Intracranial artery plaque magnetic resonance image enhancement processing model and construction method

Abstract

The invention provides an intracranial artery plaque magnetic resonance image enhancement processing model which comprises a data preprocessing module, a generator module, an attention mechanism module, a discriminator module and a loss function module. The data preprocessing module is used for preprocessing the intracranial artery plaque magnetic resonance plain scanning image; the generator module is used for converting the input 7T magnetic resonance plain-scan image into a magnetic resonance enhanced image; the attention mechanism module is used for enhancing the attention of the model on different feature channels; the discriminator module is used for discriminating the input magnetic resonance plain-scan image and the magnetic resonance enhanced image generated by conversion and judging the authenticity of the image; and the loss function module is used for calculating loss between the generated image and the real image and guiding model optimization. According to the method, the 7T magnetic resonance intracranial artery plaque image can be flatly scanned, the enhanced 7T magnetic resonance intracranial artery plaque image is generated, and the generated enhanced image is basically equal to the image effect after a contrast agent is used.

Description

Technical Field

[0001] The present invention relates to the technical field of clinical medical imaging image enhancement, and particularly relates to an intracranial artery plaque magnetic resonance image enhancement processing model and a construction method thereof. Background Art

[0002] Acute ischemic stroke (AIS) has high incidence, disability rate, mortality rate and recurrence rate, often leaving different degrees of sequelae to patients. Intracranial atherosclerotic stenosis (ICAS) is an important factor in the pathogenesis and recurrence of AIS. Given the relatively high clinical recurrence risk of AIS, timely evaluation of the intracranial artery plaques of patients is crucial for diagnosing ICAS, treating stroke and preventing stroke recurrence. It can not only prevent the occurrence of ischemic stroke, but also help improve the clinical treatment efficacy of ischemic stroke.

[0003] Intracranial artery magnetic resonance is the main clinical technique for judging the nature of plaques. At present, two factors are required for clinically judging the nature of cerebral atherosclerotic plaques. On the one hand, ultra-high resolution is required for plaque imaging. On the other hand, magnetic resonance contrast agents need to be injected into patients to change the local relaxation characteristics of tissues, improve the imaging contrast and compare with surrounding tissues, so as to judge the nature of intracranial artery plaques. However, most magnetic resonance contrast agents are gadolinium agents, which may cause allergic reactions, nephrogenic systemic fibrosis, gadolinium deposition in the brain after repeated use and other related risks, and have adverse effects such as increasing the examination time of patients.

[0004] Therefore, there is an urgent need for an image processing technology that can perform ultra-high resolution processing on image data to achieve the effect of enhanced scanning without injecting contrast agents, that is, the nuclear magnetic resonance images obtained by conventional scanning can also reflect the situation of intracranial artery plaques. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention proposes an intracranial artery plaque magnetic resonance image enhancement processing model and a construction method thereof to solve the technical problem that there is an urgent need for an image processing technology that performs enhanced scanning by injecting contrast agents and the nuclear magnetic resonance images obtained by conventional scanning can also reflect the situation of intracranial artery plaques in the existing technology.

[0006] The technical solution adopted by the present invention is as follows:

[0007] In the first aspect, an intracranial artery plaque magnetic resonance image enhancement processing model is provided, including: a data preprocessing module, a generator module, an attention mechanism module, a discriminator module and a loss function module;

[0008] The data preprocessing module is used to preprocess the intracranial artery plaque magnetic resonance plain scan images to provide input data for the model;

[0009] A generator module for converting an input non-contrast magnetic resonance image into a contrast-enhanced magnetic resonance image, where the non-contrast magnetic resonance image is a 7T magnetic resonance image, and the contrast-enhanced magnetic resonance image is a magnetic resonance image after using a contrast agent;

[0010] An attention mechanism module for enhancing the model's attention to different feature channels during the conversion of the non-contrast magnetic resonance image and the contrast-enhanced image by the generator;

[0011] A discriminator module for discriminating the input non-contrast magnetic resonance image and the converted contrast-enhanced magnetic resonance image to determine the authenticity of the images;

[0012] A loss function module for calculating the loss between the generated image and the real image to guide model optimization.

