Abdominal MRI image segmentation system, method, electronic device and storage medium

Abstract

Embodiments of the present invention disclose an abdominal MRI image segmentation system, method, electronic device, and storage medium. The abdominal MRI image segmentation method includes: constructing a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image; constructing a VAT segmentation model and a SAT segmentation model; inputting the training set into the VAT segmentation model and the SAT segmentation model for training, respectively, to obtain trained VAT segmentation models and SAT segmentation models; segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result; and performing fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result. This abdominal MRI image segmentation method solves the problem in the prior art that abdominal MRI image segmentation training requires a large number of manually labeled MRI images.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an abdominal MRI image segmentation system, method, electronic device and storage medium. Background Art

[0002] Obesity has been shown to be positively correlated with the risk of developing a variety of diseases. Quantitative analysis of abdominal adipose tissue (visceral adipose tissue, VAT) and subcutaneous adipose tissue (SAT) can help better understand and assess a patient's condition. Currently, several non-invasive imaging techniques are available to identify adipose tissue (SAT and VAT), such as computed tomography (CT), dual-energy X-ray absorptiometry (DEXA), and magnetic resonance imaging (MRI). DEXA is currently the most widely used method for monitoring body fat, but it does not provide any anatomical information and cannot segment VAT and SAT.

[0003] MRI is a safer imaging modality than CT because it does not use potentially harmful ionizing radiation. Although MRI is safer and can provide richer and more accurate anatomical information, segmenting adipose tissue in MRI images is significantly more difficult than in CT images. The Hounsfield unit (HU)-based threshold segmentation method, currently widely used in CT images, cannot be directly applied to adipose tissue segmentation in MRI images.

[0004] Traditional manual methods require professionally trained physicians to draw the boundaries of the entire abdominal region and delineate the VAT and SAT regions separately. This manual method requires a significant amount of manual effort and time and is highly subjective. Semi-automatic methods use adjustable thresholds, boundary enhancement, and other image processing tools to help observers identify adipose tissue. Compared with manual segmentation methods, this method is relatively time-saving and labor-saving, but it still takes a professionally trained physician an average of 5-10 minutes to process one slice. An entire abdominal scan typically consists of 15 to 50 slices, which, when processed using semi-automatic segmentation methods, generates a large amount of image processing tasks. Both manual and semi-automatic methods require manual intervention by the physician, which is not only time-consuming and labor-intensive, but also leads to differences between and within observers due to subjective factors, thereby undermining the reliability of the data.

[0005] Developing a high-precision automatic segmentation algorithm by using appropriate feature selection and machine learning techniques (such as deep learning) can save doctors' manual effort and time, but such algorithm training will require a large number of manually labeled MRI images.

[0006] There is an urgent need for a method to train an automated segmentation model for VAT and SAT without manual labeling of MRI images. Summary of the Invention

[0007] The purpose of the embodiments of the present invention is to provide an abdominal MRI image segmentation system, method, electronic device and storage medium to solve the problem in the prior art that abdominal MRI image segmentation requires a large number of manually labeled MRI images during training.

[0008] To achieve the above objectives, an embodiment of the present invention provides an abdominal MRI image segmentation method, the method specifically comprising:

[0009] Obtain abdominal MRI images;

[0010] Constructing a recurrent generative adversarial network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent generative adversarial network, and constructing a training set based on the synthetic CT image;

[0011] Construct VAT segmentation model and SAT segmentation model;

[0012] Inputting the training set into the VAT segmentation model and the SAT segmentation model for training respectively, to obtain a trained VAT segmentation model and a trained SAT segmentation model;

[0013] Segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result;

[0014] Based on the segmentation result, a fat volume quantitative analysis is performed on the abdominal MRI image to obtain a fat volume quantitative analysis result.

[0015] On the basis of the above technical solution, the present invention can also be improved as follows:

[0016] Furthermore, the constructing of a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image includes:

[0017] The cyclic adversarial generative network includes a first generator, a second generator, a first discriminator and a second discriminator;

[0018] Based on the first generator from I B Generate I BA, based on the second generator from I A Generate I AB ;

[0019] Based on the first discriminator from I BA Identification I A , based on the second discriminator from I AB Identification I B ;

[0020] Wherein, the I A is an MRI image, the I B is a CT image, the I AB To synthesize MRI images, I BA For synthetic CT images.

