Ultrasonic sound velocity image generation method and system based on T1 weighted MRI image

Abstract

The invention relates to an ultrasonic sound velocity image generation method and system based on a T1 weighted MRI image. The method comprises the following steps: acquiring a T1 weighted MRI image and a CT image of a historical subject as sample data; constructing a deep learning model, wherein the deep learning model comprises a first-stage network and a second-stage network; training a first-stage network through the sample data, and enabling the first-stage network to generate a CT image when the T1 weighted MRI image is input; training a second-stage network through the sample data and the CT image generated by the first-stage network, and enabling the second-stage network to generate an ultrasonic sound velocity image when the T1 weighted MRI image and the CT image are input. According to the technical scheme, a high-quality ultrasonic sound velocity image can be effectively generated, ionizing radiation caused by CT scanning is avoided, and the method can be used for planning regulation and control of ultrasonic energy in sonodynamic therapy, so that the accuracy and efficiency of therapy are improved.

Description

Technical Field

[0001] The present application relates to the technical field of medical image generation, and particularly to a method and system for generating an ultrasonic sound velocity image based on a T1-weighted MRI image. Background Art

[0002] The ultrasonic sound velocity image reflects the characteristics of the tissue's ultrasonic wave conduction velocity and is an important parameter for applications such as ultrasonic imaging and ultrasonic therapy. In the sonodynamic therapy of brain tumors, the accurate ultrasonic sound velocity distribution in the skull is crucial for sound field simulation, energy focusing, and treatment parameter optimization. Inaccurate sound velocity information will lead to inaccurate energy focusing, affecting the treatment effect and even causing adverse reactions.

[0003] Traditional methods for obtaining the ultrasonic sound velocity in the skull mainly rely on computer tomography of the brain, that is, estimating the sound velocity at the corresponding skull position through the HU value of the scanned CT (Computed Tomography) image and obtaining the sound velocity value of the corresponding brain soft tissue through a T1-weighted MRI (Magnetic Resonance Imaging) image. However, the above methods rely on CT scans, which will bring unnecessary ionizing radiation to patients and result in high treatment costs. To avoid ionizing radiation, existing deep learning-based methods usually generate pseudo-CT images based on MRI and then obtain the corresponding sound velocity values using the pseudo-CT images. However, due to the error between the pseudo-CT image and the real CT image, the error of the sound velocity value will be further increased.

[0004] Therefore, how to accurately obtain the sound velocity value of the skull part from MRI images without increasing the additional radiation dose is an urgent problem to be solved in this field. Summary of the Invention

[0005] To at least overcome to some extent the problem that generating pseudo-CT images based on MRI in the related art will further increase the error of the sound velocity value, the present application provides a method and system for generating an ultrasonic sound velocity image based on a T1-weighted MRI image.

[0006] The solution of the present application is as follows:

[0007] According to the first aspect of the embodiments of the present application, a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image is provided, including:

[0008] Obtaining the T1-weighted MRI image and CT image of historical subjects as sample data;

[0009] Constructing a deep learning model, the deep learning model including a first-stage network and a second-stage network;

[0010] Train the first-stage network with the sample data so that when the first-stage network inputs a T1-weighted MRI image, it generates a CT image;

[0011] Train the second-stage network with the sample data and the CT image generated by the first-stage network so that when the second-stage network inputs a T1-weighted MRI image and a CT image, it generates an ultrasonic velocity image.

[0012] Preferably, the method further includes:

[0013] Obtain the T1-weighted MRI image of the subject;

[0014] Input the T1-weighted MRI image of the subject into the deep learning model;

[0015] Generate a CT image of the subject according to the T1-weighted MRI image of the subject through the first-stage network of the deep learning model;

[0016] Generate an ultrasonic velocity image of the subject according to the T1-weighted MRI image of the subject and the CT image of the subject generated by the first-stage network through the second-stage network of the deep learning model.

[0017] Preferably, after obtaining the T1-weighted MRI image and CT image of the historical subject, the method further includes:

[0018] Perform data desensitization on the T1-weighted MRI image and CT image of the historical subject.