[0013] Furthermore, the architecture of the intracranial artery plaque magnetic resonance image enhancement processing model is an attention-guided Pix2Pix GAN.

[0014] Furthermore, the preprocessing includes collecting, dividing, format converting, registering, resampling, slice selecting, cropping, normalizing, and quality screening of the magnetic resonance images.

[0015] Furthermore, the generator module is based on the U-Net++ structure, uses an Attention Block to extract features of the input image and focus on important regions or features, and generates a contrast-enhanced magnetic resonance image using the input non-contrast magnetic resonance image.

[0016] Furthermore, the attention mechanism module adopts the ScSE combined with the ECA-Net attention mechanism;

[0017] When adopting the ScSE attention mechanism, the convolutional block slides on the input image or feature map to perform weighted summation operations to extract image features; fuses features from different sources through connection operations; adds a pooling layer after the convolutional layer to reduce the data spatial dimension, reduce the computational load, and increase the robustness of the features; uses transposed convolution for upsampling to generate a high-resolution output image; uses the pre-trained ResNet-50 as the basic network architecture for feature extraction;

[0018] When adopting the ECA-Net attention mechanism, it fuses global context information through a squeezing operation, then calculates the size of the adaptive convolutional kernel, calculates the weights of the channels using one-dimensional convolution, and finally uses the Sigmoid activation function to map the weights between 0 and 1, and then multiplies the reshaped weight values with the original feature map to obtain feature maps under different weights.

[0019] Furthermore, the discriminator module adopts a Patch GAN structure to distinguish the differences between source-domain and target-domain images, enhancing the constraints during the image translation process.

[0020] Furthermore, the loss function of the loss function module selects the focal frequency loss.

[0021] In a second aspect, a method for constructing an intracranial artery plaque magnetic resonance image enhancement processing model is provided, which is used to construct the intracranial artery plaque magnetic resonance image enhancement processing model described in the first aspect. The construction process includes:

[0022] Taking the magnetic resonance plain scan image as the input and the magnetic resonance enhanced image corresponding to the magnetic resonance plain scan image as the output, training the attention-guided Pix2Pix GAN neural network;

[0023] During training, the mini-batch stochastic gradient descent method is used to update the model weights. Both the generator and the discriminator use the Adam optimizer; Spectral Normalization is used in the normalization layer to prevent gradient vanishing or explosion by restricting the spectral norm of the weight matrix.

[0024] In a third aspect, an electronic device is provided, including:

[0025] One or more processors;

[0026] A storage device for storing one or more programs;

[0027] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for constructing the intracranial artery plaque magnetic resonance image enhancement processing model described in the first aspect.

[0028] In a fourth aspect, a computer program product is provided, including computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the method for constructing the intracranial artery plaque magnetic resonance image enhancement processing model described in the first aspect are implemented.

[0029] As can be seen from the above technical solutions, the beneficial technical effects of the present invention are as follows:

[0030] 1. The magnetic resonance image enhancement processing model for intracranial artery plaques is based on the pix2pix GAN framework, and generates images through the adversarial training of a generator and a discriminator. The generator attempts to generate synthetic images similar to real enhanced images, while the discriminator distinguishes between real images and generated images. The continuous game between the two makes the images generated by the generator closer and closer to real images. By virtue of the morphological and gray-scale differences caused by the susceptibility characteristics of different types of plaques themselves, as well as the completely matching data before and after enhancement in the same sequence, enhanced 7T magnetic resonance intracranial artery plaque images can be generated based on non-contrast-enhanced 7T magnetic resonance intracranial artery plaque images. The generated enhanced images are basically equivalent to the image effects after using contrast agents.

