A method, device, computer equipment and storage medium for generating simulated CT images

Abstract

The present invention is applicable to the computer field and provides a method for generating a simulated CT image, comprising: obtaining an MR scan image; processing the MR scan image to obtain multiple mdixon sequence images; determining feature information of each image point in the scan image based on the sequence images; determining the probability of a component category corresponding to each point based on the feature information of each point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm; determining the CT value corresponding to each point based on the probability of the component category corresponding to each point; and generating a simulated CT image based on the CT value. The simulated CT image generation method provided in an embodiment of the present invention, after inputting an MR scan image, generates multiple mdixon sequence images through processing, then extracts feature information of each image point from the multiple mdixon sequence images, combines the four-element medical image classification model generated by pre-training, and directly determines the CT value corresponding to each image point, thereby generating a simulated CT image. The method processes less data and effectively improves processing efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of computers, and in particular relates to a method and device for generating a simulated CT image, a computer device, and a storage medium. Background Art

[0002] CT (Computed Tomography) and MR (Magnetic Resonance) are two commonly used imaging techniques in medicine. CT uses X-ray beams to scan a layer of a certain thickness within a part of the human body. A detector receives the X-rays that pass through the layer, converts them into visible light, and then converts them into electrical signals through photoelectric conversion. This is then converted to digital form by a digitizer and input into a computer for processing. MR applies video pulses of a specific frequency to the human body in an electrostatic field, exciting hydrogen protons in tissue and causing nuclear magnetic resonance. When the radio frequency pulse is terminated, the protons induce MR signals during their relaxation process. The MR signals are then processed through reception, spatial encoding, and image reconstruction to produce MR images.

[0003] In existing diagnostic processes, it's often necessary to simulate possible CT images based on MR images. However, existing CT simulation techniques rely on matching corresponding points between images and then generating CT images through computational simulation. This requires a large amount of computational data, resulting in low image processing efficiency and poor performance of the simulated CT images.

[0004] It can be seen that the existing technology for generating images by simulating CT images still has technical problems such as low image processing efficiency and poor generation effect. Summary of the Invention

[0005] The purpose of the embodiments of the present invention is to provide a method for generating a simulated CT image, aiming to solve the technical problems of low image processing efficiency and poor generation effect in the existing simulated CT image generation technology.

[0006] The embodiment of the present invention is implemented as follows: a method for generating a simulated CT image, comprising:

[0007] acquiring an MR scan image to be converted into a simulated CT image;

[0008] Processing the MR scan image to obtain a plurality of MDIXON sequence images of the MR scan image, wherein the plurality of MDIXON sequence images include a fat map, a water map, a positive phase map, and a negative phase map;

[0009] determining feature information of each image point in the MR scan image according to a plurality of mdixon sequence images, wherein the feature information includes a spatial coordinate of each image point and an MR value in the plurality of mdixon sequence images;

[0010] Determining the probability of a component category corresponding to each image point based on the feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm, wherein the component categories include fat, soft tissue, air, and bone;

[0011] Determining the CT value corresponding to each image point according to the probability of the component category corresponding to each image point;

[0012] A simulated CT image is generated according to the CT value.

[0013] Another object of an embodiment of the present invention is to provide a simulated CT image generating device, characterized in that the method includes:

[0014] an MR scan image acquisition unit, configured to acquire an MR scan image to be converted into a simulated CT image;

[0015] an MR scan image processing unit, configured to process the MR scan image to obtain a plurality of MDIXON sequence images of the MR scan image, wherein the plurality of MDIXON sequence images include a fat map, a water map, a positive phase map, and a negative phase map;

[0016] a feature information determining unit, configured to determine feature information of each image point in the MR scan image based on a plurality of MDIXON sequence images, wherein the feature information includes a spatial coordinate of each image point and an MR value in the plurality of MDIXON sequence images;

[0017] a classification unit, configured to determine a probability of a component category corresponding to each image point based on feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm, wherein the component categories include fat, soft tissue, air, and bone;

[0018] a CT value determining unit, configured to determine a CT value corresponding to each image point according to a probability of a component category corresponding to each image point; and

[0019] The simulated CT image generating unit is configured to generate a simulated CT image according to the CT value.

[0020] Another object of an embodiment of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the simulated CT image generation method as described above.

[0021] Another object of an embodiment of the present invention is a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the steps of the simulated CT image generation method as described above.

[0022] An embodiment of the present invention provides a method for generating a simulated CT image. The method processes an MR scan image into multiple MDIXON sequence images, including a fat map, a water map, a positive phase map, and a negative phase map. Feature information of each image point in the original MR scan image is then determined based on each sequence image. The feature information includes the spatial coordinates of each image point and its MR value in the multiple MDIXON sequence images. The probability of a component category corresponding to each image point is determined based on the feature information of each image point and a pre-trained quaternary medical image classification model generated based on a hierarchical support vector machine algorithm. The component categories include fat, soft tissue, air, and bone. The corresponding CT value is determined based on the probability of each component category, thereby directly determining a simulated CT image. The method provided by the embodiment of the present invention can, upon inputting an MR scan image, automatically process the MR image and directly output a corresponding CT image, effectively improving processing efficiency. Furthermore, the quaternary medical image classification model used in the processing process is pre-trained using a large amount of sample data, effectively ensuring accuracy during the processing and improving the quality of the generated CT image. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a method for generating a simulated CT image provided by an embodiment of the present invention;

