Single-modality medical image registration method, system and computer-readable storage medium
The Active-Demons model is improved through the adaptive fractional differential method, which solves the problem of accurate registration of blurred edges and unclear textures in medical image registration, and achieves efficient medical image registration, which is suitable for efficacy evaluation and pathological research.
Patent Information
- Application Number
- CN202111450099.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-01
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-12-01
AI Technical Summary
The existing Active-Demons model is difficult to effectively register images with blurred edges, unclear textures and insignificant features when processing medical images, and is prone to falling into local optimal values and difficult to achieve high levels in medical imaging-assisted clinical tumor treatment and efficacy evaluation. Accuracy results.
Adaptive fractional differential method is used to initialize the Active-Demons model, draw the amplitude relationship diagram of the order-to-order differential image signal at different frequencies, configure the adaptive fractional step function, calculate the deformation vector field of the floating image and Perform Gaussian filtering, perform bilinear interpolation operation until the preset accuracy or maximum iteration number is reached, and the registered image is obtained.
Quickly find fractional steps suitable for treating blurred edges, unclear textures and insignificant features. The output registration results are accurate and the alignment is accurate, and it is suitable for the efficacy evaluation and pathological research of medical imaging.
Smart Images

Figure CN114255265B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical image processing, and in particular to a registration method and system for single-modality medical images and a computer-readable storage medium. Background Art
[0002] Image registration is the process of spatially aligning two or more images of the same scene acquired directly by the same sensor or different sensors at different time periods and observation angles. The key lies in finding the spatial transformation between the registered images so that the corresponding points are in the same spatial coordinate system.
[0003] Image registration technology has now been applied to various industries, such as material mechanics, geographic information systems, and medical image processing. Depending on the specific application, some focus on the registration results, while others focus on studying transformations to obtain detailed information about the image. In the medical field, image registration can be used for auxiliary diagnosis, pathological analysis, intelligent diagnosis and treatment, etc.
[0004] Precision analysis technology of medical images is the basis and prerequisite for realizing intelligent medical diagnosis and treatment, and is a core "bottleneck" technology. Efficient and accurate medical image analysis technology can timely and accurately predict and identify patients' diseases, scientifically plan treatment plans, and quantitatively evaluate treatment effects.
[0005] Image registration methods can be divided into two categories: feature-based and grayscale-based registration methods. Feature-based methods can be further divided into two types: based on internal image features and external image features. External feature-based methods are generally based on spatial positioning. Before surgery, artificial markers that are easy to display images are placed on the surface of the organ to be registered. During surgery, tracking devices are used to track and locate them. The relative positions of the markers are used to perform registration analysis on the images before and during surgery. The method based on internal image features is to find the geometric transformation relationship between the registered images by determining the feature matching relationship between the reference image and the target image. The internal features of the image mainly include corners, contours, boundaries, surfaces, textures, etc.
[0006] The grayscale-based method is to optimize a similarity metric through iteration to find the geometric transformation between the registered images. Common similarity metrics include mean square distance, mutual information, etc. Such similarity metrics generally need to meet the following assumptions: the corresponding structures of the floating image and the reference image have the same grayscale value, so they are widely used in single-modal registration. Among them, the image registration method based on the Demons model is a classic grayscale-based registration method. The Demons model obtains the gradient information of the driving deformation from the reference image, and its improved method Active-Demons model assumes that the diffusion of the driving deformation is bidirectional, which not only generates a force that diffuses a deformed "floating" target to a corresponding "reference" target, but also generates a force that diffuses this "reference" target to the corresponding "floating" target. This force is called "positive force". The Active-Demons model introduces "positive force" on the basis of the Demons model, which makes it easier to obtain high-precision registration results. However, the Active-Demons model, which uses the gradient information of the reference image as the driving force for registration, has difficulty in effectively registering medical images with weak edges, weak textures, and insignificant features. This is mainly because it is easy to fall into the local optimal value when solving the functional extremum problem.
[0007] Nowadays, intelligent medical imaging-assisted clinical tumor treatment (e.g., microwave ablation, radiofrequency ablation and other minimally invasive interventional treatments) is increasingly valued by medical institutions at all levels, becoming the fourth type of treatment after surgery, radiotherapy and chemotherapy. In order to find the best treatment and prevent over-medicalization, the efficacy evaluation of multiple treatments is also receiving more attention.