[0021] Furthermore, the constructing of a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image includes:

[0022] performing abdominal tissue mask calculation on the abdominal MRI image;

[0023] Performing intensity inhomogeneity correction on the abdominal MRI image after abdominal tissue mask calculation;

[0024] Perform image clustering processing on the abdominal MRI images after intensity inhomogeneity correction;

[0025] The composite CT image is obtained by performing image conversion processing on the abdominal MRI image after image clustering processing.

[0026] Furthermore, the performing abdominal tissue mask calculation on the abdominal MRI image includes:

[0027] Based on the iterative region growing algorithm, abdominal tissue voxels are separated from air background and unimportant limbs to obtain new voxels;

[0028] Calculate the average value μ of the air background voxels b and standard deviation σ b ;

[0029] Calculate new voxel values ​​for new voxels;

[0030] Determine whether the new voxel value is less than μ b +3σ b , when the new voxel value is less than μ b +3σ b When the new voxel value is greater than μ b +3σ bWhen the new voxel is determined as tissue, the image morphology algorithm is used to obtain the largest tissue area representing the abdomen.

[0031] Furthermore, the performing intensity non-uniformity correction on the abdominal MRI image after the abdominal tissue mask calculation includes:

[0032] The observed inhomogeneous MRI image is calculated by Equation 1;

[0033] I b =I c B+n Formula 1;

[0034] Among them, I b is the observed inhomogeneous MRI image, B is the bias field with intensity variation, I c is the corrected image, n is the additional noise;

[0035] The bias field is corrected based on the local entropy minimization method by using a higher b The voxels with the mean intensity are fitted with a 2D polynomial function to I b To predict the initial bias field B0;

[0036] Calculate the initial correction image by formula 2;

[0037] I c 0=I b / B0 formula 2;

[0038] Among them, I c 0 is the initial correction image, B0 is the initial bias field, I c 0. Start entropy optimization of the first region R1 from the node with the highest B0 value;

[0039] The entropy optimization process of the second region R2 is started at the node with the second highest B0 value, and the corrected I in R2 is determined. c The average intensity value of 0 is the same as the corrected I in R1 c 0 are similar. If so, R1 and R2 are merged before entropy optimization. Otherwise, R2 is subjected to separate entropy optimization.

[0040] Furthermore, the image clustering process is performed on the abdominal MRI image after intensity inhomogeneity correction, including:

[0041] The MRI images after intensity inhomogeneity correction were automatically classified into abdominal adipose tissue and abdominal non-fat tissue using the fuzzy c-means algorithm with Euclidean norm.

[0042] Furthermore, the step of performing fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result includes:

[0043] quantifying the abdominal fat tissue according to the volume data, selecting a middle slice as a starting slice in an MRI sequence, and using a bias field calculated from the starting slice to provide initial values ​​for the remaining slices;

[0044] The active contours calculated in each slice were used as initial contours in adjacent slices, and VAT and SAT were calculated from one slice by multiplying the number of voxels by the voxel volume;

[0045] After all slices were processed, the VAT and SAT of each slice were stacked according to the slice position and thickness to construct a 3D image representing the total volume of abdominal VAT and SAT, and the total measurement of VAT, SAT, and TAT was obtained by summing the fat volume of each slice in the sequence.

[0046] An abdominal MRI image segmentation system, comprising:

[0047] An acquisition module, used for acquiring abdominal MRI images;

[0048] The first building block is used to construct a recurrent adversarial generative network;

[0049] Converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent generative adversarial network;

[0050] A second construction module is used to construct a training set based on the synthetic CT image;

[0051] The third building module is used to build a VAT segmentation model and a SAT segmentation model;

[0052] A training module, configured to input the training set into the VAT segmentation model and the SAT segmentation model for training, thereby obtaining a trained VAT segmentation model and a trained SAT segmentation model;

[0053] Segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result;

[0054] An analysis module is configured to perform a fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result.

[0055] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the method are implemented when the processor executes the computer program.

[0056] A non-transitory computer-readable storage medium stores a computer program, which implements the steps of the method when executed by a processor.