[0024] Figure 2 A flowchart of the steps of an image point classification method provided by an embodiment of the present invention;

[0025] Figure 3 A flowchart of the steps for training and generating a four-element medical image classification model provided by an embodiment of the present invention;

[0026] Figure 4 A flowchart of the steps for training and generating multiple binary medical image classification models provided by an embodiment of the present invention;

[0027] Figure 5 A schematic structural diagram of a simulated CT image generating device provided by an embodiment of the present invention;

[0028] Figure 6 A schematic structural diagram of a classification unit provided in an embodiment of the present invention;

[0029] Figure 7A schematic structural diagram of a four-element medical image classification model training and generation unit provided in an embodiment of the present invention;

[0030] Figure 8 A schematic diagram of the structure of a binary medical image classification model training and generation module provided by an embodiment of the present invention.

[0031] Figure 9 This is a performance comparison chart of the simulated CT image generation method provided by an embodiment of the present invention and the traditional simulated CT image generation method. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0033] It is understood that the terms "first," "second," etc., used herein may be used to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script without departing from the scope of this application.

[0034] In order to solve the technical problems of low image processing efficiency and poor generation effect in the existing CT simulation image generation technology, the embodiment of the present invention provides a CT simulation image method with high image processing efficiency and high image generation quality. After receiving the MR scan image, the characteristic information in the MR scan image is extracted, and the characteristic information is used as input. In combination with a pre-trained four-element medical image classification model, the probability of each image point corresponding to the component category is directly determined. The four-element medical image classification model is pre-trained based on the hierarchical support vector machine algorithm and generated using sample data. Finally, the corresponding CT value of each image point is determined according to the probability of the corresponding component category, thereby directly generating a simulated CT image. The entire simulated CT image generation process has a small amount of data processing and high processing efficiency, and the simulated CT image finally generated has high quality.

[0035] like Figure 1 As shown, in one embodiment, a method for generating a simulated CT image is proposed, which may specifically include the following steps:

[0036] Step S102: Acquire an MR scan image to be converted into a simulated CT image.

[0037] In the embodiment of the present invention, a CT image is composed of a certain number of pixels of varying grayscales, ranging from black to white, arranged in a matrix. These pixels actually reflect the X-ray absorption coefficient of the corresponding voxel, i.e., the CT value. Therefore, it is only necessary to determine the CT value corresponding to each image point in the MR image to determine the corresponding simulated CT image.

[0038] Step S104: Process the MR scan image to obtain a plurality of mdixon sequence images of the MR scan image.

[0039] In an embodiment of the present invention, the MDIXON is a fat suppression technology that utilizes a spin echo sequence to collect positive phase and reverse phase echo signals of water and fat protons at different echo times. By adding and subtracting the two non-passing signals, the effect of water and fat separation is achieved, that is, multiple MDIXON sequence images can be obtained, including fat images, water images, positive phase images, and reverse phase images.

[0040] Step S106 : determining feature information of each image point in the MR scan image according to a plurality of mdixon sequence images.

[0041] In an embodiment of the present invention, the feature information includes the spatial coordinates of each image point and the MR value in a plurality of MDixon sequence images.

[0042] In an embodiment of the present invention, for each image point in the MR scan image, the spatial coordinates of the image point can be represented by a three-dimensional coordinate. At the same time, the MR value corresponding to the image point in each mdixon sequence image is combined with the spatial coordinates of the image point as the feature information of the image point.

[0043] Step S108 , determining the probability of the component category corresponding to each image point according to the feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm.

[0044] In an embodiment of the present invention, the component categories include fat, soft tissue, air, and bone. That is, after inputting a set of feature information, the probability of the image point being fat, soft tissue, air, or bone can be directly determined based on the feature information.

[0045] As a preferred embodiment of the present invention, since the classification model generated by the support vector machine algorithm training is used for binary classification, in order to realize the four-element medical image classification model, we use the hierarchical support vector machine algorithm for training, that is, first train and generate multiple binary medical image classification models, and then construct a decision tree of the multiple binary medical image classification models to generate a four-element medical image classification model.

[0046] Step S110 : determining the CT value corresponding to each image point according to the probability of the component category corresponding to each image point.

[0047] In the embodiment of the present invention, the probability of the component category corresponding to each image point is used as a weight, and a weighted operation is performed on the standard CT value of each component category to determine the CT value of each image point.

[0048] Step S112: generating a simulated CT image according to the CT value.

[0049] In the embodiment of the present invention, the CT value is generally between -1000Hu and 1000Hu. Among them, the common CT value of dense bone is 1000Hu, the CT value of water is 0Hu, and the CT value of air is -1000Hu. After knowing the CT value of each image point, a CT image can be simply determined. This CT image is a simulated CT image.

