Image processing method and image processing device

Abstract

An image processing method comprises: a first image acquisition step of acquiring a first image; a deviation field acquisition step of acquiring a deviation field representing intensity non-uniformity in the first image; a segmentation step of segmenting the first image; and an output step of outputting a second image corrected for intensity non-uniformity based on the deviation field and a result obtained by segmenting the first image, wherein the segmentation step comprises edge-preserving smoothing processing.

Description

Technical Field

[0001] The invention relates to an image processing method and an image processing device. Background Art

[0002] Images generated by MRI (Magnetic Resonance Imaging) technology have the problem of intensity inhomogeneity. The pixel value of the MRI image affected by intensity inhomogeneity is expressed by the following formula:

[0003] I=B·J+n, (1)

[0004] Where I represents the pixel value of the MRI image affected by the intensity inhomogeneity, B represents the bias field, which represents the intensity inhomogeneity, J represents the pixel value when it is not affected by the intensity inhomogeneity, and n represents noise. In addition, J is referred to as the "ideal pixel value" below.

[0005] Intensity non-uniformity in MRI images can affect the accuracy of image diagnosis, so it is desirable to correct it.

[0006] As a method for correcting the intensity unevenness of an MRI image to obtain an image with an ideal pixel value J, there is an intensity unevenness correction method based on segmentation. For MRI images, since a more accurate segmentation result can obtain a more accurate intensity unevenness estimation result, and a more accurate intensity unevenness estimation result can also obtain a more accurate segmentation result, the intensity unevenness correction method based on segmentation gradually obtains more accurate intensity unevenness estimation results and segmentation results by repeatedly iteratively updating the segmentation results and the intensity unevenness estimation results.

[0007] However, the intensity non-uniformity correction method based on segmentation is easily affected by noise, which affects the accuracy of segmentation and intensity non-uniformity correction. In this regard, non-patent document 1 discloses a method for intensity non-uniformity correction based on segmentation, which uses the spatial information of the MRI image in updating the segmentation result and the non-uniformity estimation result, thereby reducing the impact of noise. Patent document 1 also discloses a method for intensity non-uniformity correction based on segmentation, which reduces the impact of noise by utilizing the information of adjacent pixels.

[0008] Prior art literature

[0009] Non-patent literature

[0010] Non-patent literature 1: Wenchao Cui, Yi Wang, Yangyu Fan, Yan Feng, and TaoLei. 2013. Localized FCM clustering with spatial information for medical imagesegmentation and bias field estimation. Journal of Biomedical Imaging 2013, Article 13 (January 2013)

[0011] Patent Literature

[0012] Patent document 1: CN103996193A Summary of the invention

[0013] The problem to be solved by the embodiments shown in this specification and the drawings is to correct the intensity non-uniformity of MRI images with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The diagram shows an example of image segmentation of a brain MRI image.

[0015] Figure 2 This is a block diagram showing the structure of the image processing device according to the first embodiment.

[0016] Figure 3 This is a flowchart showing a basic filter processing method of the image processing method used by the image processing device of the first embodiment.

[0017] Figure 4 This is a flowchart showing a basic filter processing method of the image processing method used by the image processing device of the second embodiment.

[0018] Figure 5 This is an effect diagram showing the effect of filtering processing of membership function values ​​in the image processing method of embodiment 2.

[0019] Figure 6 This is a diagram comparing the effect of the membership function value filtering process of the image processing method of the second embodiment and the effect of the filtering process of the first embodiment.

[0020] Figure 7 It is a diagram comparing the segmentation results of the image processing method of embodiment 2, the image processing method of embodiment 1, and the image processing method of the prior art.

[0021] Figure 8 This is a diagram showing the effect of the image processing method of embodiment 2.

[0022] Fig. 9 This is a block diagram showing the structure of an image processing device according to a third embodiment.

[0023] Fig.10 This is a flowchart showing an image processing method used by the image processing device according to the third embodiment.

[0024] Fig.11 This is a block diagram showing the structure of an image processing device according to a fourth embodiment.

[0025] Fig.12 This is a flowchart showing an image processing method used by the image processing device according to the fourth embodiment.