[0057] The embodiments of the present invention have the following advantages:

[0058] The abdominal MRI image segmentation method of the present invention obtains an abdominal MRI image;

[0059] A recurrent adversarial generative network is constructed, and based on the recurrent adversarial generative network, the abdominal MRI image is converted into an equivalent synthetic CT image, and a training set is constructed based on the synthetic CT image; a VAT segmentation model and a SAT segmentation model are constructed; the training set is input into the VAT segmentation model and the SAT segmentation model for training, respectively, to obtain a trained VAT segmentation model and a SAT segmentation model; the abdominal MRI image is segmented based on the trained VAT segmentation model and the SAT segmentation model to obtain a segmentation result; and fat volume quantitative analysis is performed on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result. This solves the problem in the prior art that abdominal MRI image segmentation requires a large number of manually labeled MRI images during training. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0061] The structures, proportions, sizes, etc. illustrated in this specification are intended only to complement the contents disclosed herein and to facilitate understanding and reading by persons familiar with the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall still fall within the scope of the technical contents disclosed herein.

[0062] Figure 1 Flowchart of the abdominal MRI image segmentation method of the present invention;

[0063] Figure 2 This is an architecture diagram of the abdominal MRI image segmentation system of the present invention;

[0064] Figure 3 Schematic diagram of converting the abdominal MRI image into an equivalent synthetic CT image based on a recurrent generative adversarial network according to the present invention;

[0065] Figure 4 Schematic diagram of the VAT segmentation model and the SAT segmentation model of the present invention;

[0066] Figure 5 This is a schematic diagram of the results of subcutaneous and visceral fat segmentation according to the present invention;

[0067] Figure 6 This is a schematic diagram of the visualization results of abdominal fat volume reconstructed using MRI slices in the present invention.

[0068] Figure 7 This is a schematic diagram of the physical structure of the electronic device provided by the present invention.

[0069] The accompanying drawings are denoted as follows:

[0070] Acquisition module 10 , first construction module 20 , second construction module 30 , third construction module 40 , training module 50 , analysis module 60 , electronic device 70 , processor 701 , memory 702 , bus 703 . DETAILED DESCRIPTION

[0071] The following describes the implementation of the present invention using specific embodiments. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. Obviously, the embodiments described are only a portion of the present invention, not all of it. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0072] Example

[0073] Figure 1 FIG. 1 is a flow chart of an embodiment of the abdominal MRI image segmentation method of the present invention, as shown in FIG. Figure 1 As shown, an abdominal MRI image segmentation method provided by an embodiment of the present invention includes the following steps:

[0074] S101, acquiring an abdominal MRI image;

[0075] Specifically, magnetic resonance imaging (MRI) is a new examination technology that uses the principle that atomic nuclei with magnetic moments can produce transitions between energy levels under the action of a magnetic field.

[0076] MRI is achieved through the action of an external high-frequency magnetic field, in which substances in the body radiate energy to the surrounding environment to generate signals. The imaging process is similar to image reconstruction and CT, except that MRI does not rely on external radiation, absorption and reflection, nor on gamma radiation from radioactive substances in the body. Instead, it uses the interaction between the external magnetic field and the object to form an image. The high-energy magnetic field is harmless to the human body.

[0077] Compared with CT, the main advantages of MRI are:

[0078] ① Ionizing radiation causes no radioactive damage or biological damage to brain tissue.

[0079] ②It can directly produce cross-sectional, sagittal, coronal and various oblique tomographic images.

[0080] ③ There are no artifacts such as radiation hardening in CT images.

[0081] ④ It is not affected by bone images and can satisfactorily display small lesions in the posterior cranial fossa and brainstem. It has high diagnostic value for the top of the skull, parasagittal sinus, lateral fissure structures and widely metastatic tumors.

[0082] ⑤ It can show the pathological process of the disease more extensively and with clearer structure than CT. It can detect isodense lesions that appear completely normal on CT, and is particularly useful for detecting demyelinating diseases, encephalitis, infectious demyelination, ischemic lesions, and low-grade gliomas.

[0083] S102, constructing a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image;

[0084] Specifically, compared to CT images, it is very difficult to segment fat tissue in MRI images. The threshold segmentation method based on Hounsfield units (HU), which is currently widely used in CT images, cannot be directly applied to fat tissue segmentation in MRI images.

[0085] Therefore, to convert abdominal MRI images into equivalent synthetic CT images;

[0086] The cyclic adversarial generative network includes a first generator, a second generator, a first discriminator and a second discriminator:

[0087] Based on the first generator from I B Generate I BA , based on the second generator from I A Generate I AB ;

[0088] Based on the first discriminator from I BA Identification I A , based on the second discriminator from I AB Identification I B ;

[0089] Wherein, the I A is an MRI image, the I B is a CT image, the I AB To synthesize MRI images, IBA For synthetic CT images.