[0050] An embodiment of the present invention provides a method for generating a simulated CT image. The method processes an MR scan image into multiple MDIXON sequence images, including a fat map, a water map, a positive phase map, and a negative phase map. Feature information of each image point in the original MR scan image is then determined based on each sequence image. The feature information includes the spatial coordinates of each image point and its MR value in the multiple MDIXON sequence images. The probability of a component category corresponding to each image point is determined based on the feature information of each image point and a pre-trained quaternary medical image classification model generated based on a hierarchical support vector machine algorithm. The component categories include fat, soft tissue, air, and bone. The corresponding CT value is determined based on the probability of each component category, thereby directly determining a simulated CT image. The method provided by the embodiment of the present invention can, upon inputting an MR scan image, automatically process the MR image and directly output a corresponding CT image, effectively improving processing efficiency. Furthermore, the quaternary medical image classification model used in the processing process is pre-trained using a large amount of sample data, effectively ensuring accuracy during the processing and improving the quality of the generated CT image.

[0051] In a preferred embodiment, since the amount of image point data contained in the MR scan image is too large, if each image point is classified one by one, the processing efficiency will be low. Therefore, an image classification optimization method based on the K-nearest neighbor algorithm is provided to improve the efficiency of image point classification. Figure 2 As shown, step S106 specifically includes the following steps:

[0052] Step S202 : determining a plurality of sample image points in the MR scan image according to a preset sampling rule.

[0053] In the embodiment of the present invention, the preset sampling rule may be simple random sampling, or fixed-point sampling may be used to ensure the comprehensiveness of the data.

[0054] In the embodiment of the present invention, preferably, the number of sample image points is about 20,000, which can effectively ensure the final image effect while reducing the amount of image point classification calculation.

[0055] Step S204 : determining the probability of the component category corresponding to each sample image point according to the feature information of each sample image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm.

[0056] In the embodiment of the present invention, similarly, it is necessary to determine the probability of the component category corresponding to each sample image point, where the component categories include fat, soft tissue, air, and bone.

[0057] Step S206 : determining a plurality of reference image points closest to each remaining image point in the MR scan image from the plurality of sample image points, and determining a distance between each remaining image point and the plurality of reference image points.

[0058] In the embodiment of the present invention, we use the K-nearest neighbor algorithm to calculate the component probability of the remaining image points, that is, we need to determine the K sample image points closest to the remaining image points as reference image points. Generally, the K value is preferably 5-7.

[0059] In the embodiment of the present invention, compared with the conventional K-nearest neighbor algorithm, the distance between each remaining image point and its K reference points is additionally obtained.

[0060] Step S208 : determining the influence weights of the multiple reference image points on the respective remaining image points according to the distances between the respective remaining image points and the multiple reference image points.

[0061] In an embodiment of the present invention, compared to the conventional K-nearest neighbor algorithm that directly uses the mode of the attributes of K reference points to determine the attributes of the image point to be determined, the present invention also introduces weights to determine the attributes of the remaining image points, wherein the weights are determined based on the distance between each remaining image point and its reference image point, that is, the closer the distance is to the reference image point, the greater the influence factor.

[0062] Step S210 : determining the probability of the component category corresponding to each remaining image point according to the influence weights of the multiple reference image points on each remaining image point and the probability of the component category corresponding to each reference image point.

[0063] In an embodiment of the present invention, after determining the influencing factors of multiple reference image points and the probabilities of the component categories corresponding to the multiple reference points, the probabilities of the component categories corresponding to the remaining image points can be determined by weighted calculation, thus completing the classification of the component categories of each image point in the entire MR scan image.

[0064] An embodiment of the present invention provides a method for determining the probability of a component category corresponding to each image point. By combining the K-nearest neighbor algorithm, the probabilities of the component categories corresponding to some sample points are first calculated. Then, a reference point is determined based on the distances of the remaining image points from the sample points. Finally, the probabilities of the component categories corresponding to the remaining image points are determined based on the probabilities of the component categories corresponding to the reference points. Compared with calculating each image point one by one, this method effectively reduces the amount of data processing while ensuring accuracy.

[0065] In one embodiment, Figure 3 As shown, the method for generating a four-element medical image classification model based on hierarchical support vector machine algorithm training may specifically include the following steps:

[0066] Step S302 : Generate multiple binary medical image classification models based on support vector machine algorithm training.

[0067] In the embodiment of the present invention, the model generated by the support vector machine algorithm training can generally only be a binary medical image classification model. To more clearly illustrate the binary classification model, taking the component category classification provided by the present invention as an example, when the component category includes four possibilities: fat, soft tissue, air, and bone, the binary medical classification model includes (fat, soft tissue, air, bone), (soft tissue, fat, air, bone), (fat, soft tissue, air, bone), (fat, bone, soft tissue, air), and so on, where the two parts before and after the comma in each bracket are the two categories of results.

[0068] As a preferred embodiment of the present invention, the support vector machine algorithm adopts a soft margin support vector machine algorithm. Compared with the ordinary support vector machine algorithm, the soft margin support vector machine algorithm also has a good classification effect on data containing a small number of outliers.

[0069] In the embodiment of the present invention, the specific steps based on the support vector machine algorithm and the soft margin support vector machine algorithm and the calculation formula of the algorithm are specifically referred to in Figure 4 and its explanation.

[0070] Step S304 : determining the classification correctness probabilities of the multiple binary medical image classification models.

[0071] In an embodiment of the present invention, for different binary medical image classification models (please refer to the explanation of S302 above for details), the classification accuracy is different, that is, the classification effect is different. After training and generating the binary medical image classification model, it is generally necessary to test the accuracy of each binary medical image classification model through multiple test samples, that is, input multiple test sample input vectors into the binary medical image classification model to obtain response output results, and determine the accuracy of each binary medical image classification model based on the difference between the response output results and the actual results.