[0026] Fig.13 It is a diagram showing the effect of the intensity unevenness correction method according to the fourth embodiment. DETAILED DESCRIPTION

[0027] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0028] (Implementation Method 1)

[0029] Embodiment 1 of the present invention relates to an image processing device and an image processing method. The image processing method of this embodiment classifies each pixel of the MRI image through the fuzzy c-means clustering algorithm, that is, assigns a tissue category to each pixel, thereby completing the segmentation of the MRI image. In any MRI image, if there is no influence of uneven intensity, it can be considered that the ideal pixel values ​​of pixels belonging to the same category are the same. In other words, the pixel values ​​of pixels belonging to the same tissue are the same. In this embodiment, there are a total of n categories (n is a natural number greater than 2), and the pixel value of the cluster center of category i is set to c. i (i is a natural number less than or equal to n).

[0030] The principle of MRI image segmentation based on the fuzzy c-means clustering algorithm is described below with reference to the accompanying drawings.

[0031] Figure 1 FIG. 1 is a diagram showing an example of image segmentation for a brain MRI image. Figure 1 In the figure, (a) represents the segmented brain MRI image, (b) represents the segmented cerebrospinal fluid image, wherein the white part represents the cerebrospinal fluid, (c) represents the segmented gray matter image, wherein the white part represents the gray matter, and (d) represents the segmented white matter image, wherein the white part represents the white matter.

[0032] In the fuzzy c-means clustering algorithm, the membership function for each category is set, u irepresents the membership function for category i, u i (x) represents the membership function value of pixel x for category i, which means the probability that pixel x belongs to category i. The category corresponding to the largest membership function value among the membership function values ​​of each pixel for each category is set as the category of the pixel.

[0033] For any pixel x, its membership function value should satisfy:

[0034]

[0035] Formula (2) indicates that for any pixel x, the sum of the membership function values ​​of all n categories corresponding to it is 1.

[0036] In the fuzzy c-means clustering algorithm, the pixel value c of the cluster center of each category i is continuously updated. i , the deviation field B(x) of pixel x and the membership function value u of pixel x for category i i (x) so that the value of the energy function E representing the distance between the current classification of each pixel of the image and the optimal classification gradually decreases. When the energy function E reaches the minimum value, the membership function value u of each pixel x under the optimal classification for category i is obtained. i (x), thereby determining the optimal classification of each pixel x. The energy function E is defined by formula (3):

[0037]

[0038] Among them, Ω represents the domain of pixels, I(x) represents the pixel value of pixel x that is not affected by intensity inhomogeneity, K represents the weight function, and q is a natural number coefficient greater than 1, which determines the fuzziness of the fuzzy c-means clustering algorithm.

[0039] Hereinafter, an image processing device and an image processing method according to the present embodiment will be described with reference to the drawings.

[0040] Figure 2 This is a block diagram showing the structure of the image processing device according to this embodiment.

[0041] The image processing device 100 according to the first embodiment of the present invention includes an initialization unit 110 , an update unit 120 , a filter unit 130 , a determination unit 140 , a division unit 150 , and a storage unit 160 .

[0042] The initialization unit 110 initializes various parameters used in the image processing method and stores the initialized parameters in the storage unit 160. The above parameters can be any parameters used for segmenting MRI images, including pixel values ​​of cluster centers of each category, membership function values ​​of each pixel for each category, etc.

[0043] The updating unit 120 updates the parameters stored in the storage unit 160 , and includes a cluster center updating unit 121 , a membership function updating unit 122 , and a deviation field updating unit 123 .

[0044] The filtering unit 130 filters the membership function value updated by the updating unit 120 to remove noise components contained therein.

[0045] The determination unit 140 determines whether the membership function value after filtering has converged.

[0046] When the determination unit 140 determines that the membership function value has converged, the segmentation unit 150 segments the MRI image based on the membership function value.

[0047] The storage unit 160 stores various data used in the image processing method.

[0048] Next, the flow of an image processing method used by the image processing device of the present embodiment will be described with reference to the drawings.

[0049] Figure 3 This is a flowchart showing an image processing method used by the image processing apparatus of this embodiment.

[0050] After receiving the instruction to start the MRI image segmentation process, the image processing device 100 acquires the MRI image to be segmented and starts the MRI image segmentation process. The MRI image acquired first is an MRI image having intensity inhomogeneity.

