonic diagnostic apparatus 201. The ultrasonic diagnostic apparatus 201 includes an ultrasonic probe 203 that transmits and receives an ultrasonic signal, a drive circuit 202 that generates a drive signal to be input to the ultrasonic probe 203, and a reception circuit that performs amplification and A / D conversion of the received signal. 204
, An image generation unit 205 that generates an image in which scanning line signal sequences of ultrasonic scanning are arranged in two dimensions, an image quality improvement processing unit 206 that performs image quality improvement processing, and image coordinates represented by the scanning line signal sequence It includes a scan converter 212 that performs conversion processing and interpolation processing, a display unit 213 that displays an image generated by the scan converter, and a control / storage / processing unit 220 that controls the whole and stores and processes data.

The ultrasonic probe 203 transmits an ultrasonic signal based on the drive signal to the subject 200, receives a reflected wave from the subject 200 obtained at the time of transmission, and converts it into an electrical reception signal. The types of the ultrasonic probe 203 include, for example, types called linear, convex, sector, and radial. When the ultrasonic probe 203 is a convex type, the scan converter 212 converts the rectangular image into a fan-shaped image.

  The image generation unit may perform position correction so that an image obtained by performing transmission / reception in successive time frames and the display position of the tissue are the same.

  As shown in FIG. 2D, the image quality improvement processing unit 206 includes a multi-resolution decomposition unit 207, a noise amount estimation unit 208, a noise amount correction unit 209, an intensity conversion unit 210, and a reconstruction unit 211. 1, multi-resolution decomposition processing 101, noise amount estimation processing 102, correction noise amount calculation processing 103, intensity conversion processing 104, and reconstruction processing 105 in FIG. 1 are performed.

As shown in FIG. 2C, the control / storage / processing unit 220 includes an input unit 221, a control unit 222, a storage unit 223, and a processing unit 224. The input unit 221 generates an image. The start timing, parameters related to image generation, and the like are input. The control unit 222 controls operations of the drive circuit 202, the ultrasonic probe 203, the reception circuit 204, the image quality improvement processing unit 206, and the like. The storage unit 223 stores the received signal, the image generated by the image generation unit 205, the decomposition coefficient calculated by the image quality improvement processing unit 206, the image after the image quality improvement, the display image output from the scan converter 212, and the like. The The processing unit 224 performs processing for shaping an electric signal to be input to the ultrasonic probe 203, processing for adjusting brightness and contrast at the time of image display, and the like.

  In the configuration as described above, the ultrasonic probe 203 transmits an ultrasonic signal based on the drive signal controlled by the control unit 222 of the control / storage / processing unit 220 to the subject 200, and is obtained by this transmission. A reflection signal from the subject 200 is received and converted into an electrical reception signal.

Next, after the received signal converted into an electric signal is amplified by the receiving circuit 204 and A / D converted,
The A / D converted signal is processed by the image generation unit 205 to generate an image, which is input to the image quality improvement processing unit 206. In the image quality improvement processing unit 206, the input image is subjected to the multi-resolution decomposition processing 101, the noise amount estimation processing 102, the correction noise amount calculation processing 103, the intensity conversion processing 104, and the reconstruction processing 105 as described above. Thus, a highly accurate image quality improvement process is performed, and an image quality improvement processed image is obtained. Further, the image quality improvement processing image is subjected to image coordinate conversion processing and interpolation processing by the scan converter 212 to generate an image, thereby displaying a clearer ultrasonic image with reduced noise components on the screen of the display unit 213. I can do it.

FIG. 2B is a diagram showing another embodiment of the configuration of the ultrasonic diagnostic apparatus according to the present invention. In the configuration of the ultrasonic diagnostic apparatus 201 ′ illustrated in FIG. 2B, the arrangement of the scan converter 214 and the image quality improvement processing unit 215 is different from the configuration of the ultrasonic diagnostic apparatus 201 illustrated in FIG. on the other hand,
In the configuration shown in FIG. 2B, the same components as those shown in FIG.

The ultrasonic diagnostic apparatus 201 ′ shown in FIG. 2B includes an ultrasonic probe 203 that transmits and receives an ultrasonic signal, a drive circuit 202 that generates a drive signal to be input to the ultrasonic probe 203, and a received signal. A receiving circuit 204 that performs amplification and A / D conversion, an image generation unit 205 that generates an image in which scanning line signal sequences of ultrasonic scanning are arranged in a two-dimensional form, and coordinate conversion processing of an image generated by the image generation unit 205 A scan converter 214 that performs interpolation processing, an image quality improvement processing unit 215 that performs image quality improvement processing of an image generated by the scan converter 214, a display unit 213 ′ that displays an image on which image quality improvement processing has been performed, and the entirety thereof. A control / storage / processing unit 220 ′ for controlling and storing and processing data is provided.

  In the configuration of FIG. 2A, the output image of the image generation unit 205 is an image in which scanning line signal sequences of ultrasonic scanning are expressed in parallel on the image. Therefore, by performing image quality improvement processing on the image generated by the image generation unit 205, it is possible to improve image quality degradation that depends on the scan direction. On the other hand, in the configuration of FIG. 2B, the output image of the scan converter 214 is an image having the same format as the image output to the display unit 213. Therefore, by performing image quality improvement processing on the image generated by the scan converter 214, it is possible to minimize degradation of image quality that may occur during coordinate conversion processing or interpolation processing by the scan converter 214.

  Next, multi-resolution decomposition processing performed by the multi-resolution decomposition unit 207 of the image quality improvement processing unit 206 or 215 will be described with reference to FIGS.

As described above, methods such as Wavelet transform, complex Wavelet transform, Curvelet transform, Contourlet transform, and Steerable Pyramid transform are known as multi-resolution decomposition methods. FIG. 3 shows an embodiment of the present invention showing the processing flow of the multi-resolution decomposition processing. Figures 3 (a), (b), and (c) show the Wavelet transform, complex Wavelet transform, Steerable Pyrami with four or more edge direction divisions, respectively.
It is one Example of the processing flow by the multi-resolution decomposition system based on d conversion.

First, the processing flow of the multi-resolution decomposition method based on Wavelet transform in which the number of divisions in the edge direction shown in FIG. First, a horizontal (m direction) one-dimensional low-pass filter 301 and a horizontal one-dimensional high-pass filter 302 are applied to the processed input image x [m, n], and then the output signal of each filter is longitudinally applied. A one-dimensional low-pass filter 303 and a high-pass filter 304 in the direction (n direction) are applied. The coefficients of these filters are real numbers. The order in which the horizontal filters 301 and 302 and the vertical filters 303 and 304 are applied may be reversed. Further, immediately after these filter processes, a process of thinning out pixels every other pixel (hereinafter, thinning process) may be performed. As a result, the processing input image x is decomposed into four types of resolution coefficients s ₁ , w _{1, C} , w _{1, A} , and w _{1, diag} of resolution level 1.