[0072] Step S306 : constructing a quaternary medical image classification model according to the correct classification probability and the multiple binary medical image classification models.

[0073] In an embodiment of the present invention, a quaternary medical classification model is constructed based on a decision tree algorithm according to multiple binary medical image classification models, wherein the binary medical image classification model with better reliability is placed in front, that is, the binary medical image classification model with the highest accuracy is set in the first layer, and then the two classified categories are further classified again until a quaternary classifier independently corresponds to fat, soft tissue, air and bone.

[0074] An embodiment of the present invention provides a method for generating a quaternary medical image classification model based on training generated by a hierarchical support vector machine algorithm. The method includes generating multiple binary medical image classification models based on training of the support vector machine algorithm, calculating the probability of correct classification for each binary medical image classification model, and then constructing a quaternary medical image classification model based on the binary medical image classification models. In the embodiment of the present invention, multiple binary medical image classification models are generated through training, and the more reliable binary medical image classification models are prioritized for classification based on the classification accuracy of each binary medical image classification model, thereby effectively ensuring the accuracy of the ultimately generated quaternary medical image classification model.

[0075] In one embodiment, Figure 4 As shown, the steps of the method for generating multiple binary medical image classification models based on support vector machine algorithm training specifically include:

[0076] Step S402: Acquire an MR scan sample image.

[0077] In the embodiment of the present invention, the MR scan sample image is a known MR image of a patient collected in advance.

[0078] Step S404: Process the MR scan sample image to obtain a plurality of mdixon sequence sample images of the MR scan sample image.

[0079] In an embodiment of the present invention, the plurality of mdixon sequence sample images include a fat image, a water image, a positive phase image, and a negative phase image.

[0080] In the embodiment of the present invention, step S404 is the same as the aforementioned step S104. Please refer to the explanation of the aforementioned step S104 for details.

[0081] Step S406 : determining feature information of each image point in the MR scan sample image according to the multiple mdixon sequence sample images.

[0082] In an embodiment of the present invention, the feature information includes the spatial coordinates of each image point and the MR value in a plurality of mdixon sequence sample images.

[0083] In the embodiment of the present invention, step S406 is the same as the aforementioned step S106 . Please refer to the explanation of the aforementioned step S106 for details.

[0084] Step S408 : obtaining the component category corresponding to each image point in the MR scan sample image.

[0085] In the embodiment of the present invention, since the MR scan sample image is a known MR image acquired in advance, that is, the component category corresponding to each image point in the image is known, the component category includes fat, soft tissue, air and bone.

[0086] Step S410 , based on the support vector machine algorithm and different component classification situations, different binary medical image classification models are determined according to the feature information and component categories.

[0087] In an embodiment of the present invention, the learning conditions of the support vector machine algorithm are:

[0088]

[0089] That is, when satisfied with y i (ω T x i +b)≥1, determine ω, b so that The value is the smallest. Among them, ω, b are the classification parameters in the binary classification model, x i is the input vector, y i is the output, y i Usually it takes the value 1 or -1 depending on the classification result.

[0090] As a preferred embodiment of the present invention, when the soft margin support vector machine algorithm is adopted, the learning condition of the soft margin support vector machine algorithm is:

[0091]

[0092] st…y i (ω T x i +b)≥1-ε i , ε i ≥0, i=1,2,...m

[0093] That is, when satisfied with y i (ω T x i +b)≥1-ε i Under the condition of i , making The value is the smallest. Among them, ω, b are the classification parameters in the binary classification model, ε i is an additional slack variable introduced by the soft margin support vector machine, x i is the input vector, y i is the target output result, y i It usually takes the value 1 or -1 according to the classification result. By introducing slack variables, the constraints on ω and b are relaxed, that is, outliers can also be effectively classified. However, at the same time, it is also necessary to set an additional penalty for outliers, that is, pre-set the penalty parameter C, C>0. The larger C is, the greater the penalty for outlier classification, and the smaller C is, the smaller the penalty for outlier classification.

[0094] The embodiment of the present invention discloses a method for generating multiple binary medical image classification models based on support vector machine algorithm training, and specifically provides the learning conditions of the support vector machine algorithm. It can use a small amount of training samples to train a binary medical image classification model with good classification effect, and the training process has a small amount of data processing and high training efficiency.

[0095] like Figure 5 As shown, in one embodiment, a simulated CT image generating device is provided, which may specifically include an MR scan image acquiring unit 510, an MR scan image processing unit 520, a feature information determining unit 530, a classification unit 540, a CT value determining unit 550 and a simulated CT image generating unit 560.

[0096] The MR scan image acquisition unit 510 is used to acquire an MR scan image to be converted into a simulated CT image.

[0097] In the embodiment of the present invention, a CT image is composed of a certain number of pixels of varying grayscales, ranging from black to white, arranged in a matrix. These pixels actually reflect the X-ray absorption coefficient of the corresponding voxel, i.e., the CT value. Therefore, it is only necessary to determine the CT value corresponding to each image point in the MR image to determine the corresponding simulated CT image.

[0098] The MR scan image processing unit 520 is configured to process the MR scan image to obtain a plurality of mdixon sequence images of the MR scan image.