[0051] In step S101, the initialization unit 110 obtains image information such as size information and measurement intensity of the MRI image, and performs initialization processing on each parameter based on the image information and the initial values ​​or initialization conditions set for each parameter used in this embodiment stored in the storage unit 160. The initialization unit 110 assigns the initial value to the parameter set with the initial value, and initializes the parameter that has no initial value but has the initialization condition set according to the initialization condition set for it.

[0052] In step S102 , the cluster center updating unit 121 updates the cluster center of each category based on the pixel information of the MRI image, the membership function value of each pixel on the MRI image for each preset category, and the deviation field indicating the intensity inhomogeneity of the MRI image.

[0053] In this embodiment, the pixel value c of the cluster center of category i stored in the storage unit 160 is updated according to formula (4) and formula (5): i .

[0054]

[0055]

[0056] In formula (4) and formula (5), * represents the convolution operator, and β, σ, and ρ are constant parameters. In this embodiment, a truncated Gaussian kernel is used as the weight function K, and β=1, σ=1, ρ=3, and q=2 are set, but their values ​​can also be changed according to actual needs.

[0057] By using formula (4) and formula (5), the cluster center updating unit 121 updates the cluster center according to the deviation field B(x) of each pixel x and the membership function value u for category i. i (x) The pixel value c of the cluster center of category i i The value of is updated to make it closer to the ideal pixel value that is not affected by intensity non-uniformity.

[0058] In step S103 , the deviation field updater 123 updates the deviation field of the MRI image based on the pixel information of the MRI image, the cluster centers of each category, and the membership function value of each pixel on the MRI image for each preset category.

[0059] In this embodiment, the deviation field B(x) stored in the storage unit 160 is updated according to the formula (6).

[0060]

[0061] According to formula (6), the deviation field updating unit 123 calculates the membership function value u of each pixel x for each category i according to the current membership function value u i (x) and the pixel value c of the cluster center of category i i The deviation field B(x) is updated to be closer to a deviation field that correctly represents the actual intensity inhomogeneity.

[0062] In step S104 , the membership function updater 122 updates the membership function value of each pixel on the MRI image for each preset category based on the pixel information of the MRI image, the cluster centers of each category, and the deviation field corresponding to each pixel on the MRI image.

[0063] In this embodiment, the membership function value u of each pixel x stored in the storage unit 160 is updated for each category i one by one according to formula (7): i (x).

[0064]

[0065] According to formula (7), the membership function updating unit 122 updates the pixel value c of the cluster center of category i according to the current deviation field B(x) i For the membership function value u i (x) is updated to make it closer to the membership function value under the optimal classification condition.

[0066] In this embodiment, the deviation field B(x), the pixel value c of the cluster center of each category i And the membership function value u i (x) After the update is completed, the filter unit 130 executes steps S105 to S110 to update the membership function value u of each category i. i (x) Filter to remove the noise components.

[0067] Due to the influence of the environment and equipment, there is noise in the MRI image obtained by the MRI device. Therefore, the membership function value u about the pixel x obtained based on the information of the MRI image is i (x) also contains noise, so the membership function value u needs to be i (x) is filtered to remove the noise component, and v i (x) represents the membership function value after filtering, then:

[0068] v i (x) = u i (x)-n i (x) (8)

[0069] where n i (x) represents the membership function value u of pixel x for category i i (x) contains noise components.

[0070] Steps S105, S109, and S110 of the image processing method of this embodiment perform edge-preserving smoothing processing. In this embodiment, guided image filtering is used to calculate the membership function value u of each pixel x for each category i. i (x) is filtered to obtain the membership function value v after filtering i (x). Guided filtering is an edge-preserving image filtering technique, which is usually used to perform edge-preserving smoothing on grayscale images or RGB images. It uses a guided image to filter the original image so as to reduce the noise in the image while preserving the edge of the image. The output image after filtering is locally a linear transformation of the guided image.

[0071] In this embodiment, for each category i among all n categories, the membership function value u of the category isi (x) Arrange the pixels in the order of their corresponding pixels into a membership function value image U composed of membership function values ​​with the same shape and size as the original MRI image i , and the membership function value image U i As the guidance image G of category i i , based on guided filtering using the guidance image G i For the membership function value image U i When the guide image G i The filtered image is the membership function value image U i At the same time, guided filtering works as an edge-preserving smoothing operator.

