Image processing device and image processing method

Abstract

To accurately correct the intensity non-uniformity of an image.SOLUTION: An image processing device according to an embodiment comprises: an acquisition part; an updating part; a filter part; and a division part. The acquisition part acquires a first image. The updating part updates a parameter used in division processing of the first image so as to correct the intensity non-uniformity in the first image. The filter part removes noise components included in the parameter by performing processing of edge preservation smoothing. The division part divides the first image on the basis of the parameter from which the noise components are removed.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The embodiments disclosed in this specification and the drawings relate to an image processing device and an image processing method. [Background technology]

[0002] Images generated by magnetic resonance imaging (MRI) technology have a problem of intensity inhomogeneity, and the pixel values ​​of an MRI image affected by intensity inhomogeneity are expressed by the following equation (1).

number

[0003] In equation (1), I is the pixel value of the MRI image affected by the intensity inhomogeneity, B is the bias field indicating the intensity inhomogeneity, J is the pixel value when not affected by the intensity inhomogeneity, and n is the noise. Note that J will be simply called the "ideal pixel value" below.

[0004] Since intensity inhomogeneity in MRI images affects the accuracy of image diagnosis, it is desirable to correct it. As one of the techniques for correcting intensity inhomogeneity in MRI images to obtain an image having pixel values ​​J that are not affected by intensity inhomogeneity, a segmentation-based intensity inhomogeneity correction method is used. Since a more accurate segmentation result can obtain a more accurate intensity inhomogeneity estimation result for an MRI image, and a more accurate intensity inhomogeneity estimation result can also obtain a more accurate segmentation result, the segmentation-based intensity inhomogeneity correction method repeatedly updates the segmentation result and the intensity inhomogeneity estimation result to gradually obtain more accurate intensity inhomogeneity estimation results and segmentation results.

[0005] However, the segmentation-based intensity non-uniformity correction method is easily affected by noise, which affects the accuracy of segmentation and intensity non-uniformity correction. In contrast, Non-Patent Document 1 discloses a segmentation-based intensity non-uniformity correction method, which uses spatial information of MRI images to update segmentation results and non-uniformity estimation results, thereby reducing the influence of noise. Patent Document 1 also discloses a segmentation-based intensity non-uniformity correction method, which uses information from adjacent pixels to reduce the influence of noise. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] China Patent Publication No. 103996193 [Non-patent literature]

[0007] [Non-Patent Document 1] Wenchao Cui,Yi Wang,Yangyu Fan,Yan Feng,and Tao Lei.2013.Localized FCM clustering with spatial information for medical image segmentation and bias field estimation.Journal of Biomedical Imaging 2013,Article 13(January 2013) Summary of the Invention [Problem to be solved by the invention]

[0008] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to accurately correct intensity non-uniformity in an image. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0009] An image processing device according to an embodiment includes an acquisition unit, an update unit, a filter unit, and a division unit. The acquisition unit acquires a first image. The update unit updates parameters used in a division process of the first image so that intensity non-uniformity in the first image is corrected. The filter unit removes noise components contained in the parameters by performing edge-preserving smoothing processing. The division unit divides the first image based on the parameters from which the noise components have been removed. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram showing an example of image segmentation of a brain MRI image. [Diagram 2] FIG. 2 is a block diagram showing the configuration of the image processing apparatus according to the first embodiment. [Diagram 3] FIG. 3 is a flowchart showing an image processing method applied to the image processing device according to the first embodiment. [Figure 4] FIG. 4 is a flowchart showing an image processing method applied to the image processing device according to the second embodiment. [Diagram 5] FIG. 5 is a diagram showing the effect of the filtering process on membership function values ​​in the image processing method according to the second embodiment. [Figure 6] FIG. 6 is a diagram comparing the effect of the filtering process on membership function values ​​in the image processing method according to the second embodiment with the effect of the filtering process according to the first embodiment. [Figure 7] FIG. 7 is a diagram comparing division results of the image processing method according to the second embodiment, the image processing method according to the first embodiment, and an image processing method according to the conventional technology. [Figure 8] FIG. 8 is a diagram showing the effect of the image processing method according to the second embodiment. [Figure 9] FIG. 9 is a block diagram showing the configuration of an image processing apparatus according to the third embodiment. [Figure 10] FIG. 10 is a flowchart showing an image processing method applied to the image processing device according to the third embodiment. [Figure 11] FIG. 11 is a block diagram showing the configuration of an image processing apparatus according to the fourth embodiment. [Figure 12] FIG. 12 is a flowchart showing an image processing method applied to the image processing device according to the fourth embodiment. [Figure 13] FIG. 13 is a diagram showing the effect of the intensity non-uniformity correcting method according to the fourth embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, embodiments of the technology disclosed in the present application will be described with reference to the drawings.

