Modal Conversion Method for Artificial Intelligence-Based Quantitative Magnetic Resonance Imaging

Abstract

The present invention relates to the field of penetration testing security, and relates to a method for modal conversion of quantitative magnetic resonance imaging based on artificial intelligence, including data acquisition and fitting; data generation; data preprocessing; establishing a deep neural model. The present invention uses deep learning technology to achieve quantitative MRI modal conversion of conventional magnetic resonance scanning sequence modal data with arbitrary different contrasts as input. By introducing magnetic resonance sequence parameter encoding through a deep learning network, the present invention can achieve efficient conversion of conventional MRI images with arbitrary contrasts to quantitative MRI parameter maps, significantly reducing the time required for multiple scans and complex fitting calculations in traditional methods. Compared with the prior art, the acquisition efficiency of quantitative MRI data is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and relates to a method for modal conversion of quantitative magnetic resonance imaging based on artificial intelligence. Background Art

[0002] Quantitative magnetic resonance imaging (qMRI) technology has been widely used in clinical and research fields in recent years, especially playing an important role in disease diagnosis, disease monitoring, and brain function research. qMRI provides quantitative information of brain tissue under different physical parameters by generating quantitative images such as T1 relaxation time (T1map), T2 relaxation time (T2map), and proton density map (PDmap). These information can reveal detailed pathological and physiological characteristics that are difficult to present in traditional MRI images.

[0003] In terms of disease diagnosis, qMRI can be used to distinguish different types of tumors and accurately evaluate the size and boundary of tumors. In the early diagnosis of stroke, qMRI can identify ischemic and hemorrhagic lesion areas and quantify the degree of brain tissue damage. In addition, for chronic or acute neurological diseases such as multiple sclerosis and brain infections, qMRI provides important basis for disease monitoring and treatment effect evaluation by quantitatively analyzing the number, size, and distribution of lesions.

[0004] In the diagnosis and treatment process of children with congenital heart disease, preoperative brain development assessment is crucial because these patients usually face a high risk of neurodevelopmental abnormalities. qMRI can provide objective biomarkers for clinicians by quantifying the relaxation times of T1 and T2 in brain tissue, so as to identify and monitor changes in brain development.

[0005] The implementation of existing qMRI technology relies on multiple data acquisitions of the same anatomical structure under different pulse sequence parameters (such as inversion time TI, echo time TE, flip angle, etc.). The quantitative parameter map is calculated by voxel nonlinear fitting based on the corresponding signal model. However, this multiple acquisition method leads to a significant extension of the scanning time, increasing the cost of MRI examination and the demand for computing resources. For patients who cannot remain stationary for a long time, the feasibility of this acquisition method is also limited.

[0006] Therefore, it is necessary to develop an artificial intelligence technology based on magnetic resonance imaging principle to optimize the modal conversion method and system of quantitative MRI, so as to solve the problems of long scanning time and poor patient compliance existing in traditional qMRI, and thus improve the application efficiency and accuracy of qMRI technology in clinical and research. Summary of the Invention

[0007] To overcome the many limitations of traditional qMRI techniques in clinical applications, the present invention proposes a method for converting modalities of quantitative magnetic resonance imaging based on artificial intelligence. The present invention combines deep learning techniques with physical models to achieve fast and efficient conversion of quantitative parameters for magnetic resonance images with different contrasts.

[0008] The present invention is achieved through the following technical solutions:

[0009] A method for converting modalities of quantitative magnetic resonance imaging based on artificial intelligence, comprising the following steps:

[0010] S1. Data acquisition and fitting: Use a magnetic resonance device to scan a target object, collect modality data with different sequence parameters, and obtain quantitative MRI image data of the target object by fitting according to physical formulas;

[0011] S2. Data generation: Generate conventional magnetic resonance scan sequence modality data with different sequence scan parameters based on the formula for synthesizing quantitative MRI image data;

[0012] S3. Data preprocessing: Preprocess the conventional magnetic resonance scan sequence modality data through unified resolution, extreme value removal, Z-Score normalization, and data augmentation to ensure data consistency and standardization;

[0013] S4. Establish a deep neural model: Use deep learning techniques to achieve quantitative MRI modality conversion for input conventional magnetic resonance scan sequence modality data with arbitrary different contrasts.

[0014] Further, the quantitative MRI image data in step S1 includes multi-dimensional diffusion weighted imaging modality data such as T1map, T2map, and PDmap.