[0072] For the membership function value image U i , assuming that in the m×m window ω centered at pixel k ik The filtered membership function value v of any pixel x in the window is i (x) are all related to the guidance image G i The pixel value G of pixel x i (x)(equal to the membership function value image U i The pixel value u of pixel x i (x)) is linearly related, then:

[0073]

[0074] where a ik and b ik It is a window ik The guided filter coefficients, in the window ω ik Where is a constant.

[0075] Here, in order to find the condition that v is kept under the premise of maintaining the above linear relationship i (x) and u i (x) to minimize the difference between the guiding filter coefficient a ik and b ik , we need to find the guided filter coefficient a that minimizes the cost function F ik and b ik The cost function F and v of all pixels i (x) and u i The sum of the distances of (x) is positively correlated, defined by formula (10):

[0076]

[0077] Among them, ∈ is to avoid the guided filter coefficient a ik The regularization term coefficient is set too large.

[0078] In step S105 , the filter unit 130 calculates, for each pixel, a first guided filter coefficient and a second guided filter coefficient for guided filtering using the membership function value of the pixel and the membership function values ​​of other pixels in the window in a window of a certain size centered on the pixel.

[0079] In this embodiment, the equations (11) and (12) are used to calculate the value of the pixel k in the window ω for each combination of category i and pixel k. ik The first guided filter coefficient a that minimizes the cost function F ik and the second guided filter coefficient b ik .

[0080]

[0081]

[0082] where |ω ik | indicates ω ik The number of pixels in ik and σ 2 ik Respectively expressed in ω ik The pixel value G of the guide image within i (x) is the mean and variance of the value. Represents ω ik The membership function value u of category i in i The average value of (x).

[0083] After obtaining all the guided filter coefficients a for each pixel k ik and b ik After that, due to the membership function value image U i In the example, any pixel k not only exists in the window ω centered on itself, ik In the case of , there are also multiple windows ω centered around the pixels adjacent to it. ip ~ω iq (pixel p~q∈ω ik ), so in different windows, the membership function value v after filtering calculated by formula (9) i (k) is also different, and the membership function value v after filtering cannot be directly calculated by formula (9) i (k).

[0084] In step S109, the guided filter coefficients a within these windows are used ip ~a iq and b ip ~b iq The average and According to formula (13), the membership function value v after filtering is calculated i (k), thereby filtering the membership function value of each pixel according to the first guided filter coefficient and the second guided filter coefficient, and using the filtered result to update the membership function value.

[0085]

[0086] In step S110, the filter unit 130 performs filtering on the n filtered membership function values ​​v corresponding to each pixel k of the MRI image according to formula (14). i (k) Scaling is performed so that the sum of these values ​​becomes 1, and the membership function value stored in the storage unit 160 is updated using the scaled membership function value.

[0087]

[0088] In step S111, the determination unit 140 determines whether the cluster centers have converged as a whole based on the distance between the pixel values ​​of the cluster centers of each category calculated in step S110 and the pixel values ​​of the cluster centers stored previously. For example, when the mean square error between the pixel values ​​of the cluster centers of each category calculated this time and the pixel values ​​of the cluster centers of each category calculated last time is less than a preset threshold, it is considered that the cluster centers have converged. If it is determined that the cluster centers have converged, the process proceeds to step S112, otherwise, the process proceeds to step S102.

[0089] In step S112, the segmentation unit 150 obtains the membership function value of each pixel k of the MRI image for each category i, and for each pixel, takes the category to which the largest membership function value among the corresponding n membership function values ​​belongs as the category of the pixel, and performs segmentation based on the classification results of each pixel.

[0090] Through the filtering processing of this embodiment, the membership function value of each pixel for each classification is filtered through guided filtering within each iteration, so that an MRI image segmentation result that suppresses the influence of noise and retains fine tissue can be obtained without increasing excessive calculation amount, thereby ensuring execution speed.

[0091] (Implementation Method 2)

[0092] In this embodiment, the same reference numerals are given to the same or corresponding parts as those in Embodiment 1, and duplicate descriptions are omitted. The image processing method of this embodiment further includes steps S106 to S108, and step S109A instead of step S109.

[0093] Figure 4 This is a flowchart showing an image processing method used by the image processing apparatus of this embodiment.