[0019] Preferably, after obtaining the T1-weighted MRI image and CT image of the historical subject, the method further includes:

[0020] Perform resampling on the T1-weighted MRI image and CT image of the historical subject to unify the resolution of the T1-weighted MRI image and CT image of the historical subject;

[0021] Rigidly register the CT data of the historical subject to the resampled T1-weighted MRI data of the historical subject.

[0022] Preferably, it further includes:

[0023] Determine the ultrasonic velocity image of each pixel point as the true ultrasonic velocity image according to the rigidly registered CT data of the historical subject and the T1-weighted MRI data of the historical subject.

[0024] Preferably, training the first-stage network with the sample data includes:

[0025] Extract the features of the T1-weighted MRI image through a single-branch network;

[0026] Generate a simulated CT image based on the extracted features;

[0027] Identify the similarity between the generated simulated CT image and the real CT image; the real CT image is the CT image of a historical subject;

[0028] When the similarity between the generated simulated CT image and the real CT image is higher than a preset threshold, complete the training.

[0029] Preferably, training the second-stage network with the sample data and the CT image generated by the first-stage network includes:

[0030] Extract the features of the T1-weighted MRI image through the first branch of the two-branch fusion network;

[0031] Extract the features of the real CT image through the second branch of the two-branch fusion network;

[0032] Fuse the features extracted by the first branch and the second branch through an attention mechanism, and generate an ultrasonic sound velocity image according to the fused features.

[0033] Preferably, training the second-stage network with the sample data and the CT image generated by the first-stage network further includes:

[0034] Judge whether the network structures of the first-stage network and the second-stage network are stable;

[0035] If the network structures of the first-stage network and the second-stage network meet the preset stability conditions, use the CT image generated by the first-stage network as the input of the second branch of the two-branch fusion network, and enable the second branch of the two-branch fusion network to extract the features of the CT image generated by the first-stage network;

[0036] Fuse the features extracted by the first branch and the second branch through an attention mechanism, and generate an ultrasonic sound velocity image according to the fused features;

[0037] Identify the similarity between the generated ultrasonic sound velocity image and the real ultrasonic sound velocity image;

[0038] When the similarity between the generated ultrasonic sound velocity image and the real ultrasonic sound velocity image is higher than a preset threshold, complete the training.

[0039] Preferably, it further includes:

[0040] Evaluate the overall difference between the CT image generated by the first-stage network and the real CT image through the mean absolute error index;

[0041] Evaluate the overlapping degree of the skull regions between the CT image generated by the first-stage network and the real CT image through the Jaccard coefficient and the Dice coefficient;

[0042] Evaluate the overall difference between the ultrasonic sound speed image generated by the second-stage network through MAE and the real ultrasonic sound speed image.

[0043] According to the second aspect of the embodiments of the present application, there is provided an ultrasonic sound speed image generation system based on T1-weighted MRI images, including:

[0044] A processor and a memory;

[0045] The processor is connected to the memory through a communication bus:

[0046] Wherein, the processor is configured to call and execute the program stored in the memory;

[0047] The memory is configured to store a program, and the program is at least used to execute an ultrasonic sound speed image generation method based on T1-weighted MRI images as described in any one of the above.

[0048] The technical solution provided by the present application may include the following beneficial effects:

[0049] In this technical solution, the T1-weighted MRI images and CT images of historical subjects are used as sample data to form a training data set. By using the existing paired data, additional CT scans of patients are avoided, thereby reducing the risk of ionizing radiation, and at the same time ensuring that the training data contains the true sound speed characteristic information of the skull and soft tissues. The deep learning model is divided into a first-stage network and a second-stage network, where the first-stage network is used to generate CT images, and the second-stage network is used to generate ultrasonic sound speed images. The first stage: Generate CT images through T1-weighted MRI, making full use of the soft tissue information of MRI and the advantages of CT in bone imaging to provide an accurate anatomical reference for subsequent sound speed mapping; The second stage: Generate ultrasonic sound speed images based on T1-weighted MRI and CT images, realizing the fusion of two-modal information, effectively generating high-quality ultrasonic sound speed images, avoiding the ionizing radiation caused by CT scans, and can be used to plan the regulation of ultrasonic energy in sonodynamic therapy, thereby improving the accuracy and efficiency of treatment.