[0090] performing abdominal tissue mask calculation on the abdominal MRI image;

[0091] Performing intensity inhomogeneity correction on the abdominal MRI image after abdominal tissue mask calculation;

[0092] Perform image clustering processing on the abdominal MRI images after intensity inhomogeneity correction;

[0093] The composite CT image is obtained by performing image conversion processing on the abdominal MRI image after image clustering processing.

[0094] The calculating of the abdominal tissue mask on the abdominal MRI image comprises:

[0095] Based on the iterative region growing algorithm, abdominal tissue voxels are separated from air background and unimportant limbs to obtain new voxels;

[0096] Calculate the average value μ of the air background voxels b and standard deviation σ b ;

[0097] Calculate new voxel values ​​for new voxels;

[0098] Determine whether the new voxel value is less than μ b +3σ b , when the new voxel value is less than μ b +3σ b When the new voxel value is greater than μ b +3σ b When the new voxel is determined as tissue, the image morphology algorithm is used to obtain the largest tissue area representing the abdomen.

[0099] An image morphology algorithm is used to close the image and fill the holes, measure the size of each tissue region in the image, remove the smaller regions containing the limbs, and retain only the largest tissue region representing the abdomen. Figure 3 a shows a T1-weighted abdominal MRI slice with redundant limbs on both sides; Figure 3 b shows the abdominal tissue mask; Figure 3 c shows an image with the abdominal tissue preserved after the limbs have been removed. This mask limits the computational area of ​​the image processing step, speeding up the computation and reducing image artifacts or noise caused by torso outside the abdomen.

[0100] The performing intensity non-uniformity correction on the abdominal MRI image after the abdominal tissue mask calculation includes:

[0101] The spatial inhomogeneity of MRI coil sensitivity and main magnetic field leads to the presence of non-uniform signal intensity artifacts on MRI slices. The bias field model that causes this non-uniformity can be summarized as Equation 1:

[0102] The observed inhomogeneous MRI image is calculated by Equation 1;

[0103] I b =I c B+n formula 1;

[0104] Among them, I b is the observed inhomogeneous MRI image, B is the bias field with intensity variation, I c is the corrected image, n is the additional noise; the original T1-weighted MRI image ( Figure 3 a) Has noticeably non-uniform intensity, and if not corrected, such variations within the same tissue (i.e., fat or muscle) can often mislead the segmentation process.

[0105] The bias field is corrected based on the local entropy minimization method by using a higher b The voxels with the mean intensity are fitted with a 2D polynomial function to I b To predict the initial bias field B0;

[0106] Calculate the initial correction image by formula 2;

[0107] I c 0=I b / B0 formula 2;

[0108] Among them, I c 0 is the initial correction image, B0 is the initial bias field, I c 0. Start entropy optimization of the first region R1 from the node with the highest B0 value;

[0109] The entropy optimization process of the second region R2 is started at the node with the second highest B0 value, and the corrected I in R2 is determined. c The average intensity value of 0 is corrected by I in R1 c 0 are similar, if so, R1 and R2 are merged before entropy optimization, otherwise, R2 is subjected to separate entropy optimization. This adaptive segmented optimization process continues until all nodes are processed, and finally the corrected image I is formed. c The bias field correction based on the LEMS method improves the uniformity of the same tissue in MRI images and prevents overcorrection of non-fat areas ( Figure 3 d).

[0110] The image clustering process is performed on the abdominal MRI image after intensity inhomogeneity correction, comprising:

[0111] The MRI images after intensity inhomogeneity correction were automatically classified into abdominal adipose tissue and abdominal non-fat tissue using the fuzzy c-means algorithm with Euclidean norm.

[0112] The intensity-corrected MRI images are automatically classified into adipose tissue and non-adipose tissue using the fuzzy c-means (FCM) algorithm with the Euclidean norm. The energy function to be minimized during the FCM iteration is defined as:

[0113]

[0114] Where N is the total number of voxels in the MRI image, K is the number of clusters (K = 3, i.e., fat tissue, non-fat tissue, and background), and y i is the voxel intensity, c k is the central strength of the K-th cluster.

[0115] In the FCM algorithm, the fuzzy membership function and cluster centers The updates are defined as follows:

[0116]

[0117]

[0118] in, is the voxel intensity y at the nth iteration i The fuzzy membership probability of belonging to category k, is the updated category center strength value of category k in the nth iteration, b is a user-defined parameter and 0<b<1. The above function is iteratively optimized by updating the center strength of each cluster. When stability is reached, the iteration stops. Figure 3 e- Figure 3 h shows the clustering results of the MRI image after intensity inhomogeneity correction.