[0008] Therefore, a new type of single-modality medical image registration method based on the improved Active-Demons model is needed, which can solve the problem that the existing Active-Demons model cannot well register images with blurred edges, unclear textures, and unclear features. It can be well applied to efficacy evaluation research, and can also be used for the correction of artifacts of medical imaging equipment or motion artifacts of human organs, growth monitoring of organisms, and pathological research on the development and changes of diseases over time. It has important scientific value and clinical significance. Summary of the invention
[0009] In order to overcome the above technical defects, the purpose of the present invention is to provide a single-modality medical image registration method, system and computer-readable storage medium, which can be used to register medical images with blurred edges, unclear textures and unclear features.
[0010] The present invention discloses a single-modality medical image registration method, comprising the following steps:
[0011] Acquiring a medical image to be registered, and obtaining a preprocessed image after preprocessing the medical image to be registered, wherein the preprocessed image includes a reference image and a floating image;
[0012] Initializing the displacement offset in the Active-Demons model so that the displacement offset U0=0, and setting the maximum number of iterations, the similarity metric, and the preset accuracy for the similarity metric in the Active-Demons model to input the reference image and the floating image in the preprocessed image;
[0013] Plotting the order of the fractional differential versus the amplitude of the image signal of the preprocessed image at different frequencies to configure an adaptive fractional order function;
[0014] Based on the fractional differential of the adaptive fractional order function, the deformation vector field U of the floating image is calculated, and Gaussian filtering is performed on the deformation vector field U;
[0015] The floating image is deformed based on the filtered deformation vector field U, and the deformed floating image is obtained by performing a bilinear interpolation operation;
[0016] Determine whether the similarity measurement between the floating image after the bilinear interpolation operation and the reference image reaches a preset accuracy, or whether the maximum number of iterations has been reached. If so, obtain the floating image after the bilinear interpolation operation as the registration image. If not, repeatedly iterate the above steps.
[0017] Preferably, the step of acquiring a medical image to be registered and preprocessing the medical image to be registered to obtain a preprocessed image, wherein the preprocessed image includes a reference image and a floating image, comprises:
[0018] Acquire a medical image to be registered, and determine whether part or all of the medical image to be registered includes redundant information. If there is redundant information, configure the gray value of the medical image to be registered to be zero;
[0019] The medical image to be registered includes a reference image and a preprocessed image of the floating image, and at least three pairs of feature points are selected in the reference image and the floating image respectively;
[0020] Affine transformation rigid coarse registration is performed on the reference image and the floating image at the feature points to obtain a preprocessed image.
[0021] Preferably, the step of plotting the order of the fractional differential versus the amplitude of the image signal of the preprocessed image at different frequencies to configure an adaptive fractional order function comprises:
[0022] The fractional-order differential form is derived from the integer-order differential form of the Active-Demons model to obtain the expression of the fractional-order differential: And the fractional differential form D of the image signal f(x) is calculated based on the following formula: α f(x) takes Fourier transform: Where α represents the fractional order, f(x) represents the image signal of the preprocessed image, Represents the fractional differential form D α f is the Fourier transform function, Γ represents the Gamma function, ω represents the frequency, represents the derivative with respect to x, represents integration over the interval [a,x];
[0023] Calculated based on the following formula The specific expression of ω is: Based on Plot the order of fractional differential versus the amplitude of the image signal at different frequencies to obtain an adaptive fractional order function with an order between [0,1], where sgn(·) is the sign function and i is the imaginary unit;
[0024] The average value of the gradient of each pixel in the image signal f(i,j) in eight directions is defined as G(i,j). After normalizing G(i,j), 0≤|G|≤1 is obtained, and an adaptive function α=log2(|G|+1) is defined, where log2 represents the logarithm with base 2.
[0025] Preferably, the steps of calculating the deformation vector field U of the floating image based on the fractional-order differential of the adaptive fractional-order function and performing Gaussian filtering on the deformation vector field U include:
[0026] The reference image and floating image in the preprocessed image are defined as two frames in a motion image sequence, and the deformation vector field is calculated according to the constraints of the grayscale image optical flow field. Where s represents the reference image, m represents the floating image, and U is the deformation vector field vector of the floating image. represents the gradient operator;
[0027] Apply the expression of fractional differential to the deformation vector field U to obtain the deformation vector field Among them, D α s is the α-order differential of the reference image;
[0028] Based on the Active-Demons model, the following expression of the deformation vector field U is obtained:
[0029]
[0030] Among them, D α m is the α-order differential of the floating image, D α s is the α-order differential of the reference image;
[0031] Construct a fractional differential mask and rewrite the fractional differential part in the deformation vector field U into the following convolution form: And extend the above convolution form to two-dimensional space, and obtain the following formula:
[0032]
[0033]
[0034] Then discretize the fractional differential mask to obtain the following form:
[0035] The pre-processed image is processed by fractional differentiation based on convolution, where f(x,y) is the image signal and H x (x,y,α),H y (x, y, α) is the mask operator, x M ,y N is the coordinate value of the feature point in the mask operator relative to the center point of the mask, * represents the convolution operation, It means partial derivative with respect to x.