[0015] Further, the sequence scan parameters in step S2 include repetition time TR, echo time TE, and inversion time IR.

[0016] Further, the data augmentation in step S3 includes foreground cropping, background minimum setting, random flipping, and resolution unification.

[0017] Further, in step S4, a deep neural model is established using a U-Net convolutional neural network. The neural network includes multiple convolutional layers, pooling layers, upsampling layers, skip connection layers, and a parameter encoding module. The parameter encoding module is used to encode the sequence scan parameters, and the sequence scan parameters are physical parameters.

[0018] Further, step S4 specifically includes the following sub-steps:

[0019] S401. Encode the sequence scanning parameters using a U-Net convolutional neural network, map the encoded physical parameters to a high-dimensional embedding space through an independent fully connected layer to generate physical parameter embeddings, add the physical parameter embeddings and the convolutional feature maps of the corresponding modalities in the channel dimension to integrate the physical parameter information into the feature maps, and process the added feature maps through a convolutional layer;

[0020] S402. The neural network uses a multi-branch structure to process input images with different contrasts. Each branch encodes physical parameters by introducing a parameter embedding layer and fuses them with the branch feature maps; after the branch features are concatenated and channel features are fused in the intermediate layer, multi-scale features are combined in the upsampling stage, and corresponding quantitative magnetic resonance images are generated through different convolutional layers.

[0021] Furthermore, the encoding module realizes the fusion processing of physical parameters and image features through an MLP, so that the output quantitative magnetic resonance image has physical consistency.

[0022] Furthermore, the convolutional layer used in step S401 is a 1×1 convolutional layer.

[0023] Advantages of the present invention:

[0024] (1) A modality conversion method for quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention can realize the efficient conversion of conventional MRI images with arbitrary contrast to quantitative MRI parameter maps by introducing magnetic resonance sequence parameter encoding through a deep learning network, significantly reducing the time required for multiple scans and complex fitting calculations in traditional methods. Compared with the prior art, the acquisition efficiency of quantitative MRI data is greatly improved;

[0025] (2) A modality conversion method for quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention does not require the cumbersome multiple scan process in traditional qMRI technology. Only relying on conventional MRI images of single or a few scans, high-precision quantitative parameter maps can be generated through modality conversion, reducing the time cost and computational resource consumption of MRI examinations, and improving the comfort and examination compliance of patients;

[0026] (3) A modality conversion method for quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention realizes the deep fusion of physical parameters and image features by introducing a physical parameter encoding module and combining the characteristics of a deep learning network. The generated quantitative MRI images not only have high resolution, but also can ensure physical consistency and accuracy, meeting the high requirements for data accuracy in clinical and scientific research;

[0027] (4)The modal conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention uses a multi-branch U-Net convolutional neural network that supports conventional MRI images with different contrasts as inputs, has strong adaptability to complex clinical scenarios, and can generate multiple quantitative MRI parameter images through the synergistic effect of the branch structure and the physical parameter encoding module, realizing the comprehensive utilization of multi-modal information;

[0028] (5)The modal conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention significantly improves the standardization degree of the input data through unified resolution, extreme value removal, Z-Score normalization, and data augmentation in the data preprocessing link, ensuring the reliability and consistency of the modal conversion results;

[0029] (6)The modal conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention uses a 1×1 convolutional layer and an MLP module to achieve the fusion of physical parameters and image features, making the output quantitative MRI images have higher robustness and accuracy while maintaining high quality, meeting the requirements in various scenarios;

[0030] (7)The modal conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention has important application values in the fields of brain tumor diagnosis, early detection of stroke, brain development assessment, and disease monitoring of multiple sclerosis, etc. It can provide objective and accurate imaging biomarkers for clinical diagnosis and treatment, and at the same time provide high-quality data support for scientific research work. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for description in the embodiments. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0032] Figure 1 It is a flowchart of the modal conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention;

[0033] Figure 2 It is a network architecture diagram of the modal conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention;

[0034] Figure 3 It is a comparison diagram of prediction results of the modal conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention;

[0035] Figure 4Schematic diagram of a terminal device for a modality conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention;

[0036] Figure 5 Schematic diagram of a readable storage medium for a modality conversion method of quantitative magnetic resonance imaging based on artificial intelligence proposed by the present invention;