[0119] Performing image conversion processing on the abdominal MRI image after image clustering processing to obtain the synthetic CT image includes:

[0120] The C-GAN architecture is built on the GAN model and can achieve dual-domain image conversion between images from domains A (MRI) and B (CT). C-GAN uses two generators (G A From I B Generate I BA , G B From I A Generate I AB ) and two discriminators (D AFrom I BA Identification I A , D B From I AB Identification I B ). The loss function of the generator is defined using mean square error (MSE):

[0121] G loss =MSE(1,D B (I AB ))+MSE(1,D A (I BA ))+α[MSE(I ABA , I A )+MSE(I BAB , I B )]+β[MSE(I BA , I B )+MSE(I AB , I A )]

[0122] Wherein α and β are optimized to 10.0 and 2.0 respectively. A is an MRI image, the I B is a CT image, the I AB To synthesize MRI images, I BA For the synthetic CT image, I ABA and I BAB Represents the image generated after C-GAN processing.

[0123] S103, constructing a VAT segmentation model and a SAT segmentation model;

[0124] S104: Input the training set into the VAT segmentation model and the SAT segmentation model for training, thereby obtaining a trained VAT segmentation model and a trained SAT segmentation model.

[0125] Specifically, the synthetic CT images are divided into training set, test set and validation set;

[0126] The VAT segmentation model and the SAT segmentation model are trained based on the training set; the performance of the VAT segmentation model and the SAT segmentation model are respectively verified based on the validation set, and the improved VB-Net neural network model that meets the performance conditions is saved; and the segmentation results of the VAT segmentation model and the SAT segmentation model are evaluated based on the test set.

[0127] S105 , segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result.

[0128] Specifically, the VAT and SAT segmentation models are trained based on the U-Net algorithm (e.g. Figure 4 As shown in Figure 2, we use the active contour model to segment VAT and SAT by locating the abdominal muscle wall (i.e., the target contour) and minimizing the image energy. Assume that the image μ0 is formed by two regions of piecewise constant intensity approximation and have different values and Further assume that the object to be detected is represented by the value The area is represented by C0, and its boundary contour is represented by C0. Then inside C0: And outside C0: Use the following formula for fitting:

[0129] F1(C)+F2(C)=∫ inside (C)|μ0(x, y)-c1| 2 dxdy+∫ outside(C) |μ0(x,y)-c2| 2 dx dy

[0130] Where C is any other variable curve, constants c1 and c2 are the average values ​​of μ0 inside and outside C, respectively, then the boundary contour C0 is the minimum value of the fitting function:

[0131] The abdominal contour obtained in the previous region growing iteration is used as the initial contour. When the contour is stable or the maximum number of iterations is reached, the iteration stops automatically, and the final contour after optimization can enclose all the visceral regions. The final segmentation result is shown in Figure 5 As shown, the segmentation results are overlaid on the rectified MRI image.

[0132] S106 , performing fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result.

[0133] Specifically, the abdominal fat tissue is quantified according to the volume data, and in an MRI sequence, a middle slice is selected as a starting slice, and a bias field calculated using the starting slice is used to provide initial values ​​for the remaining slices;

[0134] The active contours calculated in each slice were used as initial contours in adjacent slices, and VAT and SAT were calculated from one slice by multiplying the number of voxels by the voxel volume;

[0135] After all slices were processed, the VAT and SAT of each slice were stacked according to the slice position and thickness to construct a 3D image representing the total volume of abdominal VAT and SAT, and the total measurement of VAT, SAT, and TAT was obtained by summing the fat volume of each slice in the sequence. Figure 6 a. Figure 6 b and Figure 6c shows the SAT, VAT, and TAT volumes reconstructed from multiple slices from an MRI sequence, respectively.

[0136] The abdominal MRI image segmentation method of the present invention comprises the following steps: acquiring an abdominal MRI image; constructing a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image; constructing a VAT segmentation model and a SAT segmentation model; inputting the training set into the VAT segmentation model and the SAT segmentation model for training, respectively, to obtain trained VAT segmentation models and SAT segmentation models; segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result; and performing fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result. The method solves the problem in the prior art that abdominal MRI image segmentation requires a large number of manually labeled MRI images during training.