[0036] Preferably, the step of calculating the deformation vector field U of the floating image based on the fractional differential of the adaptive fractional order function and performing Gaussian filtering on the deformation vector field U also includes:
[0037] Substitute the distances from other feature points in the neighborhood of the feature point in the mask operator to the center of the neighborhood into a two-dimensional Gaussian function to calculate and obtain a Gaussian template;
[0038] The center of the Gaussian template is aligned with the matrix of the preprocessed image, so that the Gaussian template is multiplied and added with the corresponding elements of the matrix. When the Gaussian template does not have an element somewhere in the matrix, the element at that place is filled with zero, and the window size of the Gaussian template is 10*10.
[0039] Preferably, the step of deforming the floating image based on the filtered deformation vector field U and obtaining the deformed floating image by performing a bilinear interpolation operation comprises:
[0040] Based on: U′(x)=U′(x)+U(x), the value of the deformation vector field U′ is updated, and the obtained deformation vector field correction value is added to the value of the previous deformation vector field U′ to obtain the updated value of the deformation vector field U′;
[0041] According to the updated deformation vector field U′, the floating image is deformed to m(x+U′(x)), where U′(x) is the deformation amount of the floating image and m(x) is the floating image before deformation.
[0042] Preferably, the step of deforming the floating image based on the filtered deformation vector field U and obtaining the deformed floating image by performing a bilinear interpolation operation further comprises:
[0043] The floating images are interpolated using bilinear interpolation to obtain the registered images.
[0044] The present invention also discloses a single-modality medical image registration system, comprising:
[0045] An acquisition module acquires the medical image to be registered, and obtains a preprocessed image after preprocessing the medical image to be registered, wherein the preprocessed image includes a reference image and a floating image;
[0046] An initialization module receives an initialized preprocessed image, initializes a displacement offset in an Active-Demons model so that the displacement offset U0=0, and sets a maximum number of iterations, a similarity metric, and a preset accuracy for the similarity metric in the Active-Demons model to input a reference image and a floating image in the preprocessed image;
[0047] A drawing module draws a graph of the relationship between the order of the fractional differential and the amplitude of the image signal of the preprocessed image at different frequencies to configure an adaptive fractional order function;
[0048] The processing module calculates the deformation vector field U of the floating image based on the fractional differential of the adaptive fractional order function, and performs Gaussian filtering on the deformation vector field U;
[0049] An interpolation module deforms the floating image based on the filtered deformation vector field U and obtains the deformed floating image by performing a bilinear interpolation operation;
[0050] The output module determines whether the similarity measurement between the floating image after the bilinear interpolation operation and the reference image reaches a preset accuracy, or whether the maximum number of iterations is reached. If so, the floating image after the bilinear interpolation operation is obtained as the registration image. If not, the above steps are repeatedly iterated.
[0051] The present invention further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above steps are implemented.
[0052] Compared with the prior art, the above technical solution has the following beneficial effects:
[0053] 1. By combining the adaptive fractional differential method, the most suitable fractional order for dealing with problems such as blurred edges, unclear textures, and unclear features in medical images can be quickly found;
[0054] 2. The output registration results are accurately aligned. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 A schematic diagram of a flow chart of a registration method for a single-modality medical image in accordance with a preferred embodiment of the present invention;
[0056] Figure 2 A schematic diagram of the relationship between the order of fractional differential and the amplitude of the image signal at different frequencies in accordance with a preferred embodiment of the present invention;
[0057] Figure 3 A schematic diagram of a mask operator of a fractional differential mask in accordance with a preferred embodiment of the present invention;
[0058] Figure 4 It is a schematic diagram of a registration image of a single-modality medical image in accordance with a preferred embodiment of the present invention. DETAILED DESCRIPTION
[0059] The advantages of the present invention are further described below in conjunction with the accompanying drawings and specific embodiments.
[0060] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0061] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. The singular forms of "a", "said" and "the" used in this disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0062] It should be understood that although the terms first, second, third, etc. may be used in the present disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0063] In the description of the present invention, it is necessary to understand that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.