[0031] 2. In terms of the generator, in this embodiment, the U-Net++ structure is adopted to replace the traditional U-Net structure, and its convolutional blocks are improved to make it more suitable for image generation tasks. At the same time, two attention mechanisms, ScSE and ECA–Net, are added, which can flexibly capture the dependencies between channels, enhance the feature representation ability of the model, and improve the image quality to better generate intracranial artery plaque images with complex lesions.

[0032] 3. In terms of the discriminator, the Patch GAN structure is adopted. Compared with the traditional global GAN, it can pay more attention to the local texture and style features of images and has advantages in generating high-resolution and detail-rich medical images. An additional discriminator is added to the original Pix2Pix network framework, which can distinguish the differences between source domain and target domain images and enhance the constraints during image translation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to actual scale.

[0034] Figure 1 It is a schematic diagram of the working principle of the magnetic resonance image enhancement processing model for intracranial artery plaques according to the embodiment of the present invention;

[0035] Figure 2 It is a schematic diagram of the enhanced image effect generated by using the magnetic resonance image enhancement processing model for intracranial artery plaques according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The following will describe in detail the embodiments of the technical solutions of the present invention with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and therefore are only examples and should not be used to limit the protection scope of the present invention.

[0037] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application shall have the ordinary meanings understood by those skilled in the art to which the present invention pertains.

[0038] Embodiment

[0039] The inventors found through research that in terms of plaque imaging with ultra-high resolution, the conventional magnetic resonance with a resolution of 0.8 mm isotropic based on 3T in the past has great limitations in imaging blood vessels such as the middle cerebral artery (about 2 mm), that is, only 2 layers can be imaged for one blood vessel. However, the 7T magnetic resonance has greatly improved the image resolution and can perform 0.4 mm isotropic or 0.1 mm in-plane resolution imaging, that is, 5 or more layers can be imaged for one blood vessel, which greatly improves the accuracy and data volume of the plaque-targeted model. Therefore, the technical solution of this embodiment selects 7T magnetic resonance.

[0040] This embodiment proposes a magnetic resonance image enhancement processing model for intracranial artery plaques. The architecture of this model is an Attention-Guided Pix2Pix GAN. This model can generate enhanced 7T magnetic resonance intracranial artery plaque images based on the plain 7T magnetic resonance intracranial artery plaque images. The generated enhanced images are basically equivalent to the image effects after using a contrast agent. It should be noted that the object of the image enhancement processing in this embodiment is intracranial artery plaques, which refer to a kind of deposit formed on the inner wall of cerebral arteries. These deposits are mainly composed of cholesterol, calcium salts and other substances. Over time, these substances gradually accumulate on the inner wall of cerebral arteries and finally form plaques. Intracranial artery plaques and cerebral blood vessels are two completely different concepts. Cerebral blood vessels are equivalent to a kind of pipeline, while intracranial artery plaques are equivalent to the attachments on the inner wall of the pipeline. The Pix2Pix GAN model in this embodiment fully considers the huge differences in morphology between intracranial artery plaques and cerebral blood vessels. Especially when generating targets such as intracranial artery plaques with irregular shapes, blurred edges and small areas, the magnetic resonance image enhancement processing model for intracranial artery plaques in this embodiment is divided into the following functional modules:

[0041] Data preprocessing module: Realize functions such as collection, division, format conversion, registration, resampling, slice selection, cropping, normalization and quality screening of magnetic resonance image data, and provide appropriate input data for the model. During the model construction and training process, the objects of preprocessing are magnetic resonance plain images (as the input of the model) and magnetic resonance enhanced images (as the output of the model); when using the trained model, the object of preprocessing is magnetic resonance plain images.

[0042] Generator module: Based on the U-Net++ structure and combined with the attention mechanism, it converts the input T1 image (plain scan image) into a synthetic-CE-T1 image (enhanced image).