[0137] This patent invents a deep learning algorithm for automated quantitative segmentation of adipose tissue in abdominal MRI images, which can segment and quantitatively analyze adipose tissue more efficiently, saving manual effort and time. Manual segmentation of VAT and SAT in an MRI slice takes about 5-10 minutes, while using an automated deep learning algorithm only takes 20 to 103 seconds. In the quantitative analysis of adipose tissue, manual operation requires about 1.5 hours to analyze all slices in an MRI sequence, but using an automated deep learning algorithm to process an MRI sequence only takes 10 to 20 minutes. Therefore, the fully automatic algorithm based on deep learning invented by this patent greatly improves the efficiency of segmenting adipose tissue from MRI slices. In addition, the automated algorithm does not require manual operation and any prior information, which can avoid the impact of subjective differences in manual operation on the segmentation results, thereby improving the robustness of adipose tissue segmentation and quantitative analysis results.

[0138] Figure 2 FIG. 1 is a flow chart of an embodiment of an abdominal MRI image segmentation system according to the present invention; FIG. Figure 2 As shown, an abdominal MRI image segmentation system provided by an embodiment of the present invention includes the following steps:

[0139] An acquisition module 10 is used to acquire an abdominal MRI image;

[0140] A first construction module 20 is used to construct a cyclic adversarial generative network;

[0141] Converting the abdominal MRI image into an equivalent synthetic CT image based on the cyclic adversarial generative network; the cyclic adversarial generative network includes a first generator, a second generator, a first discriminator, and a second discriminator;

[0142] Based on the first generator from I B Generate I BA , based on the second generator from I A Generate I AB ;

[0143] Based on the first discriminator from I BA Identification I A , based on the second discriminator from I AB Identification I B ;

[0144] Wherein, the I A is an MRI image, the I B is a CT image, the I AB To synthesize MRI images, I BA For synthetic CT images.

[0145] performing abdominal tissue mask calculation on the abdominal MRI image;

[0146] Performing intensity inhomogeneity correction on the abdominal MRI image after abdominal tissue mask calculation;

[0147] Perform image clustering processing on the abdominal MRI images after intensity inhomogeneity correction;

[0148] The composite CT image is obtained by performing image conversion processing on the abdominal MRI image after image clustering processing.

[0149] Based on the iterative region growing algorithm, abdominal tissue voxels are separated from air background and unimportant limbs to obtain new voxels;

[0150] Calculate the average value μ of the air background voxels b and standard deviation σ b ;

[0151] Calculate new voxel values ​​for new voxels;

[0152] Determine whether the new voxel value is less than μ b +3σ b , when the new voxel value is less than μ b +3σ b When the new voxel value is greater than μ b +3σ b When the new voxel is determined as tissue, the image morphology algorithm is used to obtain the largest tissue area representing the abdomen.

[0153] The observed inhomogeneous MRI image is calculated by Equation 1;

[0154] Ib =I c B+n Formula 1;

[0155] Among them, I b is the observed inhomogeneous MRI image, B is the bias field with intensity variation, I c is the corrected image, n is the additional noise;

[0156] The bias field is corrected based on the local entropy minimization method by using a higher b The voxels with the mean intensity are fitted with a 2D polynomial function to I b To predict the initial bias field B0;

[0157] Calculate the initial correction image by formula 2;

[0158] I c 0=I b / B0 formula 2;

[0159] Among them, I c 0 is the initial correction image, B0 is the initial bias field, I c 0. Start entropy optimization of the first region R1 from the node with the highest B0 value;

[0160] The entropy optimization process of the second region R2 is started at the node with the second highest B0 value, and the corrected I in R2 is determined. c The average intensity value of 0 is the same as the corrected I in R1 c 0 are similar. If so, R1 and R2 are merged before entropy optimization. Otherwise, R2 is subjected to separate entropy optimization.

[0161] The MRI images after intensity inhomogeneity correction were automatically classified into abdominal adipose tissue and abdominal non-fat tissue using the fuzzy c-means algorithm with Euclidean norm.

[0162] A second construction module 30 is configured to construct a training set based on the synthetic CT image;

[0163] The third construction module 40 is used to construct a VAT segmentation model and a SAT segmentation model;

[0164] A training module 50 is configured to input the training set into the VAT segmentation model and the SAT segmentation model for training, thereby obtaining a trained VAT segmentation model and a trained SAT segmentation model;

[0165] Segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result;

[0166] An analysis module 60 is configured to perform a fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result. The abdominal fat tissue is quantified based on the volume data. In the MRI sequence, a middle slice is selected as a starting slice, and a bias field calculated from the starting slice is used to provide initial values ​​for the remaining slices.