[0099] In an embodiment of the present invention, the MDIXON is a fat suppression technology that utilizes a spin echo sequence to collect positive phase and reverse phase echo signals of water and fat protons at different echo times. By adding and subtracting the two non-passing signals, the effect of water and fat separation is achieved, that is, multiple MDIXON sequence images can be obtained, including fat images, water images, positive phase images, and reverse phase images.

[0100] The feature information determining unit 530 is configured to determine feature information of each image point in the MR scan image based on a plurality of mdixon sequence images.

[0101] In an embodiment of the present invention, the feature information includes the spatial coordinates of each image point and the MR value in a plurality of MDixon sequence images.

[0102] In an embodiment of the present invention, for each image point in the MR scan image, the spatial coordinates of the image point can be represented by a three-dimensional coordinate. At the same time, the MR value corresponding to the image point in each mdixon sequence image is combined with the spatial coordinates of the image point as the feature information of the image point.

[0103] The classification unit 540 is configured to determine the probability of the component category corresponding to each image point based on the feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm.

[0104] In an embodiment of the present invention, the component categories include fat, soft tissue, air, and bone. That is, after inputting a set of feature information, the probability of the image point being fat, soft tissue, air, or bone can be directly determined based on the feature information.

[0105] As a preferred embodiment of the present invention, since the classification model generated by the support vector machine algorithm training is used for binary classification, in order to realize the four-element medical image classification model, we use the hierarchical support vector machine algorithm for training, that is, first train and generate multiple binary medical image classification models, and then construct a decision tree of the multiple binary medical image classification models to generate a four-element medical image classification model.

[0106] The CT value determining unit 550 is configured to determine the CT value corresponding to each image point according to the probability of the component category corresponding to each image point.

[0107] In the embodiment of the present invention, the probability of the component category corresponding to each image point is used as a weight, and a weighted operation is performed on the standard CT value of each component category to determine the CT value of each image point.

[0108] The simulated CT image generating unit 560 is configured to generate a simulated CT image according to the CT value.

[0109] In the embodiment of the present invention, the CT value is generally between -1000Hu and 1000Hu. Among them, the common CT value of dense bone is 1000Hu, the CT value of water is 0Hu, and the CT value of air is -1000Hu. After knowing the CT value of each image point, a CT image can be simply determined. This CT image is a simulated CT image.

[0110] An embodiment of the present invention provides a simulated CT image generation device. This device processes an MR scan image into multiple MDIXON sequence images, including a fat map, a water map, a positive phase map, and a negative phase map. Feature information of each image point in the original MR scan image is then determined based on each sequence image. The feature information includes the spatial coordinates of each image point and its MR value in the multiple MDIXON sequence images. The device then determines the probability of a component category corresponding to each image point based on the feature information and a pre-trained, hierarchical support vector machine (HSVM) algorithm. The component categories include fat, soft tissue, air, and bone. The device then determines the corresponding CT value based on the probability of each component category, thereby directly determining a simulated CT image. The simulated CT image generation method provided by the embodiment of the present invention, upon inputting an MR scan image, can automatically process the MR image and directly output a corresponding CT image, effectively improving processing efficiency. Furthermore, the quaternary medical image classification model used in the processing is pre-trained using a large amount of sample data, effectively ensuring accuracy during the processing and improving the quality of the generated CT images.

[0111] like Figure 6 As shown, in one embodiment, the classification unit 530 specifically includes: a sample point determination module 631 , a sample classification module 632 , a reference point and distance determination module 633 , a weight determination module 634 and a remaining image point classification module 635 .

[0112] The sample point determination module 631 is configured to determine a plurality of sample image points in the MR scan image according to a preset rule.

[0113] In the embodiment of the present invention, the preset sampling rule may be simple random sampling, or fixed-point sampling may be used to ensure the comprehensiveness of the data.

[0114] In the embodiment of the present invention, preferably, the number of sample image points is about 20,000, which can effectively ensure the final image effect while reducing the amount of image point classification calculation.

[0115] The sample classification module 632 is used to determine the probability of the component category corresponding to each sample image point based on the feature information of each sample image point and the four-element medical image classification model generated by training based on the hierarchical support vector machine algorithm.

[0116] In the embodiment of the present invention, similarly, it is necessary to determine the probability of the component category corresponding to each sample image point, where the component categories include fat, soft tissue, air, and bone.

[0117] The reference point and distance determination module 633 is configured to determine, from the plurality of sample image points, a plurality of reference image points closest to the remaining image points in the MR scan image and to determine the distance between the remaining image points and the plurality of reference image points.

[0118] In the embodiment of the present invention, we use the K-nearest neighbor algorithm to calculate the component probability of the remaining image points, that is, we need to determine the K sample image points closest to the remaining image points as reference image points. Generally, the K value is preferably 5-7.

[0119] In the embodiment of the present invention, compared with the conventional K-nearest neighbor algorithm, the distance between each remaining image point and its K reference points is additionally obtained.

[0120] The weight determination module 634 is configured to determine the influence weights of the multiple reference image points on the respective remaining image points according to the distances between the respective remaining image points and the multiple reference image points.