[0064] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection of two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.
[0065] In the following description, the suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention, and have no specific meanings. Therefore, "module" and "component" can be used interchangeably.
[0066] See also Figure 1 , in accordance with a single-modality medical image registration method in a preferred embodiment of the present invention, in this embodiment, the registration method comprises the following steps:
[0067] S100: Acquire a medical image to be registered, and preprocess the medical image to be registered to obtain a preprocessed image, wherein the preprocessed image includes a reference image and a floating image;
[0068] The purpose of image registration is to align the floating image to be calibrated with the reference image as a template. Therefore, what is collected in this embodiment is mainly the medical image to be registered, and after preprocessing, the reference image and the floating image are obtained.
[0069] S200: Initializing the displacement offset in the Active-Demons model so that the displacement offset U0=0, and setting the maximum number of iterations, the similarity metric and the preset accuracy for the similarity metric in the Active-Demons model to input the reference image and the floating image in the preprocessed image;
[0070] Demons algorithm is a registration method developed based on optical flow theory. Like other optical flow algorithms, this algorithm has the constraints of constant brightness and small motion. Active-Demons model adds the gradient of floating image to the calculation of offset. In the original algorithm, the internal force driving diffusion is the gradient of reference image. At this time, the gradient of floating image is added as the new internal force, thus accelerating the convergence speed of iteration. In this step, the displacement offset in Active-Demons model is initialized so that the displacement offset U0=0, and the maximum number of iterations, similarity measure and preset accuracy for the similarity measure in Active-Demons model are set to input the reference image and floating image in the preprocessed image. Similarity measure is a criterion for measuring the quality of each transformation result, which is used to evaluate the registration result and provide a basis for the next step of the search strategy. Similarity transformation is closely related to feature space and search space. Different feature spaces often correspond to different similarity measures; and the value of similarity measure will directly determine the selection of registration transformation and judge whether the image is correctly registered under the current transformation model. Usually, the anti-interference ability of the registration algorithm is determined by the extracted features and the similarity measure. Commonly used similarity measures include mutual information, normalized information, joint entropy, correlation, Euclidean distance, gradient cross-correlation, etc.
[0071] S300: drawing a graph of the relationship between the order of fractional differential and the amplitude of the image signal of the preprocessed image at different frequencies, so as to configure an adaptive fractional order function;
[0072] Fractional differentiation is a branch of calculus that performs fractional differentiation of functions. For example, for x n beg Derivative: First, for x n After taking the first-order derivative, we get nx n-1 , after the second-order derivative is n(n-1)x n-2 ...then when m<n, the mth-order derivative is n(n-1)(n-2)···(n-m+1)x n-m , that is Since the concept of factorial can be extended: n! = Γ(n+1), for x n After taking the m-th order derivative, we get make Then we get: In particular, x The derivative is sin(x) The derivative is The amplitude relationship diagram of the image signal (or image function) of the preprocessed image at different frequencies (see Figure 2) is drawn to design an adaptive fractional-order function that can solve the problems of weak edges, weak textures and insignificant features.
[0073] S400: calculating a deformation vector field U of the floating image based on a fractional-order differential of an adaptive fractional-order function, and performing Gaussian filtering on the deformation vector field U;
[0074] After obtaining the above optimal fractional order, the floating image is subjected to fractional differential processing to calculate the deformation vector field U of the floating image. The deformation vector field U can also be called the displacement vector field (DVF), which is the field composed of the displacement of each pixel of the image. After Gaussian filtering, the displacement vector of each pixel on the floating image is obtained.
[0075] S500: deforming the floating image based on the filtered deformation vector field U, and obtaining the deformed floating image by performing a bilinear interpolation operation;
[0076] After the deformation vector field U is obtained, each pixel of the floating image will be deformed according to the deformation vector field U, and the deformed floating image will be obtained by performing a bilinear interpolation operation. Bilinear interpolation is also called bilinear interpolation. Mathematically, bilinear interpolation is an extension of linear interpolation on a two-dimensional rectangular grid. It is used to interpolate two-variable functions (such as x and y). The core idea is to perform linear interpolation in two directions respectively. At the same time, bilinear interpolation also refers to a method of using a straight line connecting two known quantities to determine the value of an unknown quantity between the two known quantities. For example, if you want to get the value of the unknown function f at point P = (x, y), assuming that the known function f is at Q 11 =(x1,y1),Q 12 =(x1,y2),Q 21 =(x2,y1),Q 22 =(x2,y2) values of four points. First, perform linear interpolation in the x direction to obtain f(R1) and f(R2).