[0012] (First embodiment) The first embodiment relates to an image processing device and an image processing method. In the image processing method according to this embodiment, each pixel of an MRI image is classified by fuzzy c-means clustering, that is, the MRI image is divided by specifying a certain type of tissue class for each pixel. In any MRI image, if there is no influence of intensity non-uniformity, the ideal pixel values ​​of pixels belonging to the same class are considered to be the same. In other words, pixels belonging to the same tissue are considered to have the same pixel value. In this embodiment, a total of n classes (n is a natural number of 2 or more) are provided, and the pixel value of the cluster center of class i is calculated by c i (i is a natural number less than or equal to n).

[0013] The principle of image processing based on fuzzy c-means clustering will be described below with reference to the drawings.

[0014] FIG. 1 is a diagram showing an example of image segmentation of a brain MRI image.

[0015] In Figure 1, (a) shows a brain MRI image undergoing segmentation, (b) shows an image of the segmented cerebrospinal fluid, in which the white parts represent the cerebrospinal fluid, (c) shows an image of the segmented gray matter, in which the white parts represent the gray matter, and (d) shows an image of the segmented white matter, in which the white parts represent the white matter.

[0016] In fuzzy c-means clustering, a membership function is set for each of multiple classes. i denotes the membership function for class i. i (x) indicates the value output by the membership function of pixel x for class i (hereinafter, the membership function value), and means the probability that pixel x belongs to class i. For each pixel, the class corresponding to the maximum membership function value among the membership function values ​​for each class is determined as the class of that pixel.

[0017] For any pixel x, its membership function value must satisfy equation (2) below.

number

[0018] Equation (2) shows that for any pixel x, the sum of the membership function values ​​of all n classes corresponding to it is one.

[0019] In fuzzy c-means clustering, the pixel value c of the cluster center of each class i is i , the bias field B(x) of pixel x, and the membership function value u of pixel x for class i. i By continuously updating (x), the value of the energy function E, which indicates the distance between the current classification of each pixel in the image and the optimized classification, gradually decreases, and the membership function value u of each pixel x for class i at the time when the value of the energy function E is minimized is i An optimized classification for each pixel x is identified by obtaining (x), where the energy function E is defined as follows:

number

[0020] In equation (3), Ω denotes the domain of pixels, I(x) denotes the pixel value of pixel x affected by intensity non-uniformity, K denotes the weighting function, and q is a natural number coefficient greater than 1 that determines the fuzzy degree of fuzzy c-means clustering.

[0021] Hereinafter, an image processing apparatus and an image processing method according to the present embodiment will be described with reference to the drawings.

[0022] FIG. 2 is a block diagram showing the configuration of the image processing apparatus according to the first embodiment.

[0023] The image processing device 100 according to the first embodiment 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.

[0024] The initialization unit 110 initializes each parameter used in image processing, and stores each initialized parameter in the storage unit 160. Here, each parameter is any parameter used in segmentation processing of an MRI image, such as a pixel value of each cluster center, a membership function value of each pixel for each cluster center, a bias field representing intensity non-uniformity in an MRI image, etc.

[0025] The update unit 120 updates each parameter stored in the storage unit 160 , and includes a cluster center update unit 121 , a membership function update unit 122 , and a bias field update unit 123 .

[0026] The filter section 130 filters the membership function values ​​updated by the update section 120 to remove noise components contained in the membership function values.

[0027] The determining section 140 determines whether or not the filtered membership function values ​​have converged.

[0028] When the determining unit 140 determines that the membership function values ​​have converged, the dividing unit 150 divides the MRI image based on the membership function values.