[0037] In the figure, 200 - terminal device, 210 - memory, 211 - RAM, 212 - cache, 213 - ROM, 214 - program / utilities, 215 - program modules, 220 - processor, 230 - bus, 240 - external device, 250 - I / O interface, 260 - network adapter, 300 - program product. Specific embodiments

[0038] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in combination with embodiments and the accompanying drawings. The illustrative embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0039] Embodiment 1

[0040] Reference Figure 1 , the modality conversion method of quantitative magnetic resonance imaging based on artificial intelligence includes the following steps:

[0041] S1. Data acquisition and fitting: Use a magnetic resonance device to scan a target object, collect modality data with different sequence parameters, and obtain quantitative MRI image data of the target object through fitting according to physical formulas;

[0042] Specifically, in this embodiment, a 3T magnetic resonance device is used to scan the brain of a patient, and multi-modal brain MRI data with different sequence parameters are collected, including pulse sequence parameters such as repetition time TR, echo time TE, and inversion time IR. The collected image data are used to train, validate, and test a deep learning model. Based on known physical formulas, multi-dimensional diffusion-weighted imaging (MDME) modality data such as T1map, T2map, and PDmap of each patient are generated through non-linear fitting.

[0043] In this embodiment, the time for the longitudinal magnetization vector in the tissue to recover to the equilibrium state is described by T1map, and the fitting is performed using the following formula:

[0044]

[0045] Among them, represents the signal intensity at the inversion time ; represents the signal intensity at the equilibrium state; represents the longitudinal relaxation time; represents the flip time, that is, by collecting the signal intensities under different and fitting to obtain the value.

[0046] In this implementation, the time for the decay of the transverse magnetization vector in the tissue due to the loss of coherence is described by T2map and fitted using the following formula:

[0047]

[0048] where, represents the signal intensity at the echo time ; represents the initial signal intensity; represents the transverse relaxation time; represents the echo time, that is, by collecting multiple signals and fitting to obtain the value.

[0049] In this embodiment, the proton density in the tissue is represented by PDmap, which is the normalized value of the signal intensity , and is obtained by normalizing the signal intensities of the T1 and T2 fitting formulas:

[0050]

[0051] The correction factor described above is constrained by the sensitivity of the receiving coil, scanning sequence parameters (such as TR and TE), etc.

[0052] Perform skull stripping on the generated T1map, T2map, and PDmap images to remove irrelevant background information and retain the effective image region of the brain tissue. Randomly divide the processed data into a training set, a validation set, and a test set, which are used for model training, parameter optimization, and performance evaluation respectively.

[0053] S2. Data generation: Generate conventional magnetic resonance scanning sequence modal data with different sequence scanning parameters based on the formula for synthetic quantitative MRI image data;

[0054] Specifically, according to the fitting formulas of T1map, T2map, and PDmap, calculate the signal intensities under different parameters. The specific formulas are as follows:

[0055] T1-weighted imaging generation formula

[0056]

[0057] T2-weighted imaging generation formula

[0058]

[0059] PD weighted imaging generation formula

[0060]

[0061] Among them, the T1, T2, and PD values have been obtained through fitting; TR, TE, and TI are the set scanning conditions; finally, the signal intensity images under different parameter combinations are output 。

[0062] In this embodiment, the modal data is generated for each scanning parameter through the following steps:

[0063] (1) Load the fitted quantitative MRI data, that is, load the quantitative MRI image data of T1map, T2map, and PDmap, where each pixel corresponds to the T1, T2, and PD parameters of a tissue;

[0064] (2) Obtain a set of TR, TE, and TI parameter combinations according to the actual application;

[0065] (3) Use the above formula to calculate the signal intensity of each pixel for each set of parameters , and generate the corresponding modal image;

[0066] (4) Repeat the above process to traverse all parameter combinations and generate multi-modal T1-weighted, T2-weighted, and PD-weighted image data.

[0067] S3. Data preprocessing: Preprocess the modal data of the conventional magnetic resonance scanning sequence by unifying the resolution, removing extreme values, Z-Score normalization, and data augmentation to ensure data consistency and standardization;

[0068] Specifically, in this embodiment, cubic spline interpolation or nearest neighbor interpolation is used to adjust the image size;

[0069] Since there may be extreme values in the MRI data, such as signal anomalies caused by noise, removing extreme values can remove these abnormal values and prevent them from having an adverse impact on model training. In this embodiment, percentiles such as the 1st and 99th percentiles are used to calculate the effective signal range, and the values outside the range are clipped to the threshold range. At the same time, the pixel values exceeding the threshold are set to the threshold itself.