[0050] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0052] Figure 1It is a schematic flowchart of a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application;

[0053] Figure 2 It is a schematic diagram of the network structure of a deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application;

[0054] Figure 3 It is a schematic diagram of the GF structure of a deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application;

[0055] Figure 4 It is a schematic diagram of the network structure of the second stage of a deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application;

[0056] Figure 5 They are CT images and ultrasonic sound velocity images generated by a deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application;

[0057] Figure 6 It is a simulation diagram of an acoustic dynamic therapy sound field in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application;

[0058] Figure 7 It is a schematic diagram of the structure of a system for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application.

[0059] Reference numerals: Processor - 21; Memory - 22. Detailed implementation manners

[0060] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0061] Embodiment 1

[0062] A method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image includes:

[0063] S11: Obtain the T1-weighted MRI image and CT image of historical subjects as sample data;

[0064] It should be noted that after obtaining the T1-weighted MRI images and CT images of historical subjects, the method further includes:

[0065] Performing data desensitization on the T1-weighted MRI images and CT images of historical subjects.

[0066] In specific practice, cranial T1-weighted MRI data and CT data from multiple (such as 100) subjects are collected. These data are obtained using a standard scanning protocol in a clinical environment. And these data need to be anonymized to ensure patient privacy.

[0067] Brain T1-weighted MRI data uses magnetic resonance imaging technology. By selecting specific echo time (TE) and repetition time (TR), the T1 relaxation time differences of various tissues in the image are prominently displayed. This imaging method mainly reflects the recovery rates of fat and water molecules in tissues, thus being able to clearly distinguish soft tissue structures. It is mainly used for detailed imaging of brain structures, disease diagnosis (such as tumors, brain atrophy, etc.) and surgical planning.

[0068] The T1-weighted MRI images in this embodiment are obtained based on a 3.0T magnetic resonance scanner, using a T1-weighted sequence. The scanning parameters include: TR (repetition time) = 500 ms, TE (echo time) = 10 ms, field of view = 220 mm x 220 mm, matrix size = 256 x 256, slice thickness = 1 mm.

[0069] CT imaging reconstructs three-dimensional images through the projection data of X-rays at multiple angles, which are processed by a computer. It is commonly used in examinations such as craniocerebral trauma, fractures, calcified lesions, and angiography.

[0070] The CT images in this embodiment are obtained based on a multi-slice spiral CT scanner. The scanning parameters include: tube voltage = 120 kVp, tube current = 200 mA, slice thickness = 1 mm.

[0071] It should be noted that after obtaining the T1-weighted MRI images and CT images of historical subjects, the method further includes:

[0072] Resampling the T1-weighted MRI images and CT images of historical subjects to unify the resolutions of the T1-weighted MRI images and CT images of historical subjects;

[0073] Rigidly registering the CT data of historical subjects to the resampled T1-weighted MRI data of historical subjects.

[0074] In specific practice, resampling is performed using linear interpolation or trilinear interpolation on MRI and CT images, resampling them to a voxel size of 1mm x 1mm x 1mm to make them have the same resolution, and the size of all slice images is 256x256.

[0075] Rigidly register the CT data to the resampled T1-weighted MRI data. Mutual information is used as the similarity metric for registration, and the Iterative Closest Point (ICP) algorithm or an optimization algorithm based on gradient descent is used for registration.