[0094] The processing of steps S101 to S105 is the same as that of the first embodiment, and thus the description thereof is omitted.

[0095] Different from the first embodiment, in this embodiment, for the membership function value image U i , according to the information of pixel k and its neighboring pixels, for window ω ip ~ω iq The guided filter coefficients a in ip ~a iq and b ip ~b iq Calculate the weights and use the weighted sum of the filter coefficients a' ik and b' ik Calculate the membership function value v after filtering i (k).

[0096] In step S106 , the filter unit 130 calculates the coefficient of variation for each pixel based on the membership function values ​​of the pixels in the window centered on the pixel.

[0097] In this embodiment, each guide image G is calculated one by one by formula (15): i The coefficient of variation CV is calculated for each pixel in the image. The coefficient of variation CV is a value that reflects the degree of data dispersion.

[0098]

[0099] In formula (15), CV ik Represents the guidance image G of category i i The coefficient of variation of pixel k in reflects the discrete degree of the membership function value around pixel k.

[0100] In step S107 , the filter unit 130 calculates the weight of each pixel based on the coefficient of variation.

[0101] In this embodiment, the coefficient of variation CV calculated in step S106 is used to calculate the guide image G for each category i according to formula (16). i The weight w of each pixel in .

[0102]

[0103] In formula (16), w ik Represents the guidance image G of category i i The weight of pixel k in norm. i Represents the normalization function, which is obtained by iThe values ​​of all coefficients of variation CV are normalized to map their values ​​to the range of 0 to 1. According to formula (16), if the coefficient of variation CV of pixel k is ik The larger the value, the greater the discreteness between the neighboring pixels of pixel k, and the weight w ik The smaller the coefficient of variation CV is, the ik The larger the weight w is, ik The closer it is to 0; if the coefficient of variation CV of pixel k ik The smaller the pixel k is, the smaller the discreteness between the neighboring pixels is, and the weight w ik The larger the coefficient of variation (CV) is, the ik The smaller the weight w ik The closer it is to 1.

[0104] In step S108, the filtering unit 130 obtains the third guided filter coefficient for each pixel by summing the product of the weight of the pixel in a window centered on the pixel and the first guided filter coefficient and the product of the weights of other pixels and the first guided filter coefficient, and obtains the fourth guided filter coefficient by summing the product of the weight of the pixel in a window centered on the pixel and the second guided filter coefficient and the product of the weights of other pixels and the second guided filter coefficient.

[0105] In this embodiment, the membership function value image U is calculated for each category i one by one using formulas (17) and (18): i The third guided filter coefficient a' is the weighted sum of the pixels in ik and the fourth guided filter coefficient b' ik .

[0106]

[0107]

[0108] Next, in step S109A, similar to step S109 in the first embodiment, the filter unit 130 filters the membership function value according to the calculated guided filter coefficients, that is, filters the membership function value according to the third guided filter coefficients and the fourth guided filter coefficients.

[0109] At this time, the membership function value v of each pixel after filtering is calculated for each category i one by one through formula (19): i (k).

[0110] v i (k) = a′ ik G i (k)+b′ ik (19)

[0111] The processing after step 109A is the same as that in embodiment 1, and thus the description thereof is omitted.

[0112] In implementation mode 2, since the membership function value is filtered using the weighted summation filter coefficient according to the information of adjacent pixels when calculating the filter coefficient used to filter the membership function value of each pixel, compared with implementation mode 1, the MRI image segmentation result that further suppresses the influence of noise and retains fine tissue can be obtained.

[0113] The following Figures 5 to 8 The effects of the first and second embodiments will be described.

[0114] Figure 5 : is an effect diagram showing the effect of the membership function value filtering process of the image processing method of embodiment 2. Figure 5 In the figure, (a), (b), and (c) represent the membership function value images of cerebrospinal fluid, gray matter, and white matter before filtering, respectively, and (d), (e), and (f) represent the membership function value images of cerebrospinal fluid, gray matter, and white matter after filtering, respectively. The closer the color of a pixel is to white, the closer its membership function value is to 1. It can be seen that the membership function value filtering processing of the image processing method of implementation mode 2 can effectively reduce the noise contained in the membership function value while retaining fine tissues, making the membership function value more accurate.