[0077]
[0078]
[0079] Then perform linear interpolation in the y direction to obtain f(P).
[0080]
[0081] The integrated formula is as follows:
[0082]
[0083] S600: Determine whether the similarity measure between the floating image after the bilinear interpolation operation and the reference image reaches a preset accuracy, or whether the maximum number of iterations is reached. If so, obtain the floating image after the bilinear interpolation operation as the registration image. If not, repeatedly iterate the above steps.
[0084] According to the maximum number of iterations and the similarity measure set in step S200, it is determined whether the similarity measure between the floating image after the bilinear interpolation operation and the reference image reaches the preset accuracy, or whether the maximum number of iterations is reached. When any of the above conditions is met, it is determined that the desired image registration is completed, so that the floating image after the bilinear interpolation operation is obtained as the registered image and output. If any of the above conditions is not met, steps S300-S500 are repeatedly iterated until any of the above conditions is met.
[0085] Through the above steps, the corresponding optical flow field model is established for the medical image to be registered. At the same time, the influence of the order of the fractional differential on the amplitude of the image at different frequencies is combined to design the corresponding adaptive fractional order function. Secondly, the gradient in the expression of the deformation vector field U calculated by the optical flow field model is replaced by the fractional differential. Thirdly, the Gaussian filter is used to smooth the deformation vector field U, and then act on the floating image in the image to be registered. Then, the corrected image is interpolated. The above process is iterated until the similarity measure value between the reference image and the floating image meets the predetermined requirements, and the final registration result is obtained, which has important scientific significance and clinical application value in the precise analysis of medical images, prediction and identification of diseases, planning of treatment plans, and quantitative evaluation of treatment effects.
[0086] In a preferred embodiment, step S100 includes:
[0087] S110: Acquire a medical image to be registered, and determine whether part or all of the medical image to be registered includes redundant information. If redundant information is present, configure the gray value of the medical image to be registered to be zero;
[0088] Redundant information refers to whether there is useless information in certain areas of the medical image to be registered, such as background, debris, etc. If there is such useless information, its gray value is set to zero.
[0089] S120: The medical image to be registered includes a reference image and a preprocessed image of the floating image, and at least three pairs of feature points are selected in the reference image and the floating image respectively;
[0090] The selection of feature points can be random or targeted at points within the organ region of the medical image to be registered.
[0091] S130: Performing affine transformation rigid coarse registration on the reference image and the floating image at the feature points to obtain a preprocessed image.
[0092] Affine transformation, also known as affine mapping, refers to a linear transformation of a vector space followed by a translation in geometry, transforming it into another vector space. The core idea of rigid coarse registration is as follows: one of the images to be registered is set as a reference, called the standard image; the other image is spatially transformed based on the standard image, called the floating image. The process of rigid coarse registration can be decomposed into a series of rotation and translation steps. The floating image undergoes a series of transformations to achieve consistency in spatial position with the corresponding pixels on the standard image. Generally speaking, the images to be registered that participate in rigid registration can be either rigid single-modality images or rigid multi-modality images, such as the registration between CT-MRI. This paper adopts a pixel-based image registration method, which uses some grayscale statistical information of the image itself without the need to extract image features in advance and perform other complex segmentation operations, and estimates the image transformation based on the entire image.
[0093] Preferably or optionally, step S300 includes:
[0094] S310: The fractional-order differential form is derived from the integer-order differential form of the Active-Demons model to obtain the expression of the fractional-order differential: And the fractional differential form D of the image signal f(x) is calculated based on the following formula: α f(x) takes Fourier transform: The fractional order differential in the present invention adopts the RL definition, α represents the fractional order, f(x) represents the image signal of the preprocessed image, Represents the fractional differential form D α f is the Fourier transform function, Γ represents the Gamma function, ω represents the frequency, represents the derivative with respect to x, represents integration over the interval [a,x];
[0095] S320: Calculated based on the following formula The specific expression of ω is: Based on Plot the order of the fractional derivative versus the amplitude of the image signal at different frequencies (see Figure 2 ) to obtain an adaptive fractional order function with an order between [0,1], where sgn(·) is the sign function and i is the imaginary unit;
[0096] S330: Define the average value of the gradient of each pixel in the image signal f(i,j) in eight directions as G(i,j), normalize G(i,j), and obtain 0≤|G|≤1, and define the adaptive function α=log2(|G|+1), where log2 represents the logarithm with base 2. The adaptive function α=log2(|G|+1) can solve the problems of weak edges, weak textures and insignificant features of the image to be registered. When G is larger, that is, corresponding to the edge and texture areas, the degree of enhancement of the mid-high frequency area (edge and texture area) should be greater. From the amplitude relationship diagram, it can be seen that the order should be as large as possible in the [0,1] interval; when G is smaller, that is, corresponding to the smooth area, the degree of retention of the low frequency area (smooth area) should be greater. From the amplitude relationship diagram, it can be seen that the order should be as small as possible in the [0,1] interval.