[0029] The storage unit 160 stores various data used in the image processing method.

[0030] Hereinafter, the flow of the image processing method used in the image processing device according to the present embodiment will be described with reference to the drawings.

[0031] FIG. 3 is a flowchart showing an image processing method applied to the image processing device according to the first embodiment.

[0032] After receiving a command to start image processing, the image processing device 100 acquires an MRI image to be divided and starts image processing. The MRI image acquired first is an MRI image having intensity non-uniformity. The image processing device 100 includes an acquisition unit (not shown) that acquires the MRI image.

[0033] In step S101, the initialization unit 110 acquires image information such as dimensional information and measurement intensity of the MRI image, and initializes each parameter based on the image information and the initial value or initialization condition set for each parameter used in this embodiment stored in the storage unit 160. For parameters for which an initial value is set, the initialization unit 110 initializes the parameter by assigning the initial value, and for parameters for which no initial value is set and initialization conditions are set, the initialization unit 110 initializes the parameter according to the initialization condition.

[0034] In step S102, the cluster center update unit 121 updates the cluster center of each class based on pixel information of the MRI image, membership function values ​​for preset classes of each pixel on the MRI image, and a bias field representing intensity non-uniformity of the MRI image.

[0035] In this embodiment, the cluster center update unit 121 updates the pixel value c of the cluster center of class i stored in the storage unit 160 by the following equations (4) and (5). i Update.

number

number

[0036] In formulas (4) and (5), * denotes a convolution operator, and β, σ, and ρ are constant parameters. In this embodiment, the truncated Gaussian kernel is the weight function K, and β=1, σ=1, ρ=3, and q=2, but these values ​​can be changed according to actual needs.

[0037] The cluster center update unit 121 calculates the current bias field B(x) and the membership function value u for class i by using equations (4) and (5). i Based on (x), the pixel value c of the cluster center of class i is i By updating the value of i is made to approach an ideal pixel value without the influence of intensity non-uniformity.

[0038] In step S103, the bias field update unit 123 updates the bias field of the MRI image based on the pixel information of the MRI image, the cluster center of each class, and the membership function value of each pixel on the MRI image for each preset class.

[0039] In this embodiment, the bias field update unit 123 updates the bias field B(x) stored in the storage unit 160 by the following equation (6).

number

[0040] The bias field update unit 123 calculates the membership function value u for each class i of the current pixel x by using the formula (6). i (x), and the pixel value c of the cluster center of class i i By updating the bias field B(x) based on the above, the bias field B(x) approaches a bias field that accurately represents the actual intensity non-uniformity.

[0041] In step S104, the membership function update unit 122 updates the membership function value for each pixel on the MRI image with respect to each preset class, based on the pixel information of the MRI image, the cluster center of each class, and the bias field of the MRI image.

[0042] In this embodiment, the membership function update unit 122 updates the membership function value u of each pixel x stored in the storage unit 160 by the following equation (7): i Update (x) for each class i.

number

[0043] The membership function update unit 122 updates the current bias field B(x) and the pixel value c of the cluster center of class i by using the formula (7). i Based on the membership function value u i By updating (x), the membership function value u i (x) to approach the membership function value of the more optimized classification.

[0044] In this embodiment, as described above, the bias field B(x) and the pixel value c i , and the membership function value u i After the update of (x) is completed, the filter unit 130 executes steps S105 to S110 to obtain the updated membership function value u i By filtering (x), the membership function value u i The noise components contained in (x) are removed.

[0045] The MRI image obtained by the MRI device contains noise due to the influence of the environment and the equipment. Therefore, the membership function value u for pixel x obtained based on the information of the MRI image i Since (x) also contains noise, the membership function value u is used to remove the noise component. i (x) needs to be filtered. The filtered membership function value is v i If (x), then v i (x) is expressed by the following equation (8).

number

[0046] In formula (8), n i (x) is the membership function value u of pixel x for class i i This indicates the noise component contained in (x).

[0047] In steps S105 to S110 of the image processing method according to this embodiment, the filter unit 130 performs edge-preserving smoothing. In this embodiment, the filter unit 130 calculates a membership function value u i By filtering (x), the filtered membership function value v i (x) is obtained. GIF is an edge-preserving image filtering technique, which is usually applied to the process of edge-preserving smoothing of grayscale or RGB images. It uses a guide image to filter the original image to reduce noise in the image under the premise of preserving the edges of the image, and the filtered output image is locally a linear transformation of the guide image.