[0115] Figure 6 1 is a diagram comparing the effect of the membership function value filtering process of the image processing method of the second embodiment and the effect of the filtering process of the first embodiment. Figure 6 In the figure, (a) and (b) respectively represent the gray matter membership function value image processed by the membership function value filtering of the image processing method of embodiment 2 and the gray matter membership function value image processed by the membership function value filtering of embodiment 1, and (c) and (d) respectively represent the segmentation result of the image processing method of embodiment 2 and the segmentation result of the image processing method of embodiment 1. It can be seen that the membership function value filtering of the image processing method of embodiment 2 can further reduce the noise contained in the membership function value and retain the fine tissue compared with the membership function value filtering of embodiment 1, so that a more accurate segmentation result can be obtained.

[0116] Figure 7 1 is a diagram comparing the segmentation results of the image processing method of embodiment 2, the image processing method of embodiment 1, and the image processing method of the prior art. Figure 7 In the figure, (a) represents the original MRI image, (b), (c), and (d) represent the segmentation results of the image processing method of embodiment 1, the image processing method of the prior art, and the image processing method of embodiment 2, respectively. Figure 8This is a diagram showing the effect of the image processing method of embodiment 2.

[0117] pass Figure 7 and Figure 8 It can be seen that due to the influence of noise, there are a lot of misclassifications in the MRI image segmentation results in the prior art. The MRI image segmentation results of implementation mode 1 remove the influence of noise and improve the misclassification of fine tissues. In the MRI image segmentation results of implementation mode 2, based on the technical effect of implementation mode 1, the misclassification caused by noise is more effectively suppressed and the fine tissues are retained.

[0118] Through implementation mode 2 of the present invention, within each iteration (executing steps S102 to S110 once is regarded as one iteration), the membership function value of each pixel for each classification is filtered by guided filtering, so that, similar to implementation mode 1, an MRI image segmentation result that suppresses the influence of noise and retains fine tissues can be obtained.

[0119] In addition, in implementation mode 2, since the membership function value is filtered using the weighted summation filter coefficient according to the information of adjacent pixels when calculating the filter coefficient used to filter the membership function value of each pixel, compared with implementation mode 1, the MRI image segmentation result that further suppresses the influence of noise and retains fine tissue can be obtained.

[0120] (Implementation 3)

[0121] Embodiment 3 of the present invention relates to an image processing device and an image processing method. In the description of Embodiment 3, the same reference numerals are given to the same or corresponding parts as those in Embodiment 1 or 2, and duplicate descriptions are omitted.

[0122] Fig. 9 This is a block diagram showing the structure of an image processing device according to a third embodiment.

[0123] The MRI segmentation apparatus 100B according to the third embodiment is different from the MRI segmentation apparatus 100 according to the second embodiment in that the updating unit 120B does not include the deviation field updating unit 123 .

[0124] Fig.10 This is a flowchart showing an image processing method used by the image processing device according to the third embodiment.

[0125] The image processing method of embodiment 3 does not have step S103 compared to embodiment 2. In the MRI segmentation method of embodiment 3, when the true value of the deviation field is known, for example, when the true value of the deviation field is stored in the storage unit 160, the membership function value is calculated using the true value of the deviation field pre-stored in the storage unit 160, and therefore there is no need to update the deviation field.

[0126] Through implementation mode 3 of the present invention, the membership function value of each pixel for each classification is filtered through guided filtering within each iteration, so that an MRI image segmentation result that suppresses the influence of noise and retains fine tissue can be obtained. Compared with implementation mode 1, since step S103 is removed, the calculation required for the algorithm can be reduced and the execution speed can be accelerated.

[0127] (Implementation 4)

[0128] Embodiment 4 of the present invention relates to an image processing device and an image processing method. The image processing method of this embodiment is a method for correcting MRI image intensity non-uniformity based on segmentation.

[0129] The image processing method involved in embodiment 4 includes: a first image acquisition step, acquiring a first image; a deviation field acquisition step, acquiring a deviation field representing intensity unevenness in the first image; a segmentation step, segmenting the first image; and an output step, outputting a second image corrected for intensity unevenness based on the deviation field and a result obtained by segmenting the first image, wherein the segmentation step includes edge-preserving smoothing processing.

[0130] In the description of the fourth embodiment, the same reference numerals are used for the same or corresponding parts as those in the first to third embodiments, and duplicate descriptions are omitted.