[0097] Preferably or optionally, step S400 includes:
[0098] S410: defining the reference image and the floating image in the preprocessed image as two frames in a motion image sequence, and calculating the deformation vector field according to the constraint condition of the grayscale image optical flow field Where s represents the reference image, m represents the floating image, and U is the deformation vector field vector of the floating image. represents the gradient operator;
[0099] S420: Apply the expression of fractional differential to the deformation vector field U to obtain the deformation vector field Among them, D α s is the α-order differential of the reference image, that is, the deformation vector field U is obtained by utilizing the fractional-order differential obtained in the previous step.
[0100] S430: Based on the Active-Demons model, the following expression of the deformation vector field U is obtained:
[0101]
[0102] Among them, D α m is the α-order differential of the floating image, D α s is the α-order differential of the reference image;
[0103] S440: construct a fractional-order differential mask, and rewrite the fractional-order differential part in the deformation vector field U into the following convolution form: Since the image to be registered is a two-dimensional signal, it is necessary to expand the fractional differential part of the convolution form to the two-dimensional space. Therefore, after expanding the above convolution form to the two-dimensional space, the following formula is obtained:
[0104]
[0105]
[0106] S450: discretize the fractional-order differential mask to obtain the following form:
[0107] The pre-processed image is processed by fractional differentiation based on convolution, where f(x,y) is the image signal and H x (x,y,α),H y (x, y, α) is the mask operator (such as Figure 3 shown), x M ,y N is the coordinate value of the feature point in the mask operator relative to the center point of the mask, * represents the convolution operation, It means partial derivative with respect to x.
[0108] Furthermore, step S400 further includes:
[0109] S460: Substitute the distances from other feature points in the neighborhood of the feature point in the mask operator to the center of the neighborhood into a two-dimensional Gaussian function to calculate and obtain a Gaussian template;
[0110] The Gaussian function is discretized, and the obtained Gaussian function value is used as the coefficient of the Gaussian template.
[0111] S470: Align the center of the Gaussian template with the matrix of the preprocessed image so that the Gaussian template is multiplied with the corresponding elements of the matrix and then added, wherein when the Gaussian template does not have an element somewhere in the matrix, the element at that place is filled with zero, and the window size of the Gaussian template is 10*10.
[0112] Preferably or optionally, step S500 includes:
[0113] S510: based on: U′(x)=U′(x)+U(x), the value of the deformation vector field U′ is updated, and the obtained deformation vector field correction value is added to the value of the previous deformation vector field U′ to obtain the updated value of the deformation vector field U′;
[0114] That is, in step S510 , the deformation vector field U′ is continuously iterated for correction.
[0115] S520: According to the updated deformation vector field U′, the floating image is deformed to m(x+U′(x)), wherein U′(x) is the deformation amount of the floating image and m(x) is the floating image before deformation.
[0116] Preferably or optionally, step S600 includes: S610: using a bilinear interpolation method to perform an interpolation operation on the deformed floating image to obtain a registration image (such as Figure 4 shown).
[0117] The present invention also discloses a registration system for single-modality medical images, comprising: an acquisition module, which acquires the medical image to be registered, and obtains a preprocessed image after preprocessing the medical image to be registered, wherein the preprocessed image includes a reference image and a floating image; an initialization module, which receives the initialized preprocessed image, and initializes the displacement offset in the Active-Demons model so that the displacement offset U0=0, and sets the maximum number of iterations, the similarity measure and the preset accuracy for the similarity measure in the Active-Demons model to input the reference image and the floating image in the preprocessed image; and a drawing module, which draws a graph of the order of the fractional differential against the preprocessed image. The image signal is an amplitude relationship diagram at different frequencies to configure an adaptive fractional order function; the processing module calculates the deformation vector field U of the floating image based on the fractional differential of the adaptive fractional order function, and performs Gaussian filtering on the deformation vector field U; the interpolation module deforms the floating image based on the filtered deformation vector field U, and obtains the deformed floating image by performing a bilinear interpolation operation; the output module determines whether the similarity measurement between the floating image after the bilinear interpolation operation and the reference image reaches a preset accuracy, or whether the maximum number of iterations is reached. If so, the floating image that performs the bilinear interpolation operation is obtained as the registration image, if not, the above steps are iterated repeatedly.