[0048] In this embodiment, the filter unit 130 calculates the membership function value u iBy arranging (x) in the order of corresponding pixels, a membership function value image U consisting of membership function values, whose shape and size are the same as those of the original MRI image, is generated. i and the membership function value image U i Class i guide image G i The guide image G i Using the membership function value image U by GIF i Filter the guide image G i and 、 Membership function value image U corresponding to the image to be filtered i If and are identical, GIF acts as an edge-preserving smoothing operator.

[0049] Membership function value image U i For a window ω of size m × m centered on pixel k, ik In the window, the filtered membership function value v of any pixel x is i (x) is the guide image G i Pixel value of pixel x in G i (x) (membership function value image U i The membership function value u i If we assume that the relationship is linear with (x), then it can be expressed as follows:

number

[0050] In formula (9), a ik and b ik is the window ω ik is the GIF coefficient of the window ω ik It is a constant in

[0051] Here, assuming that the above linear relationship is maintained, v i (x) and u i GIF coefficient a that minimizes the difference with (x) ik and b ik To find the GIF coefficient a that minimizes the cost function F,ik and b ik The cost function F is the sum of all pixels v i (x) and u i It is proportional to the sum of the distances (x) and is defined by the following equation (10).

number

[0052] In formula (10), a ik 2 ∈ is the GIF coefficient a ik is the regularization term coefficient that is set to prevent from being too large.

[0053] In step S105, the filter unit 130 calculates, for each pixel, a first GIF coefficient and a second GIF coefficient for GIF in a window of a certain size centered on the pixel, based on the membership function value of the pixel and the membership function values ​​of other pixels in the window.

[0054] In this embodiment, the filter unit 130 calculates a window ω ik The first GIF coefficient a that minimizes the cost function F in ik and the second GIF coefficient b ik Request.

number

number

[0055] In equations (11) and (12), |ω ik | is ω ik indicates the number of pixels in μ ik and σ 2 ik are ω ik Pixel value of the guide image in G i The mean and variance of (x) are shown. ik(￣) is ω ik The membership function value u of class i in i The average value of (x) is shown.

[0056] GIF coefficient a for each pixel k ik and b ik After obtaining the membership function value image U i In the above, any pixel k is defined by a window ω ik In addition to the pixels in ip ~ω iq also exists (pixel p~q∈ω ik ), the filtered membership function values ​​v calculated by Eq. (9) in different windows are i (k) is also different, and the filtered membership function value v i (k) cannot be calculated directly using equation (9).

[0057] In step S109, the filter unit 130 filters the GIF coefficients a ip ~a iq and b ip ~b iq The average value of a ik (with ￣) and b ik Using (with ￣), the filtered membership function value v i By calculating (k) using the following equation (13), the membership function value of each pixel is filtered based on the first GIF coefficient and the second GIF coefficient, and the membership function value is updated using the filtering result.

number

[0058] In step S110, the filter unit 130 calculates n filtered membership function values ​​v corresponding to each pixel k of the MRI image according to the following equation (14): iBy scaling (k), their sum is made 1, and the membership function values ​​stored in storage unit 160 are updated using the scaled membership function values.

number

[0059] In step S111, the determination unit 140 determines whether or not the cluster centers have converged overall based on the difference between the pixel values ​​of the cluster centers of each class calculated in step S102 and the pixel values ​​of the cluster centers stored previously. For example, the determination unit 140 determines that the cluster centers have converged when the mean square error between the pixel values ​​of the cluster centers of each class calculated this time and the pixel values ​​of the cluster centers of each class calculated last time is smaller than a preset threshold. If it is determined that the cluster centers have converged, the process proceeds to step S112, and if it is not determined that the cluster centers have converged, the process proceeds to step S102.