[0076] Preprocessing the T1-weighted MRI and CT data to make them have the same resolution, and rigidly registering the CT data to the resampled T1-weighted MRI data is mainly to ensure that the two-modal data are completely spatially aligned, so as to achieve accurate pixel-level mapping and multi-modal information fusion. The specific reasons are as follows:

[0077] The acquisition parameters, pixel sizes, and resolutions of MRI and CT may vary. If the resolutions are not unified, the same anatomical structures between different modalities may appear as different sizes or scales on the images, resulting in no one-to-one correspondence between pixels, which in turn affects subsequent deep learning model training and image fusion. When the resolutions are consistent, the deep learning model can capture the same structural information at the same scale when extracting features, which helps the model better learn the mapping relationship between the two modalities. Unifying the resolutions is an important step in data preprocessing, which can reduce the interference caused by data sources, acquisition devices, and parameter differences, thereby improving the robustness of the model.

[0078] Due to factors such as the patient's position and device angle during acquisition, different imaging modalities (MRI and CT) often have geometric differences such as translation and rotation in space. Through rigid registration, these translation and rotation errors can be corrected, so that the same anatomical structures in the two images can accurately overlap. Rigid registration ensures that the bone and calcification regions in the CT image correspond spatially with the soft tissue structures in the MRI image, providing an accurate anatomical reference for subsequent use of CT information to guide pseudo-CT generation and ultrasound sound speed mapping. In deep learning tasks, especially pixel-level image generation and conversion, the accurate alignment of input data is a prerequisite for ensuring model performance. The registered data can significantly reduce noise and errors, helping the model more effectively learn the mapping relationship between different modalities.

[0079] It should be noted that it also includes:

[0080] Based on the CT data of historical subjects and the T1-weighted MRI data of historical subjects after rigid registration, determine the ultrasound sound speed image of each pixel point as the true ultrasound sound speed image.

[0081] By combining the information of the registered CT image and the T1-weighted MRI image, the ultrasonic sound velocity value corresponding to each pixel point is accurately determined, providing real comparison data for the verification link in the subsequent model training.

[0082] S12: Construct a deep learning model, where the deep learning model includes a first-stage network and a second-stage network;

[0083] Figure 2 It is a schematic diagram of the network structure of the deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application; Figure 3 It is a schematic diagram of the GF structure of the deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application; Figure 4 It is a schematic diagram of the network structure of the second stage of the deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided by an embodiment of the present application.

[0084] As Figure 2 shown, the first-stage network refers to the Stage One block diagram in Figure 2 , and the second-stage network refers to the Stage Two block diagram in Figure 2 .

[0085] For the specific structure of the second-stage network, refer to Figure 4 .

[0086] S13: Train the first-stage network with sample data so that when the first-stage network inputs a T1-weighted MRI image, it generates a CT image;

[0087] Specifically, training the first-stage network with sample data includes:

[0088] Extract the features of the T1-weighted MRI image through a single-branch network;

[0089] Generate a simulated CT image based on the extracted features;

[0090] Discriminate the similarity between the generated simulated CT image and the real CT image; the real CT image is the CT image of historical subjects;

[0091] When the similarity between the generated simulated CT image and the real CT image is higher than a preset threshold, the training is completed.

[0092] Specifically, the first-stage network is used to generate simulated CT images (pseudo-CT images). The generator is constructed using a U-Net structure, which includes skip connections to preserve local anatomical features. Specifically, it includes a generator G and a discriminator D. Among them, the generator G adopts an encoder-decoder structure. The encoder consists of multiple convolutional layers and pooling layers, which are used to extract the features of the MRI image. The decoder consists of multiple deconvolutional layers and upsampling layers, which are used to generate pseudo-CT images. The generator uses ReLU as the activation function and uses the Sigmoid function in the last layer to limit the pixel value range of the pseudo-CT image to [0,1]. The discriminator D consists of five convolutional layers and activation functions, which are used to judge whether the input image is a synthetic image or a real image, and to distinguish the generated pseudo-CT image from the real CT image. Using a convolutional neural network structure, it outputs a probability value indicating the probability that the input image is a real CT image.