[0131] Fig.11 This is a block diagram showing the structure of an image processing device according to a fourth embodiment.

[0132] The image processing device 200 according to the fourth embodiment is different from the image processing device 100 according to the first or second embodiment in that the image processing device 200 does not include the dividing unit 150 but includes an intensity unevenness correcting unit 170 .

[0133] The intensity unevenness correction unit 170 performs intensity unevenness correction on each pixel of the MRI image based on the calculated deviation field for each pixel.

[0134] Fig.12 This is a flowchart showing an image processing method used by the image processing device according to the fourth embodiment.

[0135] The image processing method of the fourth embodiment is different from the image processing method of the second embodiment in that steps S111 and S112 are replaced with steps S111D and S112D.

[0136] In step S111D, the determination unit 140 determines whether the deviation field corresponding to each pixel of the MRI image has converged as a whole based on the difference between the value of each deviation field calculated in step S103 and the value of the previous deviation field. For example, it is considered to have converged when the mean square error of the deviation field is less than a preset threshold. If it is determined to have converged, the process proceeds to step S112D, otherwise it proceeds to step S102.

[0137] In step S112D, the intensity non-uniformity correction unit 170 performs intensity non-uniformity correction on each pixel of the MRI image according to the calculated deviation field for each pixel, and divides the measured intensity of the pixel by the deviation field to obtain a value as the actual intensity value of the pixel. After all pixels are corrected, the process ends.

[0138] Fig.13 It is an effect diagram showing the effect of the intensity unevenness correction method of implementation mode 4, wherein (a) represents the original MRI image, (b) and (c) respectively represent the intensity unevenness correction result of this implementation mode and the corrected MRI image, and (d) and (e) respectively represent the intensity unevenness correction result of the prior art and the corrected MRI image.

[0139] Through implementation mode 4 of the present invention, the membership function value of each pixel for each classification is filtered through guided filtering within each iteration (executing steps S102 to S110 once is regarded as one iteration), so that accurate membership function values ​​that suppress the influence of noise and retain fine tissues can be obtained, and accurate deviation fields can be calculated through accurate membership function values, so as to better correct intensity non-uniformity of MRI images.

[0140] In addition, the above-mentioned implementation modes can be summarized as the following technical solutions.

[0141] Technical Solution 1

[0142] An image processing method comprises: a first image acquisition step of acquiring a first image; a deviation field acquisition step of acquiring a deviation field representing intensity non-uniformity in the first image; a segmentation step of segmenting the first image; and an output step of outputting a second image corrected for intensity non-uniformity based on the deviation field and a result obtained by segmenting the first image, wherein the segmentation step comprises edge-preserving smoothing processing.

[0143] Technical Solution 2

[0144] According to the image processing method described in Technical Solution 1, the edge-preserving smoothing process is performed by using a guided filter corresponding to a guide image that is an image to be smoothed.

[0145] Technical Solution 3

[0146] According to the image processing method described in technical solution 2,

[0147] In the guided filtering, a guided filtering parameter is calculated according to the weight calculated for each pixel, and the guided filtering is performed based on the guided filtering parameter.

[0148] Technical Solution 4

[0149] According to the image processing method described in Technical Solution 1, in the segmentation step, the pixels contained in the first image are classified based on a membership function, and the membership function is used to classify the pixels into a plurality of predetermined categories, and the categories are set for each tissue of the subject contained in the first image.

[0150] Technical Solution 5

[0151] According to the image processing method described in Technical Solution 4, the membership function is set for each of the categories in fuzzy c-means clustering.

[0152] Technical Solution 6

[0153] According to the image processing method described in Technical Solution 2, in the segmentation step, the pixels contained in the first image are classified based on a membership function, and the membership function is used to classify the pixels into a plurality of predetermined categories. The image that is the object of the smoothing processing is an image generated based on the value output by the membership function for each of the categories.

[0154] Technical Solution 7

[0155] According to the image processing method described in Technical Solution 6, the guide image is the same as the image that is the object of the smoothing process.

[0156] Technical Solution 8

[0157] According to the image processing method described in Technical Solution 6, it also includes a step of updating the membership function based on the cluster centers set for the multiple categories and the deviation field.

[0158] Technical Solution 9

[0159] According to the image processing method described in Technical Solution 8, it also includes a step of updating the deviation field based on the values ​​output by the cluster center and the membership function.