Decomposition coefficients s ₁ is the horizontal and vertical direction represent the components are both low frequency, the s ₁ resolution level
Called 1 low frequency resolution factor. Also, the decomposition coefficient w _{1, C} is a component with a low frequency in the horizontal direction and a high frequency in the vertical direction, the decomposition coefficient w _{1, A} is a component with a high frequency in the horizontal direction and a low frequency in the vertical direction, and the decomposition coefficient w _{1, d
iag} represents a component having a high frequency in both the horizontal and vertical directions. The decomposition coefficient w _{1, C} reacts strongly to the high frequency edge along the horizontal direction, and the decomposition coefficient w _{1, A} reacts strongly to the high frequency edge along the vertical direction.
In addition, the decomposition coefficient w _{1, diag} reacts strongly to the high-frequency edge along the oblique 45 degree direction, but the oblique 45 degree.
The edges along the degree direction include two types of edges: an edge parallel to the straight line m = n and an edge parallel to the straight line m = -n.

Therefore, for w _{1 and diag} , a filter 305 that passes an edge parallel to the straight line m = n and blocks an edge parallel to the straight line m = -n, and conversely, an edge parallel to the straight line m = n is provided. By applying a filter 306 that cuts off and passes through an edge parallel to the straight line m = −n, resolution coefficients w _{1, D} and w _{1, B} at resolution level 1 are calculated. The decomposition coefficients w _{1, A} , w _{1, B} , w _{1, C} , w _{1, D} are referred to as resolution level 1 high frequency decomposition coefficients. FIG. 4 (a) shows dominant frequency components included in each decomposition coefficient in the multi-resolution decomposition method using Wavelet transform in which the number of divisions in the edge direction is 4 or more. The frequencies f _m and f _n are
Each represents a horizontal frequency and a vertical frequency. For example, w _{1, A} includes many components having a high horizontal frequency f _{m and a} low vertical frequency f _n .

Subsequently, in the same manner as the processing performed on the processing input image x, the decomposition coefficient s ₁ is subjected to filter processing, whereby the resolution coefficient s ₂ , w _{2, A} , w _{2, B} , resolution level 2 is obtained. Calculate w _{2, C} and w _{2, D.} In FIG. 3 (a), only the resolution coefficient of resolution level 2 is calculated, but by repeating the same process recursively, the resolution coefficient of higher resolution level is calculated. As shown in FIG. 4 (a), the higher the resolution level, the more the lower frequency components are included.

When the high-frequency decomposition coefficient at the highest resolution level calculated by the multi-resolution decomposition process is w _J−1 , J is called the highest resolution level (J = 3 in the example of FIG. 3A). Low frequency decomposition coefficient s _J-1 is w _J,
The low-frequency resolution coefficient s _J-1 = w _{J, A} is the resolution level J resolution coefficient, resolution level j (j = 1, ...
, J-1) is called the resolution coefficient of resolution level j. The resolution coefficient w _{j, o} in the resolution level j and the edge direction o is a vector having a scalar value w _{j, o} [m, n] for each position (m, n). In addition, the decomposition coefficient at all resolution levels is represented by w. In FIG. 3 (a), the number of divisions in the edge direction at resolution levels 1 and 2 is four. However, the number of divisions in the edge direction is not necessarily four. There may be. The same is true for FIGS. 3B and 3C described below.

Next, the processing flow of the multi-resolution decomposition method based on the complex wavelet transform which is an embodiment of the present invention shown in FIG. 3B will be described. First, in the horizontal direction (m direction) with respect to the processed input image x [m, n]
After applying the one-dimensional low-pass filters 321 and 322 and the horizontal one-dimensional high-pass filters 323 and 324, the vertical (n-direction) one-dimensional low-pass filters 325 and 326 and the high-pass are subsequently applied to the output signals of the filters. Filters 327 and 328 are applied. In this embodiment, two different types of filters are used in each filter process, but the present invention is not limited to this. The coefficients of these filters are generally complex numbers.

Similar to the example of FIG. 3A, the order of applying the horizontal filter and the vertical filter may be reversed, or a thinning process may be performed immediately after these filter processes. Decomposition coefficients s _{1, A} , s obtained by applying a horizontal one-dimensional low-pass filter and a vertical one-dimensional low-pass filter
_{1, B is referred} to as a resolution level 1 low frequency decomposition coefficient. Next, processing for calculating the sum and difference of the two input signals is performed by the ΣΔ block 329 on the calculated decomposition coefficients other than the low-frequency decomposition coefficient, and the high-frequency decomposition coefficients w _{1, A} , w _{1 of} resolution level 1 are applied. _{, B} , w _{1, C} , w _{1, D} , w _{1, E} , w _{1, F} are calculated.
Subsequently, as with the processing performed on the processing input image x, the low frequency decomposition coefficient s _{1 of} resolution level _{1
, A} and s _{1 and B} are subjected to a filtering process to obtain resolution level 2 resolution coefficients s _{2 and A} and s _{2 and B}
, W2 _{, A} , w2 _{, B} , w2 _{, C} , w2 _{, D} , w2 _{, E} , w2 _{, F} are calculated. Below, by recursively processing,
Calculate the resolution factor for the higher resolution level. FIG. 4B shows dominant frequency components included in each decomposition coefficient in the multi-resolution decomposition method based on the complex wavelet transform. Six types of high-frequency resolution coefficients at each resolution level react strongly to six different edge directions.

Next, the processing flow of the multi-resolution decomposition method based on Steerable Pyramid conversion, which is an embodiment of the present invention, shown in FIG. First, a high-frequency decomposition coefficient w ₁ of resolution level ₁ is calculated by applying a two-dimensional high-pass filter 341 that blocks only low frequency components in the vertical and horizontal directions to the processed input image x [m, n]. To do. On the other hand, by applying a two-dimensional low-pass filter 342 that passes only a component having a low frequency in both the vertical and horizontal directions to the processed input image x, a low frequency decomposition coefficient s ₁ of resolution level ₁ is calculated. A thinning process may be performed immediately after the two-dimensional low-pass filter process.

Then, in either direction of vertical and horizontal direction filter 343, 344, 345 and 346 passes only a component of the high frequency at and specific edge direction, by applying a ... to degradation coefficients s _1, The high frequency resolution coefficients w2 _{, A} , w2 _{, B} , w2 _{, C} , w2 _{, D} ,... At resolution level 2 are calculated. Further, a low-frequency decomposition coefficient s ₂ at resolution level ₂ is calculated by applying a two-dimensional low-pass filter 347 to the decomposition coefficient s ₁ . Subsequently, in the same manner as the processing performed on the decomposition coefficient s ₁ , the resolution coefficient w _{3, A} , w of resolution level 3 is applied by applying a filter to the decomposition coefficient s ₂ .
_{3, B} , w3 _{, C} ,... Are calculated, and the process is performed recursively below.