[0118] The present invention further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above steps are implemented.
[0119] It should be noted that the embodiments of the present invention have better practicability and do not impose any form of limitation on the present invention. Any technician familiar with the field may use the technical content disclosed above to change or modify it into an equivalent effective embodiment. However, any modification or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.
Claims
1. A single-modality medical image registration method, characterized in that: The following steps are involved: Acquiring a medical image to be registered, and preprocessing the medical image to be registered to obtain a preprocessed image, wherein the preprocessed image includes a reference image and a floating image; Initializing the displacement offset in the Active-Demons model so that the displacement offset U0=0, and setting the maximum number of iterations, the similarity metric, and the preset accuracy for the similarity metric in the Active-Demons model to input the reference image and the floating image in the preprocessed image; Plotting a graph of the relationship between the order of the fractional differential and the amplitude of the image signal of the preprocessed image at different frequencies to configure an adaptive fractional order function; Based on the fractional differential of the adaptive fractional order function, a deformation vector field U of the floating image is calculated, and Gaussian filtering is performed on the deformation vector field U; The floating image is deformed based on the filtered deformation vector field U, and the deformed floating image is obtained by performing a bilinear interpolation operation; Determine whether the similarity measurement between the floating image after the bilinear interpolation operation and the reference image reaches the preset accuracy, or whether the maximum number of iterations is reached. If so, obtain the floating image after the bilinear interpolation operation as the registration image. If not, repeatedly iterate the above steps, where The steps of acquiring a medical image to be registered and preprocessing the medical image to be registered to obtain a preprocessed image, wherein the preprocessed image includes a reference image and a floating image, include: Acquire a medical image to be registered, and determine whether part or all of the medical image to be registered includes redundant information, and if there is redundant information, configure the gray value of the medical image to be registered to be zero; The medical image to be registered includes a reference image and a preprocessed image of the floating image, and at least three pairs of feature points are selected in the reference image and the floating image respectively; At the feature points, performing affine transformation rigid coarse registration on the reference image and the floating image to obtain a preprocessed image; The steps of calculating the deformation vector field U of the floating image based on the fractional differential of the adaptive fractional order function and performing Gaussian filtering on the deformation vector field U include: The reference image and floating image in the preprocessed image are defined as two frames in a motion image sequence, and the deformation vector field is calculated according to the constraints of the grayscale image optical flow field. Where s represents the reference image, m represents the floating image, U is the deformation vector field vector of the floating image, and ▽ represents the gradient operator; Apply the expression of fractional differential to the deformation vector field U to obtain the deformation vector field Among them, D α s is the α-order differential of the reference image; Based on the Active-Demons model, the following expression of the deformation vector field U is obtained: Among them, D α m is the α-order differential of the floating image, D α s is the α-order differential of the reference image.
2. The registration method according to claim 1, characterized in that: The step of plotting the order of the fractional differential versus the amplitude of the image signal of the preprocessed image at different frequencies to configure an adaptive fractional order function includes: deriving the fractional differential form from the integer differential form of the Active-Demons model to obtain an expression of the fractional differential: And the fractional differential form D of the image signal f(x) is calculated based on the following formula: α f(x) takes Fourier transform: Where α represents the fractional order, f(x) represents the image signal of the preprocessed image, Represents the fractional differential form D α f is the Fourier transform function, Γ represents the Gamma function, ω represents the frequency, represents the derivative with respect to x, represents integration over the interval [a,x]; Calculated based on the following formula The specific expression of ω is: Based on Plot the order of fractional differential versus the amplitude of the image signal at different frequencies to obtain an adaptive fractional order function with an order between [0,1], where sgn(·) is the sign function and i is the imaginary unit; The average value of the gradient of each pixel in the image signal f(i,j) in eight directions is defined as G(i,j). After normalizing G(i,j), 0≤|G|≤1 is obtained, and an adaptive function α=log2(|G|+1) is defined, where log2 represents the logarithm with base 2.