[0167] The active contours calculated in each slice were used as initial contours in adjacent slices, and VAT and SAT were calculated from one slice by multiplying the number of voxels by the voxel volume;

[0168] After all slices were processed, the VAT and SAT of each slice were stacked according to the slice position and thickness to construct a 3D image representing the total volume of abdominal VAT and SAT, and the total measurement of VAT, SAT, and TAT was obtained by summing the fat volume of each slice in the sequence.

[0169] The abdominal MRI image segmentation system of the present invention acquires an abdominal MRI image through an acquisition module 10; constructs a recurrent adversarial generative network through a first construction module 20; converts the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network; constructs a training set based on the synthetic CT image through a second construction module 30; constructs a VAT segmentation model and a SAT segmentation model through a third construction module 40; inputs the training set into the VAT segmentation model and the SAT segmentation model respectively through a training module 50 for training to obtain trained VAT segmentation model and SAT segmentation model; segments the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result; and performs fat volume quantitative analysis on the abdominal MRI image based on the segmentation result through an analysis module 60 to obtain a fat volume quantitative analysis result. The system solves the problem in the prior art that abdominal MRI image segmentation requires a large number of manually labeled MRI images during training.

[0170] Figure 7 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as Figure 7 As shown, the electronic device 70 includes: a processor 701 (processor), a memory 702 (memory) and a bus 703;

[0171] The processor 701 and the memory 702 communicate with each other via the bus 703.

[0172] The processor 701 is used to call program instructions in the memory 702 to execute the methods provided by the above-mentioned method embodiments, for example, including: acquiring an abdominal MRI image; constructing a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image; constructing a VAT segmentation model and a SAT segmentation model; inputting the training set into the VAT segmentation model and the SAT segmentation model for training, respectively, to obtain a trained VAT segmentation model and a SAT segmentation model; segmenting the abdominal MRI image based on the trained VAT segmentation model and the SAT segmentation model to obtain a segmentation result; and performing fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result.

[0173] This embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions. The computer instructions enable a computer to execute the methods provided by the above-mentioned method embodiments, for example, including: acquiring an abdominal MRI image; constructing a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image; constructing a VAT segmentation model and a SAT segmentation model; inputting the training set into the VAT segmentation model and the SAT segmentation model for training, respectively, to obtain trained VAT segmentation models and SAT segmentation models; segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result; and performing fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result.

[0174] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk, etc. Various storage media that can store program codes.

[0175] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0176] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or certain parts of the embodiment.

[0177] Although the present invention has been described in detail above using general descriptions and specific embodiments, it will be apparent to those skilled in the art that modifications and improvements may be made thereto. Therefore, such modifications and improvements, without departing from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. A method for segmenting abdominal MRI images, characterized in that: The method comprises: Obtain abdominal MRI images; Constructing a recurrent adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent adversarial generative network, and constructing a training set based on the synthetic CT image, including: performing abdominal tissue mask calculation on the abdominal MRI image; Performing intensity inhomogeneity correction on the abdominal MRI image after abdominal tissue mask calculation; Image clustering processing is performed on the abdominal MRI image after intensity inhomogeneity correction, including: The MRI images after intensity inhomogeneity correction were automatically classified into abdominal fat tissue and abdominal non-fat tissue using the fuzzy c-means algorithm with Euclidean norm. Performing image conversion processing on the abdominal MRI image after image clustering processing to obtain the synthetic CT image; The calculating of the abdominal tissue mask on the abdominal MRI image comprises: Based on the iterative region growing algorithm, abdominal tissue voxels are separated from air background and unimportant limbs to obtain new voxels; Calculate the average value of air background voxels and standard deviation ; Calculate new voxel values ​​for new voxels; Determine whether the new voxel value is less than , when the new voxel value is less than When the new voxel value is greater than , determining the new voxel as tissue, and using an image morphology algorithm to obtain the largest tissue area representing the abdomen; Construct VAT segmentation model and SAT segmentation model; Inputting the training set into the VAT segmentation model and the SAT segmentation model for training respectively, to obtain a trained VAT segmentation model and a trained SAT segmentation model; Segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result; Performing fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result, including: quantifying the abdominal fat tissue according to the volume data, selecting a middle slice as a starting slice in an MRI sequence, and using a bias field calculated from the starting slice to provide initial values ​​for the remaining slices; The active contours calculated in each slice were used as initial contours in adjacent slices, and VAT and SAT were calculated from one slice by multiplying the number of voxels by the voxel volume; After all slices were processed, the VAT and SAT of each slice were stacked according to the slice position and thickness to construct a 3D image representing the total volume of abdominal VAT and SAT, and the total measurement of VAT, SAT, and TAT was obtained by summing the fat volume of each slice in the sequence.