FIG. 4 (c) shows dominant frequency components included in each decomposition coefficient in the multi-resolution decomposition method based on Steerable Pyramid transformation. However, in the example of FIG. 4C, the number of divisions in the edge direction at each resolution level 2 and 3 is 8 and 4, respectively. The processing parameters in the embodiment shown in FIG. 3 include the maximum resolution level, the number of edge direction divisions for each resolution level, the edge direction to be divided, and the presence or absence of thinning after filtering. These processing parameters can be changed according to imaging conditions (type of ultrasonic probe, magnification, frequency used, presence / absence of compound method, scanning pitch), image type, and imaging target (hereinafter referred to as imaging conditions / images). The types and imaging targets are collectively referred to as image imaging information).

  Next, processing related to the calculation of the noise amount in the noise amount estimation unit 208 of the image quality improvement processing unit 206 or 215 will be described with reference to FIGS.

FIG. 5 is a diagram for explaining the noise amount estimation processing 102 in FIG. Since noise contained in an ultrasound image has different characteristics depending on the normal position, resolution level, and edge direction, in order to perform image quality improvement processing while appropriately suppressing noise, the estimated noise amount is set for each position, resolution level, and edge direction. It is effective to calculate. In the noise amount estimation processing of FIG. 5, the estimated noise amount z ′ _{j, o} [m, n] at the decomposition coefficient w _{j, o} [m, n] for each of a plurality of resolution levels j, as in processing 501, 502, and 503. n]. Where w _{j, o} [m, n] and z ' _{j, o} [
m, n] represents a decomposition coefficient and an estimated noise amount at the same resolution level j, edge direction o, and position (m, n).

For example, z ′ _{j, o} [m, n] is calculated as follows as standard deviations of decomposition coefficients in all edge directions and all positions at the same resolution level j.

Here, N _j is the number of decomposition coefficients in all edge directions and all positions at resolution level j.

As another example, the following calculation can be performed using the median of the absolute values of the decomposition coefficients.

Here, α is a constant. The estimated noise amount does not necessarily have to be calculated at all resolution levels. In this case, the correction noise amount is calculated based on the estimated noise amount at different resolution levels by the correction noise amount calculation processing.

Further, as in (Equation 1) and (Equation 2), the estimated noise amount z ′ may not be calculated for each of two or more edge directions o and two or more positions (m, n). In this case, z ′ _{j, o} [m, n] does not depend on the edge direction o or the position (m, n). It is not always necessary to calculate the estimated noise amount for a plurality of resolution levels, and the estimated noise amount z ′ _o [m, n] can also be calculated for each of a plurality of edge directions or a plurality of positions. The processing parameters in the noise amount estimation process include parameters for deciding which resolution level, edge direction, and position to estimate the noise amount, and parameters for specifying the noise amount calculation method. The processing parameter can be changed according to the image capturing information.

Next, a correction noise amount calculation method in the noise amount correction unit 209 of the image quality improvement processing unit 206 or 215 will be described with reference to FIG. A graph 601 in FIG. 6A shows an estimated noise amount z ′ _{j, o} [m, n] at each resolution level in a specific edge direction o and position (m, n) according to an embodiment of the correction noise amount calculation method. ] And the corrected noise amount z _{j, o} [m, n]. Also, the graph 602
Reliability of noise amount estimation at the resolution level j, edge direction o, and position (m, n) corresponding to the estimated noise amount z ′ _{j, o} [m, n] (hereinafter, estimated reliability) e _{j, o} [ m, n]. Estimated reliability e _j,
o [m, n] is calculated from, for example, one or more estimated noise amounts, or from the distance of estimated noise amounts calculated by different methods such as (Equation 1) and (Equation 2). It can be calculated, or can be calculated based on the decomposition coefficient in images of different frames, or a predetermined value can be used. In the case where a predetermined value is used, a method of preparing a value in a table so that an appropriate value can be used according to image capturing information may be used.

The correction noise amount is calculated based on the estimated noise amount and the reliability. In FIG. 6 (a), sets the threshold value T _e with respect to estimation reliability, and corrects only the estimated noise amount 603 corresponding to the threshold value T _e smaller estimation reliability 605, the correction noise amount 604 Calculated. The correction is obtained by interpolation processing with respect to the target estimated noise amount, using values of neighboring positions, neighboring frequencies, and other estimated noise amounts in the neighboring edge directions.

In the example of FIG. 6B, the estimated noise amount z ′ _{j, o} [m, n] and the correction noise at each position (m, n) (where n is fixed) in a specific edge direction o and resolution level j. The quantity z _{j, o} [m, n] is shown. An approximate curve 611 is obtained for the estimated noise amount z ′ _{j, o} [m, n], and a point on the obtained approximate curve is set as a correction noise amount. The approximate curve 611 is calculated based on the estimated reliability.

For example, a curve that minimizes E _z given by the following equation as a weighted least square approximation is selected.

Here, Z ′ _{j ′, o ′} [m ′, n ′] is a function representing the approximate curve 611, and depends on one or more parameters, the resolution level j ′, the edge direction o ′, and the positions m ′, n ′. It is a function that can be expressed.

The approximate curve 611 can be obtained by calculating the one or more parameters so that (Equation 3) is minimized. The sum of (Expression 3) on the right-hand side is the position m ′, n ′ in the vicinity, the frequency (resolution level) j ′ in the vicinity, the edge direction o in the vicinity with respect to the target estimated noise amount
Represents calculating the sum of '. Further, as in the example of the corrected noise amount 612, the corrected noise amount can be calculated by interpolation or approximation even at a position where the estimated noise amount has not been calculated in the noise amount estimating process described with reference to FIG.

In the example of FIG. 6C, the corrected noise amount z _{j is} obtained using the product of the estimated reliability e _{j, o} [m, n] and the estimated noise amount z ′ _{j, o} [m, n] as in the following equation. _{, o} [m, n].

This method is effective when the estimated noise amount is larger than the true noise amount as the estimated reliability is lower. Actually, when the noise amount estimation processing is performed using (Equation 1) or (Equation 2), the estimated noise amount is larger than the true noise amount because it is affected by the signal component included in the decomposition coefficient. May be a value.

Examples of the processing parameters in the correction noise amount calculation processing include parameters for specifying the estimation reliability and parameters for specifying a correction method based on the estimation reliability. These processing parameters are the image parameters. It can be changed according to imaging information.
Next, intensity conversion processing in the intensity conversion unit 210 of the image quality improvement processing unit 206 or 215 will be described with reference to FIGS.