[0060] In step S112, the division unit 150 obtains a membership function value for each class i of each pixel k of the MRI image, classifies each pixel into a class to which the maximum membership function value of the n membership function values ​​corresponding to the pixel belongs, and performs division based on the classification result of each pixel. Here, the division unit 150 divides the MRI image by classifying each pixel included in the MRI image based on a membership function for classifying pixels into a plurality of classes predetermined for each tissue of the subject.

[0061] According to the filtering process of the first embodiment, by filtering the membership function values ​​for each classification of each pixel by GIF within each iteration (one execution of steps S102 to S110 is considered as one iteration), it is possible to obtain image processing results that suppress the effects of noise and retain fine structures, and it is possible to guarantee the execution speed without significantly increasing the amount of calculation.

[0062] Second Embodiment The second embodiment relates to an image processing device and an image processing method. In the second embodiment, the same or corresponding parts as those in the first embodiment are denoted by the same reference numerals, and duplicated explanations are omitted. The image processing method according to this embodiment further includes steps S106 to S108, and includes step S109A instead of step S109.

[0063] FIG. 4 is a flowchart showing an image processing method applied to the image processing device according to the second embodiment.

[0064] The processes in steps S101 to S105 are similar to those in the first embodiment, and therefore will not be described here.

[0065] The present embodiment differs from the first embodiment in that the filter unit 130 filters the membership function value image U i Based on the information of pixel k and its neighboring pixels, ip ~ω iq Each GIF coefficient a ip ~a iq and b ip ~b iq Calculate the weights of and calculate the weighted sum filter coefficient a' ik and b' ik The filtered membership function value v i The aim is to calculate (k).

[0066] In step S106, the filter unit 130 calculates a coefficient of variation for each pixel based on the membership function value of the pixel in a window centered on that pixel.

[0067] In this embodiment, the filter unit 130 calculates the guide image G i A coefficient of variation CV is calculated for each pixel in , where CV is a value that reflects the degree of variance of the data.

number

[0068] In formula (15), CV ik is the guide image G of class i i This indicates the coefficient of variation of pixel k in and represents the degree of dispersion of the membership function values ​​around pixel k.

[0069] In step S107, the filter unit 130 calculates the weight of each pixel based on the variation coefficient.

[0070] In this embodiment, the filter unit 130 uses the coefficient of variation CV calculated in step S106 to filter the guide image G for each class i according to the following formula (16): i Calculate the weight w of each pixel in

number

[0071] In formula (16), w ik is the guide image G of class i i indicates the weight of pixel k in norm i denotes the normalization function, and the guide image G i By normalizing all the values ​​of the coefficient of variation CV, the values ​​are mapped to a range between 0 and 1. ik The larger the distribution of pixels adjacent to pixel k, the larger the weight w ik becomes smaller, and the coefficient of variation CV ik When is increased, the weight w ik approaches 0, while the coefficient of variation CV ik The smaller the distribution of the weights w ik becomes larger, and the coefficient of variation CV ik When is small, the weight w ik approaches 1.

[0072] In step S108, filter unit 130 obtains, for each pixel, a third GIF coefficient for the pixel by summing the product of the pixel's weight and the first GIF coefficient in a window centered on the pixel and the products of the weights of other pixels and the first GIF coefficient, and obtains a fourth GIF coefficient by summing the product of the pixel's weight and the second GIF coefficient in a window centered on the pixel and the products of the weights of other pixels and the second GIF coefficient.

[0073] In this embodiment, the filter unit 130 calculates the membership function value image U i The third GIF coefficient a' is the weighted sum of each pixel in ik and the fourth GIF coefficient b' ik is calculated for each class i.

number

number

[0074] Next, in step S109A, similar to step S109 in the first embodiment above, the filter unit 130 filters the membership function values ​​based on the calculated GIF coefficients, i.e., filters the membership function values ​​based on the third GIF coefficient and the fourth GIF coefficient.

[0075] At this time, the filter unit 130 calculates the filtered membership function value v of each pixel by the following equation (19): i (k) is calculated for each class i.

number

[0076] The processing from step 109A onwards is similar to that in the first embodiment, and therefore a description thereof will be omitted here.

[0077] In the second embodiment, when calculating a filter coefficient for filtering the membership function value of each pixel, filtering is performed using a filter coefficient that is a weighted sum of the membership function values ​​based on information about neighboring pixels. Therefore, compared to the first embodiment, it is possible to obtain an image processing result that further suppresses the effects of noise and retains fine texture.