[0121] In an embodiment of the present invention, compared to the conventional K-nearest neighbor algorithm that directly uses the mode of the attributes of K reference points to determine the attributes of the image point to be determined, the present invention also introduces weights to determine the attributes of the remaining image points, wherein the weights are determined based on the distance between each remaining image point and its reference image point, that is, the closer the distance is to the reference image point, the greater the influence factor.

[0122] The remaining image point classification module 635 is configured to determine the probability of the component category corresponding to each remaining image point based on the influence weights of the multiple reference image points on each remaining image point and the probability of the component category corresponding to each reference image point.

[0123] In an embodiment of the present invention, after determining the influencing factors of multiple reference image points and the probabilities of the component categories corresponding to the multiple reference points, the probabilities of the component categories corresponding to the remaining image points can be determined by weighted calculation, thus completing the classification of the component categories of each image point in the entire MR scan image.

[0124] A structural diagram of a classification unit provided by an embodiment of the present invention, by combining the K-nearest neighbor algorithm, first calculates the probabilities of component categories corresponding to some sample points, then determines the reference point based on the distances of the remaining image points from the sample points, and finally determines the probabilities of component categories corresponding to the remaining image points based on the probabilities of component categories corresponding to the reference point. Compared with calculating each image point one by one, this method effectively reduces the amount of data processing while also ensuring accuracy.

[0125] In one embodiment, a simulated CT image generating device is provided. Figure 5 The difference between the simulated CT image generation device shown in FIG. 1 and FIG. 2 is that it further includes a quaternary medical image classification model training generation unit 560, such as Figure 7 As shown, the four-element medical image classification model training generation unit 560 includes a binary medical image classification model training generation module 761 , a classification correct probability determination module 762 and a four-element medical image classification model construction module 763 .

[0126] The binary medical image classification model training and generation module 761 is used to generate multiple binary medical image classification models based on support vector machine algorithm training.

[0127] In the embodiment of the present invention, the model generated by the support vector machine algorithm training can generally only be a binary medical image classification model. To more clearly illustrate the binary classification model, taking the component category classification provided by the present invention as an example, when the component category includes four possibilities: fat, soft tissue, air, and bone, the binary medical classification model includes (fat, soft tissue, air, bone), (soft tissue, fat, air, bone), (fat, soft tissue, air, bone), (fat, bone, soft tissue, air), and so on, where the two parts before and after the comma in each bracket are the two categories of results.

[0128] As a preferred embodiment of the present invention, the support vector machine algorithm adopts a soft margin support vector machine algorithm. Compared with the ordinary support vector machine algorithm, the soft margin support vector machine algorithm also has a good classification effect on data containing a small number of outliers.

[0129] The classification correctness probability determination module 762 is used to determine the classification correctness probability of the multiple binary medical image classification models.

[0130] In an embodiment of the present invention, for different binary medical image classification models (please refer to the explanation of S761 above for details), the classification accuracy is different, that is, the classification effect is different. After training and generating the binary medical image classification model, it is generally necessary to test the accuracy of each binary medical image classification model through multiple test samples, that is, input multiple test sample input vectors into the binary medical image classification model to obtain response output results, and determine the accuracy of each binary medical image classification model based on the difference between the response output results and the actual results.

[0131] The four-element medical image classification model construction module 763 is used to construct a four-element medical image classification model according to the classification correct probability and the multiple binary medical image classification models.

[0132] In an embodiment of the present invention, a quaternary medical classification model is constructed based on a decision tree algorithm according to multiple binary medical image classification models, wherein the binary medical image classification model with better reliability is placed in front, that is, the binary medical image classification model with the highest accuracy is set in the first layer, and then the two classified categories are further classified again until a quaternary classifier independently corresponds to fat, soft tissue, air and bone.

[0133] An embodiment of the present invention provides a schematic structural diagram of a quaternary medical image classification model training and generation unit, which includes training and generating multiple binary medical image classification models based on a support vector machine algorithm, calculating the classification correctness probability of each binary medical image classification model, and then constructing a quaternary medical image classification model based on the binary medical image classification models. In this embodiment of the present invention, multiple binary medical image classification models are generated through training, and the more reliable binary medical image classification models are prioritized for classification based on the classification accuracy of each binary medical image classification model, effectively ensuring the accuracy of the ultimately generated quaternary medical image classification model.

[0134] like Figure 8 As shown, in one embodiment, the binary medical image classification model training generation module 761 specifically includes an MR scan sample image acquisition submodule 801, an MR scan sample image processing submodule 802, a feature information determination submodule 803, an MR scan sample image component category determination submodule 804, and a binary medical image classification model determination submodule 805.

[0135] The MR scan sample image acquisition submodule 801 is used to acquire MR scan sample images.

[0136] In the embodiment of the present invention, the MR scan sample image is a known MR image of a patient collected in advance.

[0137] The MR scan sample image processing submodule 802 is configured to process the MR scan sample image to obtain a plurality of mdixon sequence sample images of the MR scan sample image.

[0138] In an embodiment of the present invention, the plurality of mdixon sequence sample images include a fat image, a water image, a positive phase image, and a negative phase image.

[0139] In the embodiment of the present invention, the MR scan sample image processing submodule 802 has similar functions to the aforementioned MR scan image processing unit 520 . For details, please refer to the explanation of the aforementioned MR scan image processing unit 520 .