FIG. 7A is a diagram showing an embodiment of the present invention showing a processing flow in the intensity conversion processing 104. In the present embodiment, the image quality improvement decomposition coefficient w ′ is generated by subjecting the decomposition coefficient w to the amplitude conversion processing 701 of the decomposition coefficient. In the decomposition coefficient amplitude conversion processing 701, the absolute value of the decomposition coefficient after amplitude conversion | w ′ _{j, o} [m, n] | is, for example, the same resolution level j, edge direction o, and position (m, n).
As a function of the decomposition coefficient w _{j, o} [m, n] and the corrected noise amount z _{j, o} [m, n] in FIG.

A (p; z) is an amplitude conversion function representing amplitude conversion. The amplitude conversion function A (p; z) is a function that monotonously increases with respect to the input p.

FIG. 8 shows an example of the amplitude conversion function. A function 801 shown in FIG. 8A is a soft thresholding method which is a widely known amplitude conversion method.

Is an amplitude conversion function at. Here, T is a constant multiple of the correction noise amount z (for example, T = 3z, k = 3).
It is.

An example of another amplitude conversion function is shown as a function 811 in FIG. Decomposition coefficient amplitude conversion processing 701
By _{| w 'j, o [m} , n] |> | w j, o [m, n] | if become the signal is emphasized, conversely _{| w' j, o [m} , n] | <
If | w _{j, o} [m, n] |, the signal is suppressed. Usually, decomposition coefficients w _{j, o} [m, n] is decomposed coefficient w _'j after amplitude conversion in the case of a real _{number, o} [m, n] code decomposition coefficients w _{j of, o} [m, n] of same city as the sign and decomposition coefficients _{w j, o [m, n} ] w 'j when the _{complex, o} [m, n] phase the w _{j of, o} [m, n] the same as the phase of the However, it is not limited to this.

FIG. 7B shows another embodiment of the processing flow in the intensity conversion processing 104. In this embodiment, in addition to the decomposition coefficient amplitude conversion process 701, a preservation degree calculation process 711 and a decomposition coefficient correction process 712 based on the preservation degree are performed. In the preservation degree calculation process 711, a preservation degree C representing the degree of preservation of the decomposition coefficient is calculated based on the estimated noise amount z and the values of a plurality of decomposition coefficients, and then each decomposition coefficient correction process 712 based on the preservation degree performs each decomposition. The coefficient w is corrected based on the conservation degree C. The degree of conservation C and the corrected decomposition coefficient w ⁺ are scalar values C _{j, o} [m, n], w ⁺ _{j, for} each position (m, n), each resolution level j, and each edge direction o _{. o} A vector with [m, n]. In the following, a vector having a scalar value a _{j, o} [m, n] with respect to position (m, n), resolution level j, and edge direction o is simply represented by a.

The corrected decomposition coefficient w ⁺ _{j, o} [m, n] is the degree of conservation C _{j, o} [m, n] and decomposition coefficient w _{j at} the same resolution level j, edge direction o, and position (m, n). _{, o As} a function of [m, n], it can be expressed as

Here, F (C _{j, o} [m, n]) is a monotonically increasing function. An example of F (C _{j, o} [m, n]) is shown in FIG. F (C _{j, o}
[m, n])> 1, then the decomposition coefficients w _j of the previous _{correction, o [m,} n] decomposition coefficient after correction compared to _{^{w + j, o [m,}} n] is more of increases.

In the decomposition coefficient amplitude conversion processing 701, the decomposition coefficient w ′ _{j, o} [m, n] after amplitude conversion is calculated using the following equation (Equation 8) instead of (Equation 5).

With respect to the processing flow in the intensity conversion processing 104, still another embodiment is shown in FIG.
In this embodiment, first, a decomposition coefficient amplitude conversion process 701 and a conservation degree calculation process 711 are performed. In the decomposition coefficient amplitude conversion processing 701, the decomposition coefficient w ″ after amplitude conversion is calculated using (Equation 9).

In the preservation degree calculation process 711, the preservation degree C is calculated based on a plurality of decomposition coefficient values. Subsequently, a decomposition coefficient correction process 712 is performed based on the conservation degree C. By this processing, the corrected decomposition coefficient w ′ _{j, o} [m, n] is calculated as follows.

In the decomposition coefficient correction processing 712 based on the degree of preservation in FIGS. 7B and 7C, it is not necessary to correct all the decomposition coefficients. For example, for the decomposition coefficient w _{J at the} resolution level J, F (C _{J , o} [m, n]) = 1 and no correction may be performed. As processing parameters in the intensity conversion processing shown in the embodiment of FIG. 7, for example, parameters for determining which processing of FIGS. 7A, 7B, and 7C to apply, amplitude conversion processing 701 of the decomposition coefficient, There are processing parameters for specifying each of the preservation degree calculation processing 711 and the decomposition coefficient correction processing 712 based on the preservation degree, and these processing parameters can be changed according to the image capturing information.

  Next, the conservation degree calculation process 711 in FIG. 7 will be described with reference to FIGS.

First, a method of performing intensity conversion using a plurality of decomposition coefficients will be described with reference to FIG. FIG. 10A shows an example of the relationship between the amplitude of the high-frequency decomposition coefficient w _{j, o} [m, n] and the position m. In the edge portion, generally, the amplitude of the decomposition coefficient in the corresponding edge direction o increases. A curve 1001 represents the amplitude of the decomposition coefficient obtained when only the signal component is extracted in the edge portion 1004, assuming that the signal component and the noise component have been correctly separated. Since the amplitude of the decomposition coefficient w _{j, o} [m, n] that is actually calculated from the captured image is affected by noise, the normal curve 1
The value is out of the range of 001. As a result, there are many cases where the value is smaller than the curve, such as the decomposition coefficient 1002. On the other hand, in the flat portion 1005, since the signal component contained in the decomposition coefficient w _{j, o} [m, n] is generally small, the amplitude of the decomposition coefficient is small. However, the amplitude may increase irregularly like the decomposition coefficient 1003 due to the influence of noise.

In the case of FIG. 10A, in order to sufficiently improve the image quality, the amplitude is preserved or increased with respect to the decomposition coefficient 1002 which is the edge portion, and conversely, the decomposition coefficient 1003 which is the flat portion is changed. For this, it is effective to reduce the amplitude. Therefore, in one embodiment of the present invention, there are generally many decomposition coefficients having large amplitudes in the vicinity of the decomposition coefficient 1002 that is the edge part, and conversely, generally the decomposition coefficient is in a position in the vicinity of the decomposition coefficient 1003 that is the flat part. In particular, the intensity conversion is performed by using the values of a plurality of decomposition coefficients at nearby positions. Such a process cannot be realized by the method of performing the intensity conversion using only the amplitudes of the individual decomposition coefficients, and according to the embodiment of the present invention, it is possible to perform the image quality improvement process with higher accuracy than before. .