[0078] The effects of the first and second embodiments will be described below with reference to FIGS.

[0079] FIG. 5 is a diagram showing the effect of the filtering process on membership function values ​​in the image processing method according to the second embodiment.

[0080] 5, (a), (b), and (c) respectively show membership function value images of cerebrospinal fluid, gray matter, and white matter before filtering, and (d), (e), and (f) show membership function value images of cerebrospinal fluid, gray matter, and white matter after filtering, indicating that the closer a pixel's color is to white, the closer its membership function value is to 1. It can be seen that the filtering process of the membership function values ​​in the image processing method according to the second embodiment effectively reduces noise contained in the membership function values, provided that fine tissues are left intact, and the membership function values ​​become more accurate.

[0081] FIG. 6 is a diagram comparing the effect of the filtering process on membership function values ​​in the image processing method according to the second embodiment with the effect of the filtering process according to the first embodiment.

[0082] 6, (a) and (b) respectively show a gray matter membership function value image obtained by filtering the membership function values ​​of the image processing method according to the second embodiment and a gray matter membership function value image obtained by filtering the membership function values ​​of the image processing method according to the first embodiment, and (c) and (d) respectively show the segmentation results of the image processing method according to the second embodiment and the segmentation results of the image processing method according to the first embodiment. It can be seen that the filtering process of the membership function values ​​of the image processing method according to the second embodiment can preserve fine tissue while further reducing noise contained in the membership function values, compared to the filtering process of the membership function values ​​of the first embodiment, and can obtain more accurate segmentation results.

[0083] FIG. 7 is a diagram comparing division results of the image processing method according to the second embodiment, the image processing method according to the first embodiment, and an image processing method according to the conventional technology.

[0084] In Fig. 7, (a) shows an MRI image of the original image, and (b), (c), and (d) are diagrams showing the division results of the image processing method according to the first embodiment, the image processing method according to the conventional technology, and the image processing method according to the second embodiment, respectively. Fig. 8 is a diagram showing the effect of the image processing method according to the second embodiment.

[0085] 7 and 8, the image processing result in the conventional technology has a large amount of misclassification due to the influence of noise, whereas the image processing result in the first embodiment removes the influence of noise and improves the misclassification of fine structures. The image processing result in the second embodiment can retain fine structures while more effectively suppressing misclassification due to noise based on the technical effect of the first embodiment.

[0086] According to the second embodiment, by filtering the membership function values ​​for each classification of each pixel by GIF within each iteration (one execution of steps S102 to S110 is considered as one iteration), it is possible to obtain image processing results that suppress the effects of noise and retain fine structures, as in the first embodiment.

[0087] Furthermore, according to the second embodiment, when calculating a filter coefficient for filtering the membership function value of each pixel, filtering is performed using a filter coefficient that is a weighted sum of the membership function values ​​based on information about neighboring pixels. Therefore, compared to the first embodiment, it is possible to obtain an image processing result that retains fine texture while further suppressing the effects of noise.

[0088] (Third embodiment) The third embodiment relates to an image processing device and an image processing method. In the third embodiment, the same or corresponding parts to those in the first or second embodiment are denoted by the same reference numerals, and duplicated explanations will be omitted.

[0089] FIG. 9 is a block diagram showing the configuration of an image processing apparatus according to the third embodiment.

[0090] The image processing device 100B according to the third embodiment is different from the image processing device 100 according to the first or second embodiment in that an updating section 120B does not have a bias field updating section 123.

[0091] FIG. 10 is a flowchart showing an image processing method applied to the image processing device according to the third embodiment.

[0092] The image processing method according to the third embodiment does not include step S103, as compared with the image processing method according to the first or second embodiment. In the MRI segmentation method according to the third embodiment, when the true value of the bias field is known, for example, when the true value of the bias field is stored in the storage unit 160, the membership function value is calculated using the true value of the bias field stored in advance in the storage unit 160, so there is no need to update the bias field.