[0140] The feature information determination submodule 803 is configured to determine feature information of each image point in the MR scan sample image based on the multiple mdixon sequence sample images.

[0141] In an embodiment of the present invention, the plurality of mdixon sequence sample images include a fat image, a water image, a positive phase image, and a negative phase image.

[0142] In the embodiment of the present invention, the feature information determination submodule 803 has similar functions to the aforementioned feature information determination unit 630 . For details, please refer to the explanation of the aforementioned feature information determination unit 630 .

[0143] The MR scan sample image component category determination submodule 804 is configured to obtain the component category corresponding to each image point in the MR scan sample image.

[0144] In the embodiment of the present invention, since the MR scan sample image is a known MR image acquired in advance, that is, the component category corresponding to each image point in the image is known, the component category includes fat, soft tissue, air and bone.

[0145] The binary medical image classification model determination submodule 805 is used to determine different binary medical image classification models according to the feature information and component categories based on the support vector machine algorithm and different component category classification situations.

[0146] In an embodiment of the present invention, the learning conditions of the support vector machine algorithm are:

[0147]

[0148] That is, when satisfied with y i (ω T x i +b)≥1, determine ω, b so that The value is the smallest. Among them, ω, b are the classification parameters in the binary classification model, x iis the input vector, y i is the output, y i Usually it takes the value 1 or -1 depending on the classification result.

[0149] As a preferred embodiment of the present invention, when the soft margin support vector machine algorithm is adopted, the learning condition of the soft margin support vector machine algorithm is:

[0150]

[0151] st…y i (ω T x i +b)≥1-ε i , ε i ≥0, i=1,2,...m

[0152] That is, when satisfied with y i (ω T x i +b)≥1-ε i Under the condition of i , making The value is the smallest. Among them, ω, b are the classification parameters in the binary classification model, ε i is an additional slack variable introduced by the soft margin support vector machine, x i is the input vector, y i is the target output result, y i It usually takes the value 1 or -1 according to the classification result. By introducing slack variables, the constraints on ω and b are relaxed, that is, outliers can also be effectively classified. However, at the same time, it is also necessary to set an additional penalty for outliers, that is, pre-set the penalty parameter C, C>0. The larger C is, the greater the penalty for outlier classification, and the smaller C is, the smaller the penalty for outlier classification.

[0153] The embodiment of the present invention discloses a structural schematic diagram of a binary medical image classification model determination module, and specifically provides the learning conditions of the support vector machine algorithm. It can use a small number of training samples to train a binary medical image classification model with good classification effect, and the training process has a small amount of data processing and high training efficiency.

[0154] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are performed:

[0155] acquiring an MR scan image to be converted into a simulated CT image;

[0156] Processing the MR scan image to obtain a plurality of MDIXON sequence images of the MR scan image, wherein the plurality of MDIXON sequence images include a fat map, a water map, a positive phase map, and a negative phase map;

[0157] determining feature information of each image point in the MR scan image according to a plurality of mdixon sequence images, wherein the feature information includes a spatial coordinate of each image point and an MR value in the plurality of mdixon sequence images;

[0158] Determining the probability of a component category corresponding to each image point based on the feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm, wherein the component categories include fat, soft tissue, air, and bone;

[0159] Determining the CT value corresponding to each image point according to the probability of the component category corresponding to each image point;

[0160] A simulated CT image is generated according to the CT value.

[0161] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor performs the following steps:

[0162] acquiring an MR scan image to be converted into a simulated CT image;

[0163] Processing the MR scan image to obtain a plurality of MDIXON sequence images of the MR scan image, wherein the plurality of MDIXON sequence images include a fat map, a water map, a positive phase map, and a negative phase map;

[0164] determining feature information of each image point in the MR scan image according to a plurality of mdixon sequence images, wherein the feature information includes a spatial coordinate of each image point and an MR value in the plurality of mdixon sequence images;

[0165] Determining the probability of a component category corresponding to each image point based on the feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm, wherein the component categories include fat, soft tissue, air, and bone;

[0166] Determining the CT value corresponding to each image point according to the probability of the component category corresponding to each image point;

[0167] A simulated CT image is generated according to the CT value.

[0168] In order to intuitively demonstrate the technical performance of the above-mentioned simulated CT image generation method, an example is now completed based on the method disclosed in claim 1, using medical image data of volunteers recruited from Case Western Reserve University in the United States. During the experiment, experimental data of 8 patients were collected, and the average percentage error, root mean square error, and recall rate were compared with the traditional segmented simulated CT generation method. For specific experimental data, please refer to Figure 9 .