[0093] The generated pseudo-CT image and the real CT image of the historical subject are input into the discriminator D of the first-stage network, and a loss function is set to output the value of the loss function. Based on the value of the loss function, gradient descent is performed and optimized through multiple iterations to obtain the optimal generator for synthesizing CT images.

[0094] S14: Train the second-stage network with the sample data and the CT images generated by the first-stage network, so that when the second-stage network inputs the T1-weighted MRI image and the CT image, it generates an ultrasonic sound speed image.

[0095] Specifically, training the second-stage network with the sample data and the CT images generated by the first-stage network includes:

[0096] Extract the features of the T1-weighted MRI image through the first branch of the two-branch fusion network;

[0097] Extract the features of the real CT image through the second branch of the two-branch fusion network;

[0098] Fuse the features extracted by the first branch and the second branch through the attention mechanism, and generate an ultrasonic sound speed image according to the fused features.

[0099] Judge whether the network structures of the first-stage network and the second-stage network are stable;

[0100] If the network structures of the first-stage network and the second-stage network meet the preset stability conditions, use the CT image generated by the first-stage network as the input of the second branch of the two-branch fusion network, so that the second branch of the two-branch fusion network extracts the features of the CT image generated by the first-stage network;

[0101] Fuse the features extracted by the first branch and the second branch through the attention mechanism, and generate an ultrasonic sound speed image according to the fused features;

[0102] Identify the similarity between the generated ultrasonic velocity image and the real ultrasonic velocity image;

[0103] When the similarity between the generated ultrasonic velocity image and the real ultrasonic velocity image is higher than a preset threshold, the training is completed.

[0104] The second stage is used to generate ultrasonic velocity images. The network in the second stage adopts a U-Net structure to form a two-branch fusion network generator. Each branch adopts an independent downsampling encoder and a unified upsampling decoder after bottom-layer fusion. Extract the features of the T1-weighted MRI image through the first downsampling branch of the two-branch fusion network; extract the features of the generated CT image through the second downsampling branch of the two-branch fusion network.

[0105] In the U-Net structure generator, the features extracted by the first branch and the second branch are fused by using a cross-attention mechanism at the skip connection and the bottom layer of the network. The calculation method is: element-wise add the attention-weighted CT features and the original MRI features to achieve inter-modal feature calibration and fusion, and finally generate the ultrasonic velocity image.

[0106] The generator GF structure in this stage is a two-branch fusion structure, as Figure 3 shown. Among them, the encoder structure of the MRI branch is used to extract the features of the MRI image, and the pseudo-CT branch structure is used to extract the features of the pseudo-CT image. The two-branch encoders are similar but have different parameters. After the encoder, an attention mechanism is used to fuse the features of the MRI branch and the pseudo-CT branch. The attention mechanism structure is as Figure 4 shown. The features after virtual-real fusion are connected to the decoder to output the generated ultrasonic velocity image. The discriminator DF in this part is used to distinguish the generated ultrasonic velocity image from the real ultrasonic velocity image.

[0107] The discriminator in the second stage is a patch discriminator, which consists of five convolutional layers and activation functions, and is used to judge whether the input image is a synthetic image or a real image.

[0108] Input the generated ultrasonic velocity image and the real ultrasonic velocity image into the discriminator in the second stage, set the loss function, and output the loss function value. According to the loss function value, perform gradient descent and optimize through multiple iterations to obtain the optimal generator of the ultrasonic velocity image.

[0109] It should be noted that in this technical solution, after the network structures in the two stages reach stability, serial fine-tuning is performed. Replace the real CT image in the training process of the second stage with the CT image generated in the first stage.

[0110] In this technical solution, the deep learning model is divided into a first-stage network and a second-stage network. The first-stage network is used to generate CT images, and the second-stage network is used to generate ultrasonic velocity images.

[0111] The goal of the first-stage network is to make the generated simulated CT images as similar as possible to the real CT images.

[0112] The goal of the second-stage network is to be able to generate ultrasonic velocity images based on the input MRI images of the subject and the simulated CT images generated by the first-stage network.

[0113] During the training process, the standard loss functions of the generative adversarial network are adopted, including the generator loss and the discriminator loss.