[0070] The data augmentation in this embodiment expands the dataset by introducing random variations to increase the robustness and generalization ability of the model. The implementation method is as follows:

[0071] (1) Foreground clipping: Remove the background area and only retain the important tissues, which is implemented using simple threshold segmentation or deep learning segmentation tools.

[0072] (2) Background minimum value setting: The background pixel values are uniformly set to zero to eliminate the interference of irrelevant signals on the model.

[0073] (3) Random flipping: Horizontally or vertically flip the image to simulate different acquisition angles.

[0074] (4) Rotation and scaling: Randomly rotate by a certain angle such as -10° to +10° and slightly scale such as 90% to 110%.

[0075] Simulate Gaussian noise or motion artifacts to enhance data diversity.

[0076] S4. Establish a deep neural model and use deep learning technology to achieve quantitative MRI modality conversion for conventional magnetic resonance scan sequence modality data with arbitrary different contrasts in the input.

[0077] Reference Figure 2 , in this embodiment, the U-Net convolutional neural network structure is adopted, which includes multiple convolutional layers, pooling layers, upsampling layers and skip connection layers. At the same time, a parameter encoding module is introduced to encode physical parameters (TR, TE and IR), where the rectangular box represents the network output feature map, the gray arrow represents the convolutional layer + activation layer + normalization layer, the red arrow represents the maximum pooling, the blue arrow represents the upsampling, and the blue dotted line represents the skip connection.

[0078] Among them, the parameter encoding module includes an independent embedding layer. The independent embedding layer sets independent fully connected layers (FC layers) for each physical parameter (TR, TE and IR) to map the input parameters to a high-dimensional embedding space.

[0079] At the same time, the parameter encoding module performs an addition operation on the downsampled feature map and the corresponding parameter embedding, and realizes the deep fusion of physical parameters and image features through 1×1 convolution processing.

[0080] The neural network in this embodiment adopts a multi-branch structure. Each branch processes the input data with a specific contrast. At the middle layer of the network, the features extracted by each branch are concatenated, and the channel features are fused through 1×1 convolution. In the upsampling stage, multi-scale features are fused to generate a high-precision quantitative MRI image. The feature map outputs the corresponding T1map, T2map and PDmap images through three independent convolutional layers.

[0081] The physical model parameter encoding obtained in this embodiment is to first perform high-dimensional mapping on physical parameters (TR, TE, and IR) through an independent embedding layer to extract rich feature representations. Secondly, the fused feature map is processed by a 1×1 convolutional layer to achieve spatial information retention and channel number optimization. Finally, the fused features of all modalities are concatenated to form a comprehensive feature vector, and a quantitative MRI image is output through the final convolutional layer.

[0082] Through the above deep learning model, the network is trained on the training set and the parameters are optimized on the validation set. On the test set, the generated T1map, T2map, and PDmap images are highly consistent with the real images, verifying the effectiveness of the model.

[0083] Experimental results

[0084] Reference Figure 3 , with the real T1map, T2map, and PDmap images on the top and the images predicted and generated by the deep learning model on the bottom. It can be seen that the synthetic images are consistent with the real images in terms of resolution, texture details, and quantitative accuracy, showing high physical consistency.

[0085] Example 2

[0086] Based on Example 1, this embodiment proposes an optimization scheme for a modality conversion method of quantitative magnetic resonance imaging based on artificial intelligence.

[0087] In this embodiment, a U-Net convolutional neural network is used to encode the sequence scanning parameters, and the encoded physical parameters are mapped to a high-dimensional embedding space through an independent fully connected layer to generate physical parameter embeddings; the physical parameter embeddings are fused with the convolutional feature maps of the corresponding modalities through a dynamic weight adjustment module, and the weight adjustment module generates attention weights according to the input image features to dynamically allocate the contribution degrees of physical parameters at different spatial positions. The fused feature map is processed by a convolutional layer to achieve deep adaptive fusion of physical parameters and image features. In this embodiment, an attention mechanism is introduced to dynamically adjust the parameter weights, enhance the adaptability of the model to different anatomical regions, and improve the local accuracy of the quantitative parameter map. Specifically:

[0088] Input: Convolutional feature map , physical parameter embeddings ;

[0089] The physical parameter embeddings are extended to through a fully connected layer to calculate the attention weight matrix ;