[0093] According to the third embodiment, by filtering the membership function values ​​for each classification of each pixel by GIF within each iteration (one execution of steps S102 to S110 is considered as one iteration), it is possible to obtain image processing results that suppress the effects of noise and retain fine structures, and since step S103 has been eliminated compared to the first embodiment, the number of calculations required for the algorithm can be reduced and the execution speed can be increased.

[0094] (Fourth embodiment) The fourth embodiment relates to an image processing device and an image processing method. The image processing method according to this embodiment is a segmentation-based MRI image intensity non-uniformity correction method. In the fourth embodiment, the same reference numerals are used to designate the same or corresponding parts as those in the first to third embodiments, and duplicated explanations are omitted.

[0095] FIG. 11 is a block diagram showing the configuration of an image processing apparatus according to the fourth embodiment.

[0096] The image processing device 200 according to the fourth embodiment differs from the image processing device 100 according to the first or second embodiment in that it does not have a division section 150 but has an intensity non-uniformity correction section 170.

[0097] An intensity non-uniformity corrector 170 corrects intensity non-uniformities in the MRI image based on the calculated bias field for each pixel.

[0098] FIG. 12 is a flowchart showing an image processing method applied to the image processing device according to the fourth embodiment.

[0099] The image processing method according to the fourth embodiment is different from the image processing method according to the first or second embodiment in that it includes steps S111D and S112D instead of steps S111 and S112.

[0100] In step S111D, the determination unit 140 determines whether the bias field corresponding to each pixel of the MRI image has converged overall based on the difference between the value of each bias field calculated in step S103 and the previous bias field value. For example, if the mean square error of the bias field is smaller than a preset threshold, it is considered to have converged. If it is determined to have converged, proceed to step S112D, and otherwise proceed to step S102.

[0101] In step S112D, the intensity non-uniformity correction unit 170 corrects the intensity non-uniformity for each pixel of the MRI image by dividing the measured intensity of the pixel by the bias field based on the calculated bias field for each pixel. When the correction of all pixels is completed, the process ends.

[0102] FIG. 13 is a diagram showing the effect of the intensity non-uniformity correcting method according to the fourth embodiment.

[0103] In Figure 13, (a) shows an MRI image of the original image, (b) and (c) show the intensity non-uniformity correction result according to this embodiment and the corrected MRI image, respectively, and (d) and (e) show the intensity non-uniformity correction result according to the conventional technology and the corrected MRI image, respectively.

[0104] According to the fourth embodiment, by filtering the membership function values ​​for each classification of each pixel by GIF within each iteration (one execution of steps S102 to S110 is considered as one iteration), it is possible to obtain accurate membership function values ​​that suppress the effects of noise and retain fine structures, and since an accurate bias field can be obtained from the accurate membership function values, it is possible to perform better intensity non-uniformity correction on MRI images.

[0105] The above-described embodiments can be summarized as the following technical solutions. (Technical proposal 1) 1. An image processing method comprising: a first image acquisition step of acquiring a first image; a bias field acquisition step of acquiring a bias 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 in which intensity non-uniformity has been corrected based on the bias field and the result of segmenting the first image, wherein the segmentation step includes an edge-preserving smoothing process. (Technical proposal 2) The image processing method described in Technical Solution 1, in which the edge-preserving smoothing process is performed by guided image filtering (GIF) using a guide image corresponding to the image to be smoothed. (Technical proposal 3) In the GIF, a GIF parameter is calculated based on the weight calculated for each pixel, and the GIF is performed based on the GIF parameter. An image processing method described in Technical Proposal 2. (Technical proposal 4) The image processing method described in Technical Proposal 1, wherein the division step includes classifying pixels included in the first image based on a membership function for classifying pixels into a plurality of predetermined classes, and the classes are provided for each tissue of the subject included in the first image. (Technical proposal 5) The image processing method described in Technical Proposal 4, wherein the membership function is set for each class in fuzzy c-means clustering. (Technical proposal 6) The image processing method described in Technical Proposal 2, wherein the division step includes classifying pixels contained in the first image based on a membership function for classifying pixels into a plurality of predetermined classes, and the image to be smoothed is an image generated based on values ​​output by the membership function for each of the classes. (Technical proposal 7) The image processing method described in Technical Proposal 6, wherein the guide image is identical to the image to be smoothed. (Technical proposal 8) The image processing method described in Technical Proposal 6 further includes a step of updating the membership function based on cluster centers provided for each of the multiple classes and the bias field. (Technical proposal 9) The image processing method according to technical proposal 8, further comprising a step of updating the bias field based on the cluster centers and values ​​output by the membership functions. (Technical proposal 10) The image processing method described in Technical Proposal 8 further includes a step of determining convergence of cluster centers based on the bias field and the amount of change in the cluster centers that changes due to the updated membership function. (Technical proposal 11) The image processing method described in Technical Proposal 10, wherein the step of dividing the first image is performed after it is determined that the cluster centers have converged. (Technical proposal 12) 1. An image processing apparatus comprising: a processing circuit for acquiring a first image and a bias field representative of intensity non-uniformity in the first image, segmenting the first image, and outputting a result of segmenting the first image and a second image in which intensity non-uniformity has been corrected based on the bias field, wherein the processing circuit segments the first image using an edge-preserving smoothing process. (Technical proposal 13) 1. An image processing method comprising: a first image acquisition step of acquiring a first image; a bias field acquisition step of acquiring a bias field representing intensity non-uniformity in the first image; and a segmentation step of segmenting the first image based on the bias field, wherein the segmentation step includes an edge-preserving smoothing process.