3. The registration method according to claim 2, characterized in that: The step of calculating the deformation vector field U of the floating image based on the fractional differential of the adaptive fractional order function and performing Gaussian filtering on the deformation vector field U also includes: constructing a fractional differential mask to rewrite the fractional differential part in the deformation vector field U into the following convolution form: And extend the above convolution form to two-dimensional space, and obtain the following formula: Then discretize the fractional differential mask to obtain the following form: The pre-processed image is processed by fractional differentiation based on convolution, where f(x,y) is the image signal and H x (x,y,α),H y (x, y, α) is the mask operator, x M ,y N is the coordinate value of the feature point in the mask operator relative to the center point of the mask, * represents the convolution operation, It means partial derivative with respect to x.
4. The registration method according to claim 3, characterized in that: The step of calculating the deformation vector field U of the floating image based on the fractional differential of the adaptive fractional order function and performing Gaussian filtering on the deformation vector field U also includes: bringing the distances from other feature points in the neighborhood of the feature point in the mask operator to the center of the neighborhood into the two-dimensional Gaussian function to calculate and obtain a Gaussian template; The center of the Gaussian template is aligned with the matrix of the preprocessed image, so that the Gaussian template is multiplied and then added with the corresponding elements of the matrix, wherein when the Gaussian template does not have an element somewhere in the matrix, the element there is filled with zero, and the window size of the Gaussian template is 10*10.
5. The registration method according to claim 3, characterized in that: The steps of deforming the floating image based on the filtered deformation vector field U and obtaining the deformed floating image by performing a bilinear interpolation operation include: Based on: U′(x)=U′(x)+U(x), the value of the deformation vector field U′ is updated, and the obtained deformation vector field correction value is added to the value of the previous deformation vector field U′ to obtain the updated value of the deformation vector field U′; According to the updated deformation vector field U′, the floating image is deformed to m(x+U′(x)), where U′(x) is the deformation amount of the floating image and m(x) is the floating image before deformation.
6. The registration method according to claim 5, characterized in that: The step of deforming the floating image based on the filtered deformation vector field U and obtaining the deformed floating image by performing a bilinear interpolation operation also includes: The floating images are interpolated using bilinear interpolation to obtain the registered images.
7. A single-modality medical image registration system, characterized in that: The registration system comprises: an acquisition module for acquiring a medical image to be registered, and preprocessing the medical image to be registered to obtain a preprocessed image, wherein the preprocessed image comprises a reference image and a floating image, wherein The acquisition module acquires the medical image to be registered, and determines whether part or all of the medical image to be registered includes redundant information. If redundant information is present, the gray value of the medical image to be registered is configured to be zero; The medical image to be registered includes a reference image and a preprocessed image of the floating image, and at least three pairs of feature points are selected in the reference image and the floating image respectively; At the feature points, affine transformation rigid coarse registration is performed on the reference image and the floating image to obtain a preprocessed image; an initialization module receives the preprocessed image, and initializes the displacement offset in the Active-Demons model so that the displacement offset U0=0, and sets the maximum number of iterations, similarity measure and preset accuracy for the similarity measure in the Active-Demons model to input the reference image and the floating image in the preprocessed image; A drawing module, which draws a diagram of the relationship between the order of the fractional differential and the amplitude of the image signal of the preprocessed image at different frequencies, so as to configure an adaptive fractional order function; The processing module calculates the deformation vector field U of the floating image based on the fractional differential of the adaptive fractional order function, and performs Gaussian filtering on the deformation vector field U, wherein The processing module defines the reference image and the floating image in the preprocessed image as two frames in a motion image sequence, and calculates the deformation vector field according to the constraints of the grayscale image optical flow field. Where s represents the reference image, m represents the floating image, U is the deformation vector field vector of the floating image, and ▽ represents the gradient operator; The processing module applies the expression of fractional differential to the deformation vector field U to obtain the deformation vector field Among them, D α s is the α-order differential of the reference image; The processing module is based on the Active-Demons model and obtains the following expression of the deformation vector field U: Among them, D α m is the α-order differential of the floating image, D α s is the α-order differential of the reference image; An interpolation module deforms the floating image based on the filtered deformation vector field U and obtains the deformed floating image by performing a bilinear interpolation operation; The output module determines whether the similarity measurement between the floating image after the bilinear interpolation operation and the reference image reaches a preset accuracy, or whether the maximum number of iterations is reached. If so, the floating image after the bilinear interpolation operation is obtained as the registration image. If not, the above steps are repeatedly iterated.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps according to any one of claims 1 to 6 are implemented.