[0160] Technical Solution 10

[0161] The image processing method according to Technical Solution 8 also includes a step of determining the convergence of the cluster center based on the amount of change of the cluster center due to the deviation field and the updated membership function.

[0162] Technical Solution 11

[0163] According to the image processing method described in Technical Solution 10, the segmentation step is performed after it is determined that the cluster centers have converged.

[0164] Technical Solution 12

[0165] An image processing device has a processing circuit, which executes: the steps of obtaining a first image and a deviation field representing intensity non-uniformity in the first image; the step of segmenting the first image; and the step of outputting a second image in which the intensity non-uniformity is corrected and a result of segmenting the first image based on the deviation field, wherein the processing circuit segments the first image using edge-preserving smoothing processing.

[0166] Technical Solution 13

[0167] An image processing method comprises: a first image acquisition step for acquiring a first image; a deviation field acquisition step for acquiring a deviation field representing intensity non-uniformity in the first image; and a segmentation step for segmenting the first image based on the deviation field, wherein the segmentation step comprises edge-preserving smoothing processing.

[0168] In the above-described embodiment, the intensity inhomogeneity correction process is described for MRI images, but the image to be corrected is not limited to the MRI image. That is, the intensity inhomogeneity correction according to the embodiment of the present invention can also be applied to medical images output from a medical image diagnosis apparatus other than an MRI apparatus when an image having the same intensity inhomogeneity is obtained.

Claims

1. An image processing method, comprising: A first image acquisition step, acquiring a first image; a deviation field obtaining step, obtaining a deviation field representing intensity non-uniformity in the first image; A membership function value obtaining step, obtaining membership function values ​​of pixels included in the first image for a plurality of preset categories based on the deviation field; a guided filtering step, for a membership function value image composed of the membership function values, calculating a weight for each pixel based on information of the pixel and its neighboring pixels, calculating a guided filtering coefficient after weighted summation of each pixel according to the weight calculated for each pixel, filtering the membership function values ​​using the calculated guided filtering coefficient, thereby performing guided filtering using a guide image; a segmentation step of classifying pixels included in the first image based on the membership function value subjected to guided filtering, and segmenting the first image; as well as an output step of outputting a second image corrected for intensity inhomogeneity based on the deviation field and a result obtained by segmenting the first image, The guide image and the membership function value image are the same image.

2. The image processing method according to claim 1, In the guided filtering step, the weight for each pixel is calculated by the following equations (a) and (b): in, i represents the index of the category, k represents the index of the pixel contained in the first image, μ ik and σ 2 ik represents the mean and variance of the pixel values ​​of the guide image of category i in the window centered on pixel k, CV ik represents the coefficient of variation of pixel k in the guidance image of category i, w ik represents the weight of pixel k in the guidance image of category i.

3. The image processing method according to claim 1, The membership function value is set for each of the categories in fuzzy c-means clustering.

4. The image processing method according to claim 1, The method also includes the step of updating the membership function value based on the cluster centers set for the multiple categories and the deviation field.

5. The image processing method according to claim 4, The method also includes the step of updating the deviation field based on the cluster center and the membership function value.

6. The image processing method according to claim 4, The method further includes a step of determining the convergence of the cluster center based on the amount of change of the cluster center due to the deviation field and the updated membership function value.

7. The image processing method according to claim 6, The segmentation step is performed after it is determined that the cluster centers have converged.

8. An image processing device, A processing circuit is provided, the processing circuit performing the following steps: Obtain a first image; obtaining a deviation field representing intensity inhomogeneities in the first image; Obtaining membership function values ​​of pixels included in the first image for a plurality of preset categories based on the deviation field; For a membership function value image composed of the membership function values, for each pixel, a weight is calculated based on information of the pixel and its neighboring pixels, a guided filtering coefficient after weighted summation of each pixel is calculated according to the weight calculated for each pixel, and the membership function value is filtered using the calculated guided filtering coefficient, thereby performing guided filtering using a guide image; Based on the membership function value after guided filtering, pixels included in the first image are classified to segment the first image; as well as outputting a second image corrected for intensity inhomogeneity based on the deviation field and a result obtained by segmenting the first image, The guide image and the membership function value image are the same image.

Images