2. The abdominal MRI image segmentation method according to claim 1, characterized in that: The step of constructing a cyclic adversarial generative network, converting the abdominal MRI image into an equivalent synthetic CT image based on the cyclic adversarial generative network, and constructing a training set based on the synthetic CT image comprises: The cyclic adversarial generative network includes a first generator, a second generator, a first discriminator and a second discriminator; Based on the first generator from generate , based on the second generator from generate ; Based on the first discriminator Identification , based on the second discriminator from Identification ; Among them, the For MRI images, the For CT images, the To synthesize MRI images, For synthetic CT images.

3. The abdominal MRI image segmentation method according to claim 2, characterized in that: The performing intensity non-uniformity correction on the abdominal MRI image after the abdominal tissue mask calculation includes: The observed inhomogeneous MRI image was calculated by Equation 1; Formula 1: in, is the observed inhomogeneous MRI image, is a bias field with varying intensity, is the corrected image, is the additional noise; The bias field is corrected based on the local entropy minimization method by using a higher The voxels with the mean intensity are fitted with a 2D polynomial function to To predict the initial bias field ; Calculate the initial correction image by formula 2; Formula 2: in, is the initial correction image, is the initial bias field, Starting from the node with the highest value, the entropy optimization process of the first region R1 is performed; exist The entropy optimization process of the second region R2 is started at the node with the second highest value, and the corrected value in R2 is determined. The average intensity value is corrected by R1 If so, R1 and R2 are merged before entropy optimization. Otherwise, R2 is subjected to separate entropy optimization.

4. An abdominal MRI image segmentation system, characterized in that: include: An acquisition module, used for acquiring abdominal MRI images; The first building block is used to construct a recurrent adversarial generative network; Converting the abdominal MRI image into an equivalent synthetic CT image based on the recurrent generative adversarial network; A second construction module is used to construct a training set based on the synthetic CT image; The second building block is further configured to: performing abdominal tissue mask calculation on the abdominal MRI image; Performing intensity inhomogeneity correction on the abdominal MRI image after abdominal tissue mask calculation; Image clustering processing is performed on the abdominal MRI image after intensity inhomogeneity correction, including: The MRI images after intensity inhomogeneity correction were automatically classified into abdominal fat tissue and abdominal non-fat tissue using the fuzzy c-means algorithm with Euclidean norm. Performing image conversion processing on the abdominal MRI image after image clustering processing to obtain the synthetic CT image; The calculating of the abdominal tissue mask on the abdominal MRI image comprises: Based on the iterative region growing algorithm, abdominal tissue voxels are separated from air background and unimportant limbs to obtain new voxels; Calculate the average value of air background voxels and standard deviation ; Calculate new voxel values ​​for new voxels; Determine whether the new voxel value is less than , when the new voxel value is less than When the new voxel value is greater than , determining the new voxel as tissue, and using an image morphology algorithm to obtain the largest tissue area representing the abdomen; The third building module is used to build a VAT segmentation model and a SAT segmentation model; A training module, configured to input the training set into the VAT segmentation model and the SAT segmentation model for training, thereby obtaining a trained VAT segmentation model and a trained SAT segmentation model; Segmenting the abdominal MRI image based on the trained VAT segmentation model and SAT segmentation model to obtain a segmentation result; an analysis module, configured to perform a fat volume quantitative analysis on the abdominal MRI image based on the segmentation result to obtain a fat volume quantitative analysis result; The analysis module is also used for: quantifying the abdominal fat tissue according to the volume data, selecting a middle slice as a starting slice in an MRI sequence, and using a bias field calculated from the starting slice to provide initial values ​​for the remaining slices; The active contours calculated in each slice were used as initial contours in adjacent slices, and VAT and SAT were calculated from one slice by multiplying the number of voxels by the voxel volume; After all slices were processed, the VAT and SAT of each slice were stacked according to the slice position and thickness to construct a 3D image representing the total volume of abdominal VAT and SAT, and the total measurement of VAT, SAT, and TAT was obtained by summing the fat volume of each slice in the sequence.

5. An electronic device 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 steps of the method according to any one of claims 1 to 3 are implemented.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images