In FIG. 10 (a), the method of obtaining a good image quality improvement result by using the relevance of the decomposition coefficients in the neighboring positions has been described. Similarly, the relevance in the decomposition coefficients of different resolution levels and different edge directions is also shown. It is also effective to use it. FIG. 10B shows an example of the relationship between the resolution coefficient amplitude and the resolution level. In the figure, a point represented by a circle represents the decomposition coefficient at the edge portion, and a point represented by an x mark represents the decomposition coefficient at the flat portion. A curve 1011 represents the amplitude of the decomposition coefficient obtained when only the signal component is extracted at the edge portion assuming that the signal component and the noise component can be correctly separated. The edge portion generally has a larger amplitude than the flat portion, but due to the influence of noise, the decomposition coefficient (for example, the decomposition coefficient 1013) in the flat portion is larger than the decomposition coefficient (for example, the decomposition coefficient 1012) in the edge portion. Cases can also occur. However, by using a plurality of decomposition coefficient values including decomposition coefficients at different frequency levels, it is possible to perform image quality improvement processing with higher accuracy than before.

FIG. 10C shows an example of the relationship between the amplitude of the decomposition coefficient and the edge direction. Similarly, by using the values of a plurality of decomposition coefficients including the decomposition coefficients in the neighboring edge directions, for example, the amplitude of the decomposition coefficient 1022 at the edge portion is increased and the amplitude of the decomposition coefficient 1023 at the flat portion is decreased. Such processing is possible. As described above, in one embodiment of the present invention, improvement in image quality improvement performance is realized by utilizing the relationship between a plurality of decomposition coefficients.

  Next, the conservation degree calculation processing described with reference to FIGS. 7B and 7C will be described with reference to FIG. For the calculation of the conservation degree C, a plurality of decomposition coefficient values are used. FIG. 11 (a) is an example showing a conservation degree calculation process.

In this embodiment, first, values C ^L , C ^O , and C ^S are obtained by blocks 1101, 1102, and 1103. Where the value C ^L is the resolution factor w _{1, o} [m, n], ..., w _{J, o} [m, n] for different resolution levels in the same edge direction o and position (m, n) As a function of

Here, C ^L (...) is a function, for example,

It is expressed as

Since the processing input image x generally includes more signal components as the low frequency component, (Equation 12)
As described above, by using a resolution factor of a high resolution level corresponding to a lower frequency with respect to the target resolution level j, it is possible to identify a signal and noise with high accuracy. Similarly to the calculation of C ^L , the decomposition coefficients w _{j, A} [m, n], ..., w _{j, K} [m, n for different edge directions at the same resolution level j and position (m, n) As a function of n], the value C ^O is calculated as follows:

K is the number of divisions in the edge direction at the resolution level j. C ^O (...) is a function, for example

It is expressed as Here, o ₁ and o ₂ represent edge directions on both sides of the edge direction o.

Furthermore, the decomposition coefficients w _{j, o} [m ⁽¹⁾ , n ⁽¹⁾ ], ..., w _{j, o} [m ^(S) , n ^{(S )} as a function of, calculating the value C ^S as follows.

Where w _{j, o} [m ⁽¹⁾ , n ⁽¹⁾ ], ..., w _{j, o} [m ^(S) , n ^(S) ] are all positions in resolution level j and edge direction o Represents the decomposition coefficient. C ^S (...) is a function, for example, using specific weights a _{j, o} [m ′, n ′],

Calculate a weighted average like

FIG. 12 shows an example representing the weights a _{j, o} [m ′, n ′]. By using a _{j, o} [m ′, n ′] having a non-zero value along the edge direction, the weighted average is performed along the edge direction. Since signal components such as a tissue structure included in an image generally have similar brightness values along the edge direction, the corresponding decomposition coefficient also has a substantially similar value along the edge direction. Therefore, by performing smoothing processing by weighted averaging along the edge direction as shown in FIG. 12, it is possible to suppress the noise component without degrading the signal component.

In FIG. 3, when the thinning process is performed after the low-pass filter or the high-pass filter, the resolution coefficient w _{1, o} [m, n] of all resolution levels at the position (m, n) depends on the position (m, n). , ...
, W _{J, o} [m, n], some values may not be obtained. In such a case, after obtaining the value of the required decomposition coefficient by processing such as approximating with the decomposition coefficient at the nearest position or performing interpolation using multiple decomposition coefficients at the nearest position, save Perform the degree calculation process.

Also, when using a multi-resolution decomposition method in which the number of divisions in the edge direction differs depending on the resolution level, depending on the edge direction o, the decomposition coefficients w _{1, o} [m, n],. ., W _{J, o} [m, n] may not be obtained. In this case as well, the value of the required decomposition coefficient was obtained by processing such as approximation with the decomposition coefficient in the nearest edge direction or interpolation using a plurality of decomposition coefficients in the neighboring edge directions. A conservation degree calculation process is performed later. C ^L , C ^O , and C ^S can be functions that always return constant values, or functions that simply return w _{j, o} [m, n].

Next, at block 1104 in FIG. 11, the same resolution level j, edge direction o, position (m, n
The degree of conservation C is calculated using the values C ^L _{j, o} [m, n], C ^O _{j, o} [m, n], and C ^S _{j, o} [m, n] in the following equation.

C (...) is a function, for example

It is expressed as

In the embodiment shown in FIG. 11A, the functions C ^L (...), C ^O (...), C ^S (...) Are used in parallel, but the conservation degree calculation process shown in FIG. As in the first embodiment, the functions C ^L (...), C ^O (...), C ^S (...) may be used in series. In the embodiment of FIG. 11B, first, the block 1111 causes the decomposition coefficient w _{j, o} [m ⁽¹⁾ , n ⁽¹⁾ ],... At the same resolution level j and edge direction o to be different. , w _{j, o} [m ^(S) , n ^(S) ] as a function, the value C ^S is calculated by the calculation method shown in (Formula 15). next,
Block 1112 allows different edge direction values C ^S _{j, A} [m, n at the same resolution level j and position (m, n) using the function C ^O (...) As used in (Equation 13). ], ..., C ^S _{j, K From} [m, n], the value C ^OS is calculated as follows:

Subsequently, at block 1113, using the function C ^L (...) As used in (Equation 11), different resolution level values C ^OS _{1, o} [in the same edge direction o and position (m, n) are used. m, n], ..., C ^OS _{J, o The} degree of conservation C is calculated from [m, n].

In the embodiment of FIG. 11B, the function is applied to the decomposition coefficient w in the order of C ^S (...), C ^O (...), C ^L (. .

FIG. 11 (c) is another example showing the conservation degree calculation process. In this embodiment, the function CL (...
), C ^O (...), C ^S (...) In parallel with the processing using the processing using serial processing. First, after calculating the values C ^S and C ^O using (Equation 15) and (Equation 13) in blocks 1121 and 1122, the same resolution level j and edge direction are calculated in block 1123.
The value C ^{O + S} is calculated as follows using the values C ^o _{j, o} [m, n] and C ^S _{j, o} [m, n] at o and position (m, n).