[0114] In the first-stage network, the mean absolute error metric is used to evaluate the overall difference between the CT images generated by the first-stage network and the real CT images;

[0115] The Jaccard coefficient and the Dice coefficient are used to evaluate the overlapping degree of the skull regions in the CT images generated by the first-stage network and the real CT images.

[0116] In the second-stage network, MAE is used to evaluate the difference between the generated ultrasonic velocity images and the real ultrasonic velocity images.

[0117] In this technical solution, the T1-weighted MRI images and CT images of historical subjects are used as sample data to form a training dataset. By using the existing paired data, additional CT scans of patients are avoided, thereby reducing the risk of ionizing radiation. At the same time, it is ensured that the training data contains the real sound velocity characteristic information of the skull and soft tissues. The deep learning model is divided into a first-stage network and a second-stage network. The first-stage network is used to generate CT images, and the second-stage network is used to generate ultrasonic velocity images. First stage: Generate CT images through T1-weighted MRI, making full use of the soft tissue information of MRI and the advantages of CT in bone imaging to provide an accurate anatomical reference for subsequent sound velocity mapping; Second stage: Generate ultrasonic velocity images based on T1-weighted MRI and CT images, realizing the fusion of two-modal information, effectively generating high-quality ultrasonic velocity images, avoiding the ionizing radiation brought by CT scans, and can be used to plan the regulation of ultrasonic energy in sonodynamic therapy, thereby improving the accuracy and efficiency of treatment.

[0118] It should be noted that after the model training is completed, it can be put into use. Based on this, the method further includes:

[0119] Obtain the T1-weighted MRI image of the subject;

[0120] Input the T1-weighted MRI image of the subject into the deep learning model;

[0121] The first-stage network of the deep learning model generates a CT image of the subject based on the T1-weighted MRI image of the subject;

[0122] The second-stage network of the deep learning model generates an ultrasonic sound velocity image of the subject based on the T1-weighted MRI image of the subject and the CT image of the subject generated by the first-stage network.

[0123] Figure 5 The CT image and the ultrasonic sound velocity image are generated by the deep learning model in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided in an embodiment of the present application. Figure 6 It is a simulation diagram of an acoustic dynamic therapy sound field used in a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image provided in an embodiment of the present application. Refer to Figure 5 - Figure 6 After the trained deep learning model inputs the T1-weighted MRI image of the subject, it can automatically generate the ultrasonic sound velocity image of the subject, avoiding additional CT scans for the patient, thereby reducing the risk of ionizing radiation, and at the same time ensuring that the training data contains the true sound velocity characteristic information of the skull and soft tissues.

[0124] Embodiment 2

[0125] An ultrasonic sound velocity image generation system based on a T1-weighted MRI image. Refer to Figure 7 It includes:

[0126] A processor 21 and a memory 22;

[0127] The processor 21 is connected to the memory 22 through a communication bus:

[0128] Among them, the processor 21 is used to call and execute the program stored in the memory 22;

[0129] The memory 22 is used to store the program, and the program is at least used to execute a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image as in the above embodiment.

[0130] It can be understood that the same or similar parts in the above embodiments can be referred to each other, and the content not detailed in some embodiments can be seen in the same or similar content in other embodiments.

[0131] It should be noted that in the description of the present application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present application, unless otherwise specified, the meaning of "a plurality" refers to at least two.

[0132] Any process or method description depicted in a flowchart or otherwise described herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process. The scope of the preferred embodiments of the present application includes additional implementations where functions may be performed in a substantially simultaneous manner or in a reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0133] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well-known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0134] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0135] In addition, in each embodiment of the present application, the functional units can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0136] The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disc, etc.