[0090] , where Denoted as the Sigmoid function, Denotes element-wise addition, Denotes the spatial expansion of physical parameter embedding, Denotes the height of the input feature map, i.e., the dimension of the feature map in the vertical direction, Denotes the width of the input feature map, i.e., the dimension of the feature map in the horizontal direction, Denotes the number of channels (or depth) of the input feature map, i.e., the feature dimension contained in each pixel point, usually corresponding to the number of channels of the convolutional layer, Denotes the real number field

[0091] Dynamically fuse features into , where Denotes element-wise multiplication, and the final dynamically fused feature of this embodiment Achieves local adaptivity through weighted distribution.

[0092] In this embodiment, the data augmentation in step S3 further includes cross-modal contrast augmentation. By randomly occluding part of the modal input, such as T1-weighted or T2-weighted images, the model is forced to generate a quantitative parameter map of the missing modality based on the remaining modalities, and the learning ability of the model for multi-modal correlation features is optimized through a contrast loss function. Specifically:

[0093] Input multi-modal data , select one modality such as Perform full-image or regional occlusion, and the mask is , where Denotes the T1-weighted MRI image, Denotes the proton density-weighted MRI image;

[0094] Force the model to generate a quantitative parameter map of the missing modality based on the remaining modalities ; ;

[0095] Contrast loss function:

[0096]

[0097] where Denotes the enhanced true label, Denotes the weight coefficient, Denotes the KL divergence. In this embodiment, the multi-modal correlation feature learning is optimized through the contrast loss function.

[0098] In this embodiment, a cross-device adaptive calibration module is added during the model training stage, that is, a device parameter encoder is added before the input layer. Device parameters such as the magnetic field strength and coil type of the MRI device are encoded into high-dimensional vectors, which are then concatenated with the physical parameter embeddings. A device-independent feature representation is generated through the calibration convolutional layer to ensure the generalization performance of the model under different MRI devices. Specifically:

[0099] Input device parameters [Magnetic field strength, coil type encoding];

[0100] Encode into device embeddings through a fully connected layer and concatenate with the physical parameters and input them into the calibration convolutional layer In this embodiment, by introducing the cross-device calibration module, the device parameters are encoded and then concatenated with the physical parameters, and device-independent features are generated through the calibration convolutional layer .

[0101] Embodiment 3

[0102] Refer to Figure 4 , based on Embodiment 1, this embodiment proposes a terminal device for a modality conversion method of quantitative magnetic resonance imaging based on artificial intelligence. The terminal device 200 includes at least one memory 210, at least one processor 220, and a bus 230 connecting different platform systems.

[0103] The memory 210 may include a readable medium in the form of volatile memory, such as RAM 211 and / or cache 212 memory, and may further include ROM 213.

[0104] Among them, the memory 210 also stores a computer program, which can be executed by the processor 220, so that the processor 220 executes the application of any one of the above-mentioned modality conversion methods of quantitative magnetic resonance imaging based on artificial intelligence in the embodiments of the present application. The specific implementation manner is the same as the implementation manner and the achieved technical effects described in the above-mentioned application embodiments, and some contents will not be elaborated. The memory 210 may further include a program / utilities 214 having a set (at least one) of program modules 215. Such program modules include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0105] Correspondingly, the processor 220 can execute the above-mentioned computer program and can also execute the program / utilities 214.

[0106] The bus 230 can represent one or more of several types of bus structures, including a memory bus or a memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of the various bus structures.

[0107] The terminal device 200 can also communicate with one or more external devices 240 such as a keyboard, a pointing device, a Bluetooth device, etc., and can also communicate with one or more devices capable of interacting with the terminal device 200, and / or communicate with any device (such as a router, a modem, etc.) that enables the terminal device 200 to communicate with one or more other computing devices. Such communication can be carried out through the I / O interface 250. Moreover, the terminal device 200 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 260. The network adapter 260 can communicate with other modules of the terminal device 200 through the bus 230. It should be understood that although not shown in the figure, other hardware and / or software modules can be used in combination with the terminal device 200, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0108] Embodiment 4

[0109] Reference Figure 5 , this embodiment provides a readable storage medium for a mode conversion method of artificial intelligence-based quantitative magnetic resonance imaging. Instructions are stored on the computer-readable storage medium, and when the instructions are executed by a processor, the above-mentioned mode conversion method of artificial intelligence-based quantitative magnetic resonance imaging is implemented. Its specific implementation manner is consistent with the implementation manner and the achieved technical effects described in the embodiments of the above application, and some content will not be repeated here.