[0106] In the above-described embodiment, the process of correcting the intensity non-uniformity of an MRI image has been described, but the image to be corrected is not limited to an MRI image. In other words, when an image having similar intensity non-uniformity is acquired from a medical image output from a medical image diagnostic device other than an MRI device, the intensity non-uniformity correction according to the above-described embodiment may be applied.

[0107] According to at least one of the embodiments described above, intensity non-uniformity in an image can be corrected with high accuracy.

[0108] Although some embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations of embodiments can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope of the invention and its equivalents as described in the claims, as well as in the scope and spirit of the invention. [Explanation of symbols]

[0109] 100, 100B, 200 Image processing device 120,120B Update Department 130 Filter section 150 Division 140 Judgment section 170 Intensity non-uniformity correction section

Claims

1. An acquisition unit that acquires a first image; a bias field acquisition unit for acquiring a bias field representative of intensity non-uniformity in the first image; a membership function value acquisition unit that acquires membership function values ​​for a plurality of predetermined classes for pixels included in the first image based on the bias field; a filter unit that calculates a weight for each pixel based on information of the pixel and its neighboring pixels for a membership function value image made up of the membership function values, calculates a guided image filtering (GIF) coefficient that is a weighted sum of the pixels using the weight calculated for each pixel, and filters the membership function values ​​using the calculated GIF coefficient, thereby performing GIF using a guide image; a division unit that classifies pixels included in the first image based on the membership function value obtained by performing the GIF, and divides the first image; Equipped with The guide image is the same image as the membership function value image. Image processing device.

2. The dividing unit classifies each pixel of the first image by fuzzy c-means clustering in which the membership function value is set for each of the plurality of classes. The image processing device according to claim 1 .

3. The membership function value acquisition unit further updates the membership function value based on the bias field and the cluster center of each class.

3. The image processing device according to claim 1 or 2.

4. The bias field acquisition unit further updates the bias field based on the cluster centers and the membership function values. The image processing device according to claim 3 .

5. A determination unit that determines the convergence of the cluster center based on the bias field and the change amount of the cluster center that changes depending on the membership function value.

5. The image processing device according to claim 3 or 4.

6. The division unit divides the first image after the determination unit determines that the cluster centers have converged. The image processing device according to claim 5 .

7. acquiring a first image; obtaining a bias field representative of intensity inhomogeneities in the first image; obtaining membership function values ​​for a plurality of predetermined classes for pixels included in the first image based on the bias field; a step of calculating a weight for each pixel based on information of the pixel and its neighboring pixels for a membership function value image made up of the membership function values, calculating a guided image filtering (GIF) coefficient that is a weighted sum of the pixels using the weight calculated for each pixel, and filtering the membership function values ​​using the calculated GIF coefficient, thereby performing GIF using a guide image; classifying pixels included in the first image based on the membership function value obtained by performing the GIF, and dividing the first image; Including, The guide image is the same image as the membership function value image. Image processing methods.

Images