[0137] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0138] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A method for generating an ultrasonic sound velocity image based on T1-weighted MRI images, characterized in that, Including: Obtaining T1-weighted MRI images and CT images of historical subjects as sample data; Constructing a deep learning model, the deep learning model including a first-stage network and a second-stage network; Training the first-stage network with the sample data, such that when the first-stage network inputs a T1-weighted MRI image, it generates a CT image; Training the second-stage network with the sample data and the CT image generated by the first-stage network, such that when the second-stage network inputs a T1-weighted MRI image and a CT image, it generates an ultrasonic sound velocity image.

2. The method according to claim 1, characterized in that, The method further includes: Obtaining a T1-weighted MRI image of a subject; Inputting the T1-weighted MRI image of the subject into the deep learning model; Generating a CT image of the subject by the first-stage network of the deep learning model according to the T1-weighted MRI image of the subject; Generating an ultrasonic sound velocity image of the subject by the second-stage network of the deep learning model according to the T1-weighted MRI image of the subject and the CT image of the subject generated by the first-stage network.

3. The method according to claim 1, wherein After obtaining the T1-weighted MRI images and CT images of historical subjects, the method further includes: Performing data desensitization on the T1-weighted MRI images and CT images of historical subjects.

4. The method according to claim 1, characterized in that After obtaining the T1-weighted MRI images and CT images of historical subjects, the method further includes: Performing resampling on the T1-weighted MRI images and CT images of historical subjects to unify the resolution of the T1-weighted MRI images and CT images of historical subjects; Rigidly registering the CT data of historical subjects to the resampled T1-weighted MRI data of historical subjects.

5. The method according to claim 4, wherein Further including: Determining the ultrasonic sound velocity image of each pixel point as the true ultrasonic sound velocity image according to the rigidly registered CT data of historical subjects and the T1-weighted MRI data of historical subjects.

6. The method according to claim 5, wherein Training the first-stage network with the sample data includes: Extracting features of the T1-weighted MRI image through a single-branch network; Generating a simulated CT image according to the extracted features; Identifying the similarity between the generated simulated CT image and the true CT image; the true CT image is the CT image of historical subjects; When the similarity between the generated simulated CT image and the true CT image is higher than a preset threshold, completing the training.

7. The method according to claim 6, characterized in that, Training the second-stage network with the sample data and the CT image generated by the first-stage network includes: Extracting features of the T1-weighted MRI image through the first branch of a two-branch fusion network; Extracting features of the true CT image through the second branch of the two-branch fusion network; Fusing the features extracted by the first branch and the second branch through an attention mechanism and generating an ultrasonic sound velocity image according to the fused features.

8. The method according to claim 7, wherein Training the second-stage network with the sample data and the CT image generated by the first-stage network further includes: Judging whether the network structures of the first-stage network and the second-stage network are stable; If the network structures of the first-stage network and the second-stage network reach the preset stable conditions, the CT image generated by the first-stage network is used as the input of the second branch of the two-branch fusion network, so that the second branch of the two-branch fusion network extracts the features of the CT image generated by the first-stage network; The features extracted by the first branch and the second branch are fused through an attention mechanism, and an ultrasonic sound velocity image is generated according to the fused features; Discriminate the similarity between the generated ultrasonic sound velocity image and the real ultrasonic sound velocity image; When the similarity between the generated ultrasonic sound velocity image and the real ultrasonic sound velocity image is higher than the preset threshold, the training is completed.

9. The method according to claim 1, characterized in that, It further includes: Evaluating the overall difference between the CT image generated by the first-stage network and the real CT image through the mean absolute error index; Evaluating the overlapping degree of the skull regions in the CT image generated by the first-stage network and the real CT image through the Jaccard coefficient and the Dice coefficient; Evaluating the overall difference between the ultrasonic sound velocity image generated by the second-stage network and the real ultrasonic sound velocity image through MAE.

10. An ultrasonic sound velocity image generation system based on T1-weighted MRI images, characterized in that, It includes: A processor and a memory; The processor is connected to the memory through a communication bus: Wherein, the processor is used to call and execute the program stored in the memory; The memory is used to store the program, and the program is at least used to execute a method for generating an ultrasonic sound velocity image based on a T1-weighted MRI image according to any one of claims 1-9.

Images