[0110] Figure 5Fig. 0 shows the program product 300 provided in this embodiment for implementing the above application. It may be a portable compact disc read-only memory (CD-ROM), include program code, and can run on a terminal device, such as a personal computer. However, the program product 300 of the present invention is not limited thereto. In this embodiment, the readable storage medium may be any tangible medium that contains or stores a program, and this program can be used by or in conjunction with an instruction execution system, apparatus, or device. The program product 300 may adopt any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may, for example, but not be limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection with one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0111] The computer-readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, in which the readable program code is carried. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable storage medium may also be any readable medium other than the readable storage medium, and this readable medium can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above. The program code for performing the operations of the present invention may be written in any combination of one or more programming languages. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, executed as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0112] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only to illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A method for modal conversion of artificial intelligence-based quantitative magnetic resonance imaging, characterized in that, Including the following steps: S1. Data acquisition and fitting: Use a magnetic resonance device to scan the target object, acquire modal data with different sequence parameters, and fit to obtain the quantitative MRI image data of the target object according to physical formulas; S2. Data generation: Generate conventional magnetic resonance scan sequence modal data with different sequence scan parameters based on the formula for synthesizing quantitative MRI image data; S3. Data preprocessing: Preprocess the conventional magnetic resonance scan sequence modal data by unifying the resolution, removing extreme values, Z-Score normalization, and data augmentation to ensure data consistency and standardization; The data augmentation includes cross-modal contrast enhancement. By randomly occluding part of the modal input, such as T1-weighted or T2-weighted images, the model is forced to generate a quantitative parameter map of the missing modality based on the remaining modalities, and the learning ability of the model for multi-modal correlation features is optimized through a contrast loss function; S4. Establish a deep neural model: Use deep learning technology to achieve quantitative MRI modality conversion for the input conventional magnetic resonance scan sequence modal data with arbitrary different contrasts; In step S4, a U-Net convolutional neural network is used to establish a deep neural model. The neural network includes multiple convolutional layers, pooling layers, upsampling layers, skip connection layers, and a parameter encoding module. The parameter encoding module is used to encode the sequence scan parameters, and the sequence scan parameters are physical parameters; Step S4 specifically includes the following sub-steps: S401. Use a U-Net convolutional neural network to encode the sequence scan parameters, map the encoded physical parameters to a high-dimensional embedding space through an independent fully connected layer to generate physical parameter embeddings, add the physical parameter embeddings and the convolutional feature maps of the corresponding modalities in the channel dimension to integrate the physical parameter information into the feature maps, and process the added feature maps through a convolutional layer; S402. The neural network uses a multi-branch structure to process input images with different contrasts. Each branch encodes the physical parameters by introducing a parameter embedding layer and fuses them with the branch feature maps; after the branch features are concatenated and channel features are fused in the middle layer, multi-scale features are combined in the upsampling stage, and corresponding quantitative magnetic resonance images are generated through different convolutional layers; S403. Use an attention mechanism to dynamically adjust the parameter weights, enhance the adaptability of the model to different anatomical regions, and improve the local accuracy of the quantitative parameter map.

2. The method for modal conversion of quantitative magnetic resonance imaging based on artificial intelligence according to claim 1, wherein The quantitative MRI image data in step S1 includes multi-dimensional diffusion weighted imaging modal data such as T1map, T2map, and PDmap.

3. The method for modal conversion of artificial intelligence-based quantitative magnetic resonance imaging according to claim 1, wherein The sequence scan parameters in step S2 include repetition time TR, echo time TE, and inversion time IR.

4. The modality conversion method for quantitative magnetic resonance imaging based on artificial intelligence according to claim 1, characterized in that: The data augmentation in step S3 includes foreground cropping, background minimum setting, random flipping, and resolution unification.

5. The method for modality conversion of artificial intelligence-based quantitative magnetic resonance imaging according to claim 1, wherein The encoding module realizes the fusion processing of physical parameters and image features through MLP, so that the output quantitative magnetic resonance image has physical consistency.

6. The method for modality conversion of artificial intelligence-based quantitative magnetic resonance imaging according to claim 1, characterized in that The convolutional layer used in step S401 is a 1×1 convolutional layer.

Images