C ^{O + S} (…) is a function, for example

It is expressed as

Next, the block 1124 uses the function C ^L (...) As used in (Equation 11) to use different resolution level values C ^{O + S} _{1, in the} same edge direction o and position (m, n) _{. o The} degree of conservation C is calculated from [m, n], ..., C ^{O + S} _{J, o} [m, n].

In the embodiment of FIG. 11 (c), the functions C ^S (...), C ^O (...) Are used in parallel for the decomposition coefficient w, and then the function C ^L is used in series. It is not limited to this combination.
Next, conservation degree calculation processing using a decomposition coefficient for images in different frames will be described with reference to FIGS.

FIG. 13 shows the relationship between the amplitude of the high-frequency decomposition coefficient w _{j, o} ^(u) [m, n] and the frame u. Here, a decomposition coefficient obtained by performing multi-resolution decomposition on an image captured in the frame u is referred to as w _{j, o} ^(u) [m, n]. Also, the frame subject to image quality improvement processing is called a frame t, and the decomposition coefficient w _{j, o} ^(t) [m, n] is simply abbreviated as w _{j, o} [m, n]. In the following, simply a vector with scalar value a _{j, o} ^(t) [m, n] for position (m, n), resolution level j, edge direction o, frame u
Represented by
In FIG. 13, a point represented by a circle represents a decomposition coefficient at the edge portion, and a point represented by a cross represents a decomposition coefficient at the flat portion. A curve 1301 represents the amplitude of the decomposition coefficient obtained when only the signal component is extracted at the edge portion, assuming that the signal component and the noise component have been correctly separated. Since the amplitude of the decomposition coefficient w _{j, o} ^(u) [m, n] actually calculated from the captured image is affected by noise, it may be smaller than the curve 1301 like the decomposition coefficient 1302. On the other hand, in general, even in a flat portion that does not contain much signal components, it is affected by noise, and the amplitude may become large like the decomposition coefficient 1303.

  In order to improve the image quality, it is necessary to preserve or increase the amplitude with respect to the decomposition coefficient 1302 that is the edge portion, and conversely to reduce the amplitude with respect to the decomposition coefficient 1303 that is the flat portion. This method cannot be realized by a method of performing intensity conversion using only the amplitude of each decomposition coefficient. On the other hand, in one embodiment of the present invention, the expansion coefficient of a neighboring frame at the same resolution level, the same edge direction, and the same position as the resolution coefficient 1302 that is an edge portion generally has a large amplitude, and conversely, at a flat portion. Focusing on the fact that the expansion coefficient of a neighboring frame at the same resolution level, the same edge direction, and the same position as a certain decomposition coefficient 1303 generally has a small amplitude, the values of a plurality of decomposition coefficients including the decomposition coefficient in the neighboring frame are set. Use to convert intensity. Thereby, it is possible to improve the identification accuracy of a signal component and a noise component.

Next, the preservation degree calculation process in one embodiment of the present invention will be described with reference to FIG. The embodiment of FIG. 14 performs the same processing as the embodiment of FIG. 11, but differs from the embodiment of FIG. 11 in that the degree of conservation is calculated using a plurality of frames of decomposition coefficients. FIG. 14A is an example showing the conservation degree calculation process. In this embodiment, the values C ^T , C ^L , C ^O , and C ^S are obtained by the blocks 1101, 1102, 1103, and 1401. Here, the value C ^T is the decomposition coefficient w _{j, o} ^{(t−t ′ + 1)} [m, n], w _j of different frames at the same resolution level j, edge direction o and position (m, n) _{, o} ^{(t-t '+ 2)} [m, n] ..., w _{j, o} ^{(t) As} a function of [m, n], the following calculation is performed.

Where C ^T (...) is a function, for example

It is expressed as The value t ′ in (Equation 24) does not need to be a constant and may be variable.

Next, at block 1402, the degree of conservation C is calculated using the values C ^L , C ^O , C ^S , and C ^T as shown in the following equation.

C ⁽⁴⁾ (...) is a function, for example

It is expressed as

FIG. 14B is another example showing the conservation degree calculation process. In the embodiment of FIG. 14A, functions C ^L (...), C ^O (...), C ^S (...), C ^T (...) Are used in parallel, whereas FIG. In the embodiment, functions C ^L (...), C ^O (...), C ^S (...), C ^T (...) Are used in series as in the embodiment of FIG. First, in block 1411, the same resolution level j, edge direction o and position (m, n
) Decomposition coefficients w _{j, o} ^{(t-t '+ 1)} [m, n], w _{j, o} ^{(t-t' + 2)} [m, n], ..., w _{j, o} ^(t)
As a function of [m, n], a value ^CT is calculated by a calculation method such as (Equation 24).

Next, the block 1412 uses the function C ^S (...) As used in (Equation 15) to use different position values C ^T _{j, o} [m ^{(1) in the} same resolution level j and edge direction o. , n ⁽²⁾ ], ..., C ^T _{j, o} [m ^(S) , n ^(S) ], a value C ^ST as shown in the following equation is calculated.

Hereinafter, similarly, the blocks 1413 and 1414 perform the following calculations using the functions C ^L (...) And C ^O (...) As used in (Equation 11) and (Equation 13). Calculate the conservation degree C.

In the embodiment shown in FIG. 14B, the functions for the decomposition coefficient w are C ^T (...), C ^S (...), C ^O (...), C ^L (
...) in this order, but is not limited to this order. Furthermore, as another embodiment (not shown), a process such as a combination of serial processing and parallel processing may be used as the processing of the embodiment of FIG.

Next, the reconstruction process will be described with reference to FIG.
FIGS. 15 (a), (b), and (c) show the Wavelet transform, complex Wavelet transform, and Steerable Pyrami, respectively.
An embodiment of a reconstruction processing flow corresponding to a multi-resolution decomposition method based on d conversion will be described. These correspond to the processing flows of FIGS. 3 (a), (b), and (c), respectively. The decomposition coefficients s and w in FIG. 15 represent the decomposition coefficients after amplitude correction.

First, an embodiment diagram representing a reconstruction processing flow in the case of the Wavelet transform of FIG. In the present embodiment, reconstruction is performed by a procedure reverse to the multiresolution decomposition step described with reference to FIG. That is, first, the high frequency resolution coefficients w _{J−1, B} and w _J− 1D of the resolution level J−1 (J = 3 in this embodiment) pass through edges parallel to the straight line m = n, respectively. Straight line m = -n
And a filter 1502 that blocks edges parallel to the straight line m = n and passes through edges parallel to the straight line m = −n. The sum w _{J−1, diag} of the output obtained by the filter processing is calculated.