[0169] It should be understood that, although the various steps in the flow chart of each embodiment of the present invention are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in order, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0170] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0171] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

[0173] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for generating a simulated CT image, characterized in that: include: acquiring an MR scan image to be converted into a simulated CT image; Processing the MR scan image to obtain a plurality of MDIXON sequence images of the MR scan image, wherein the plurality of MDIXON sequence images include a fat map, a water map, a positive phase map, and a negative phase map; determining feature information of each image point in the MR scan image according to a plurality of mdixon sequence images, wherein the feature information includes a spatial coordinate of each image point and an MR value in the plurality of mdixon sequence images; Determining the probability of a component category corresponding to each image point based on the feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm, wherein the component categories include fat, soft tissue, air, and bone; Determining the CT value corresponding to each image point according to the probability of the component category corresponding to each image point; generating a simulated CT image according to the CT value; The method for generating the four-element medical image classification model generated by training based on the hierarchical support vector machine algorithm includes: Generate multiple binary medical image classification models based on support vector machine algorithm training; Determining correct classification probabilities of the multiple binary medical image classification models; Constructing a quaternary medical image classification model according to the classification correct probability and the multiple binary medical image classification models; The method for generating multiple binary medical image classification models based on support vector machine algorithm training includes: Acquire MR scan sample images; Processing the MR scan sample image to obtain a plurality of mdixon sequence sample images of the MR scan sample image, wherein the plurality of mdixon sequence sample images include a fat map, a water map, a positive phase map, and a negative phase map; determining feature information of each image point in the MR scan sample image according to the multiple mdixon sequence sample images, the feature information including the spatial coordinates of each image point and the MR value in the multiple mdixon sequence sample images; Obtaining a component category corresponding to each image point in the MR scan sample image; Based on the support vector machine algorithm and different component category classification situations, different binary medical image classification models are determined according to the feature information of each image point and the corresponding component category.

2. The method for generating a simulated CT image according to claim 1, wherein: The step of determining the probability of the component category corresponding to each image point based on the feature information of each image point and the four-element medical image classification model generated by training based on the hierarchical support vector machine algorithm specifically includes: determining a plurality of sample image points in the MR scan image according to a preset sampling rule; Determining the probability of the component category corresponding to each sample image point based on the feature information of each sample image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm; determining, from the plurality of sample image points, a plurality of reference image points closest to each remaining image point in the MR scan image and determining a distance between each remaining image point and the plurality of reference image points; determining, according to the distances between the respective remaining image points and the plurality of reference image points, influence weights of the plurality of reference image points on the respective remaining image points; The probability of the component category corresponding to each remaining image point is determined according to the influence weights of the multiple reference image points on the remaining image points and the probability of the component category corresponding to each reference image point.

3. A simulated CT image generating device, characterized in that: The device comprises: an MR scan image acquisition unit, configured to acquire an MR scan image to be converted into a simulated CT image; an MR scan image processing unit, configured to process the MR scan image to obtain a plurality of MDIXON sequence images of the MR scan image, wherein the plurality of MDIXON sequence images include a fat map, a water map, a positive phase map, and a negative phase map; a feature information determining unit, configured to determine feature information of each image point in the MR scan image based on a plurality of MDIXON sequence images, wherein the feature information includes a spatial coordinate of each image point and an MR value in the plurality of MDIXON sequence images; a classification unit, configured to determine a probability of a component category corresponding to each image point based on feature information of each image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm, wherein the component categories include fat, soft tissue, air, and bone; a CT value determining unit, configured to determine a CT value corresponding to each image point according to a probability of a component category corresponding to each image point; and a simulated CT image generating unit, configured to generate a simulated CT image according to the CT value; The CT image generation device further includes a four-element medical image classification model training generation unit, and the four-element medical image classification model training generation unit includes: A binary medical image classification model training and generation module, which is used to train and generate multiple binary medical image classification models based on the support vector machine algorithm; a classification correctness probability determination module, configured to determine the classification correctness probabilities of the plurality of binary medical image classification models; and A four-element medical image classification model construction module, configured to construct a four-element medical image classification model based on the classification correctness probability and the multiple binary medical image classification models; The binary medical image classification model training generation module includes: MR scan sample image acquisition submodule, used for acquiring MR scan sample images; an MR scan sample image processing submodule, configured to process the MR scan sample image to obtain a plurality of MDIXON sequence sample images of the MR scan sample image, wherein the plurality of MDIXON sequence sample images include a fat image, a water image, a positive phase image, and a negative phase image; a feature information determination submodule, configured to determine feature information of each image point in the MR scan sample image based on the multiple mdixon sequence sample images, the feature information including the spatial coordinates of each image point and the MR value in the multiple mdixon sequence sample images; An MR scan sample image component category determination submodule is used to obtain the component category corresponding to each image point in the MR scan sample image; The binary medical image classification model determination submodule is used to determine different binary medical image classification models according to the feature information and component categories based on the support vector machine algorithm and different component category classification situations.

4. The CT image generating device according to claim 3, wherein: The classification units include: a sample point determination module, configured to determine a plurality of sample image points in the MR scan image according to a preset rule; A sample classification module, configured to determine the probability of a component category corresponding to each sample image point based on the feature information of each sample image point and a four-element medical image classification model generated by training based on a hierarchical support vector machine algorithm; a reference point and distance determination module, configured to determine, from the plurality of sample image points, a plurality of reference image points closest to each of the remaining image points in the MR scan image and to determine a distance between each of the remaining image points and the plurality of reference image points; a weight determination module, configured to determine influence weights of the multiple reference image points on the respective remaining image points based on distances between the respective remaining image points and the multiple reference image points; and The remaining image point classification module is used to determine the probability of the component category corresponding to each remaining image point based on the influence weights of the multiple reference image points on each remaining image point and the probability of the component category corresponding to each reference image point.

5. A computer device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for generating a simulated CT image according to any one of claims 1 to 2.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method for generating a simulated CT image according to any one of claims 1 to 2.

Images