Next, a vertical one-dimensional low-pass filter 1504 and a vertical one-dimensional high-pass filter are applied to the decomposition coefficients s _J-1 , w _{J-1, C} , w _{J-1, A} , w _{J-1, diag} . After applying 1505, the sum is calculated by blocks 1506 and 1507 for the output of each filter. Subsequently, the horizontal one-dimensional low-pass filter 1508 and the high-pass filter 1509 are applied, and then the sum thereof is calculated by the block 1510, whereby the low-frequency decomposition coefficient s _{J-2 at the} resolution level _J-2.
Ask for.

As in the case of multiresolution decomposition, the coefficients of these filters are real numbers, and the order of applying the horizontal filter and the vertical filter may be reversed. Further, in the multi-resolution decomposition process described with reference to FIG. 3A, when the thinning process is performed immediately after the filter process, the lightness value is zero every other pixel immediately before the corresponding filter process in the present embodiment. Insertion processing for inserting pixels (hereinafter, interpolation processing) is performed. Thereafter, the low frequency decomposition coefficient s _j-1 at the resolution level j-1 is obtained from the resolution coefficient at the resolution level j by recursively performing the same processing as described above. Finally, an image quality improved image y is obtained by performing the same processing as described above from the resolution coefficient of resolution level 1.

  Next, a description will be given of an embodiment showing a reconstruction processing flow in the case of the complex wavelet transform of FIG.

First, the high-frequency decomposition coefficients w _{J−1, A} , w _{J−1 of the} resolution level J−1 (J = 3 in the present embodiment) are performed by a procedure reverse to the multi-resolution decomposition step described in FIG. _{, B} , w _{J-1, C} , w _{J-1, D} , w _{J-1, E} , w _J
A process of calculating the sum and difference of the input signals by the ΣΔ block 1521 is performed on _{−1 and F.} Next, the output signal of the ΣΔ block 1521 and the low frequency resolution coefficients s _{J−1, A} , s _{J− of the} resolution level J−1.
After applying the vertical one-dimensional low-pass filters 1522, 1523 and the vertical one-dimensional high-pass filters 1524, 1525 to _{1, B} , the sum is calculated by blocks 1526, 1527, 1528, 1529. Subsequently, the horizontal one-dimensional low-pass filters 1530 and 15
31 and high-pass filters 1532 and 1533, and then the sum is calculated by blocks 1534 and 1535, so that the low frequency decomposition coefficients s _{J-2, A} , s _{J-2, B at} resolution level J-2 are obtained.
Ask for.

In the embodiment diagram, two different types of filters are used for each filter process, but the present invention is not limited to this. The coefficients of these filters are complex numbers. The order of applying the horizontal filter and the vertical filter may be reversed.

In the multi-resolution decomposition process described with reference to FIG. 3B, when the thinning process is performed immediately after the filter process, the interpolation process is performed immediately before the corresponding filter process in this embodiment. Thereafter, the same processing as described above is performed recursively to obtain the low frequency decomposition coefficients s _{j−1, A} and s _{j− 1B of} the resolution level j−1 from the resolution coefficient of the resolution level j. Finally, an image quality improved image y is obtained by performing the same processing as described above from the resolution coefficient of resolution level 1.

Next, an example diagram showing a reconstruction processing flow in the case of Steerable Pyramid conversion in FIG. First, the decomposition coefficients w _{J−1, A} , w _{J−1 of the} resolution level J−1 (J = 3 in the present embodiment) are performed in the reverse procedure to the multiresolution decomposition step described in FIG. _B , w _{J-1, C} ,…
In contrast, filters 1541, 1542, 1543,... That pass only components in a specific edge direction at a high frequency in either the vertical or horizontal direction are applied. In addition, a two-dimensional low-pass filter 1544 that passes only a component having a low frequency in both the vertical and horizontal directions is applied to the decomposition coefficient s _J-1 .
Next, the sum of the outputs of the filters 1541, 1542, 1543, and 1544 is calculated by the block 1545 to obtain s _J-2 .

In the multi-resolution decomposition process described in FIG. 3C, when the thinning process is performed immediately after the two-dimensional low-pass filter process, the interpolation process is performed immediately before the corresponding two-dimensional low-pass filter process in this embodiment. Apply. Thereafter, the low frequency decomposition coefficient s _j-1 at the resolution level j-1 is obtained from the resolution coefficient at the resolution level j by recursively performing the same processing as described above. Finally, a two-dimensional high-pass filter 1 that cuts off only a signal obtained by applying the two-dimensional low-pass filter 1546 to s ₁ and a component having a low frequency in both the vertical and horizontal directions with respect to w ₁ .
An image quality improved image y is obtained by calculating the sum in block 1547 for the signal obtained by applying 545.

  Next, image quality improvement processing in which processing parameters can be changed according to image capturing information will be described with reference to FIGS.

  FIG. 16 is a diagram showing an example of the flow of image quality improvement processing in the present invention. First, in a processing parameter determination process 1601, processing parameters are determined based on imaging conditions, image types, and imaging targets, which are image imaging information. The processing parameters for each image capturing information are tabulated in advance. Furthermore, it is possible to provide a function that allows the user to adjust the value based on the processing parameter obtained from the table. Next, multi-resolution decomposition processing 101, noise amount estimation processing 102, correction noise amount calculation processing 103, intensity conversion processing 104, and reconstruction processing 105 are performed using the determined processing parameters. However, it is also possible to prepare a plurality of algorithms for performing different processes in each process and switch them according to process parameters.

  FIG. 17 is an example showing a table describing processing parameters for each image capturing information used in the processing parameter determination processing 1601 of FIG. The values in the table 1701 may be fixed in advance or may have a function that can be changed by the user. Each row of the table 1701 represents a processing parameter used in each image capturing information.

  The imaging information includes the imaging conditions in the column 1702, the type of the captured image in the column 1703, and the imaging target in the column 1704. Specifically, the imaging conditions include the type of ultrasonic probe, display magnification, frequency band used for ultrasonic transmission / reception signals, application of spatial compound method, application of frequency compound method, scanning pitch of ultrasonic transmission signal, etc. Can be mentioned. The processing parameters include parameters related to multi-resolution decomposition processing shown in column 1705, and parameters related to processing such as noise amount estimation processing, correction noise amount calculation processing, coefficient strength conversion processing, and reconstruction processing in column 1706.

The table has a format in which processing parameters for each image capturing information are uniquely determined. In this embodiment, the processing parameter described in the top row is applied among the rows that match the image capturing information in the table.

DESCRIPTION OF SYMBOLS 101 ... Multi-resolution decomposition process 102 ... Noise amount estimation process 103 ... Correction noise amount calculation process 104 ... Intensity conversion process 105 ... Reconstruction process 201 ... Ultrasonic diagnostic apparatus
DESCRIPTION OF SYMBOLS 202 ... Drive circuit 203 ... Ultrasonic probe 204 ... Reception circuit 205 ... Image generation part 206 ... Image quality improvement process part 207 ... Multi-resolution decomposition part 208 ... Noise amount estimation part 209 ... Noise amount correction part 210 ... Intensity conversion part 211 ... Re Configuration unit 212 ... Scan converter 213 ... Display unit 221 ... Input unit 222 ... Control unit
223: Storage unit 224: Processing unit 701: Amplitude conversion process of decomposition coefficient 711: Preservation degree calculation process 712: Decomposition coefficient correction process based on preservation degree

Claims (6)

A method for generating an image from an ultrasonic wave received by an ultrasonic probe of an ultrasonic diagnostic apparatus, improving the image quality of the generated image, and displaying the image with the improved image quality,
In the step of improving the image quality of the generated image,
Obtaining a decomposition coefficient for each position, for each resolution level, and for each edge direction by multi-resolution decomposition on the generated image,
Calculate an estimated noise amount based on the obtained decomposition coefficient,
Correcting the estimated noise amount using a product of the reliability of the estimated noise amount set in advance for each resolution level or edge direction and the estimated noise amount;
Intensity conversion is performed on each of the obtained decomposition coefficients using the corrected estimated noise amount information,
Reconstructing the image by performing a reconstruction process on each decomposition coefficient that has undergone the intensity conversion,
An image quality improvement method for an ultrasonic diagnostic apparatus. A method for generating an image from an ultrasonic wave received by an ultrasonic probe of an ultrasonic diagnostic apparatus, improving the image quality of the generated image, and displaying the image with the improved image quality,
In the step of improving the image quality of the generated image,
Obtaining a decomposition coefficient for each position, for each resolution level, and for each edge direction by multi-resolution decomposition on the generated image,
Calculate an estimated noise amount based on the obtained decomposition coefficient,
The estimated noise amount is calculated based on a product of the reliability of the estimated noise amount calculated using the image decomposition coefficient in different frames or the distance of the estimated noise amount calculated by a plurality of different methods and the estimated noise amount. Correct,
Intensity conversion is performed on each of the obtained decomposition coefficients using the corrected estimated noise amount information,
Reconstructing the image by performing a reconstruction process on each decomposition coefficient that has undergone the intensity conversion,
An image quality improvement method for an ultrasonic diagnostic apparatus. A method for generating an image from an ultrasonic wave received by an ultrasonic probe of an ultrasonic diagnostic apparatus, improving the image quality of the generated image, and displaying the image with the improved image quality,
In the step of improving the image quality of the generated image,
Obtaining a decomposition coefficient for each position, for each resolution level, and for each edge direction by multi-resolution decomposition on the generated image,
Estimating the amount of noise based on the obtained decomposition coefficient,
For each of the obtained decomposition coefficients, an intensity conversion is performed using the information on the estimated noise amount and information on two or more decomposition coefficients,
An image is reconstructed by performing a reconstruction process on each decomposition coefficient subjected to the intensity conversion,
Information on two or more decomposition coefficients used for the intensity conversion includes information on a decomposition coefficient that differs only in position, a decomposition coefficient that differs only in the edge direction, and a decomposition coefficient that differs only in the resolution level. Image quality improvement method for diagnostic equipment. An ultrasonic probe that emits ultrasonic waves to the subject and receives reflected waves from the subject, an image generation unit that generates an image from the reflected waves received by the ultrasonic probe, and an image generated by the image generation unit An ultrasonic diagnostic apparatus comprising image quality improvement means for improving the image quality of an image and display means for displaying an image whose image quality has been improved by the image quality improvement means,
The image quality improving means is:
A multi-resolution decomposition unit for determining a decomposition coefficient for each position, for each resolution level, and for each edge direction by multi-resolution decomposition on the image generated from the received reflected wave;
A noise amount estimation unit that calculates an estimated noise amount based on the decomposition coefficient obtained by the multi-resolution decomposition unit;
A correction noise amount calculation unit that corrects the estimated noise amount using a product of reliability and the estimated noise amount set in advance for each resolution level or for each edge direction;
An intensity conversion unit that performs intensity conversion on each of the obtained decomposition coefficients using the corrected estimated noise amount information;
An image reconstruction unit that reconstructs an image by performing reconstruction processing on each decomposition coefficient that has undergone intensity conversion in the intensity conversion unit;
An ultrasonic diagnostic apparatus comprising: An ultrasonic probe that emits ultrasonic waves to the subject and receives reflected waves from the subject, an image generation unit that generates an image from the reflected waves received by the ultrasonic probe, and an image generated by the image generation unit An ultrasonic diagnostic apparatus comprising image quality improvement means for improving the image quality of an image and display means for displaying an image whose image quality has been improved by the image quality improvement means,
The image quality improving means is:
A multi-resolution decomposition unit for determining a decomposition coefficient for each position, for each resolution level, and for each edge direction by multi-resolution decomposition on the image generated from the received reflected wave;
A noise amount estimation unit that calculates an estimated noise amount based on the decomposition coefficient obtained by the multi-resolution decomposition unit;
The estimated noise amount is calculated based on a product of the reliability of the estimated noise amount calculated using the image decomposition coefficient in different frames or the distance of the estimated noise amount calculated by a plurality of different methods and the estimated noise amount. A correction noise amount calculation unit to correct,
An intensity conversion unit that performs intensity conversion on each of the obtained decomposition coefficients using the corrected estimated noise amount information;
An image reconstruction unit that reconstructs an image by performing reconstruction processing on each decomposition coefficient that has undergone intensity conversion in the intensity conversion unit;
An ultrasonic diagnostic apparatus comprising: An ultrasonic probe that emits ultrasonic waves to the subject and receives reflected waves from the subject, an image generation unit that generates an image from the reflected waves received by the ultrasonic probe, and an image generated by the image generation unit An ultrasonic diagnostic apparatus comprising image quality improvement means for improving the image quality of an image and display means for displaying an image whose image quality has been improved by the image quality improvement means,
The image quality improving means is:
A multi-resolution decomposition unit for determining a decomposition coefficient for each position, for each resolution level, and for each edge direction by multi-resolution decomposition on the image generated from the received reflected wave;
A noise amount estimation unit for estimating a noise amount based on the decomposition coefficient obtained by the multi-resolution decomposition unit;
For each of the obtained decomposition coefficients, an intensity conversion unit that performs intensity conversion using information on the estimated noise amount and information on at least two or more decomposition coefficients;
An image reconstruction unit that reconstructs an image by performing reconstruction processing on each decomposition coefficient that has undergone intensity conversion in the intensity conversion unit;
Have
The intensity conversion unit includes, as information on two or more decomposition coefficients used for the intensity conversion, information on decomposition coefficients that differ only in position, decomposition coefficients that differ only in the edge direction, and decomposition coefficients that differ only in resolution level. An ultrasonic diagnostic apparatus characterized by the above.

Images