[0043] Attention mechanism module (ECA-Net): Functions in the generator to enhance the model's attention to different feature channels and improve the quality of the generated images.

[0044] Discriminator module: Adopts the Patch GAN structure to discriminate the input images (real CE-T1 and generated synthetic-CE-T1) and judge the authenticity of the images.

[0045] Loss function module (focal frequency loss): Calculates the loss between the generated image and the real image, guides the model optimization, and focuses on the difficult-to-process frequency components to improve the image quality.

[0046] Data flow between modules: The data preprocessing module inputs the processed data into the generator module and the discriminator module respectively. The synthetic-CE-T1 image generated by the generator module and the real CE-T1 image are input into the discriminator module for discrimination, and the discrimination result is fed back to the generator module to adjust the generation strategy. The attention mechanism module processes the features inside the generator module, and the loss function module calculates the loss based on the output of the generator module and the real image and guides the training and optimization of the model.

[0047] The construction process of the model is as follows:

[0048] 1. Data preparation

[0049] Collect paired 7T magnetic resonance T1 images (plain scan images) and 7T magnetic resonance T1-CE images (enhanced images) of the brain from October 2023 to May 2024, with a total of 300 patients (median age, 55 years; 65 [62.5%] male patients and 39 [37.5%] female patients).

[0050] Divide the data into a training set and a test set according to 8:2. Preprocess the data, including adjusting the window width and window level between the plain scan and enhanced magnetic resonance of the same patient (due to color difference), performing rigid registration on T1 and T1-CE (due to time difference), converting DCM format data to Nifti format, resampling the resolution, and selecting slices containing lesions;

[0051] The brain plaques are relatively small and the number of slices containing lesions is relatively small. Therefore, 10-20 slices containing lesions are selected for each subject. Brain plaques usually occupy fewer pixels in the image. If the feature map is too small, the features of small targets may be overcompressed or lost. Therefore, each MRI slice is cropped to a size of 256×256 pixels, converted into the png format of pictures, and the pixel values are normalized to [-1,1], and pictures with poor image quality are excluded after manual inspection.

[0052] Through this step, a suitable data set is provided for model training, the bias and noise in the data are reduced, and the data quality is improved.

[0053] 2. Build the main architecture of the model

[0054] The architecture of the magnetic resonance image enhancement processing model for intracranial artery plaques is Pix2PixGAN based on attention guidance, including a generator and a discriminator, where:

[0055] (1) Generator

[0056] Since the plaques in the cranial arteries have complex lesions, in order to reduce the loss of information in the encoding and decoding processes, the generator adopts a more densely connected U-Net++ structure and combines more dense skip connections to reduce information loss. In the U-Net++ structure, an Attention Block is used to extract the features of the input image and focus on important regions or features; the ScSE (Concurrent Spatial and Channel Squeeze&Excitation) structure is used as the attention mechanism to integrate multi-dimensional information; the convolutional block slides on the input image or feature map to perform weighted summation operations to extract image features; the features from different sources are fused through connection operations; a pooling layer (using the Max pooling method) is added after the convolutional layer to reduce the data spatial dimension, reduce the computational load and increase the robustness of the features; transposed convolution is used for upsampling to generate a high-resolution output image; the pre-trained ResNet-50 (with frozen parameters) is used as the basic network architecture for feature extraction.

[0057] The ECA-Net (Efficient Channel Attention) attention mechanism is added to the generator. Specifically, the global context information is fused through a squeeze operation (global average pooling converts the feature map from (N,C,H,W) to (N,C,1,1)), then the size of the adaptive convolutional kernel is calculated, the weights of the channels are calculated using one-dimensional convolution, and finally the weight values are mapped between (0-1) using the Sigmoid activation function. Then, the reshaped weight values are multiplied by the original feature map to obtain the feature map under different weights.

[0058] The function of the generator is to generate synthetic-CE-T1 images from the input T1 images, reduce information loss during the encoding and decoding processes, and improve the ability to generate complex cranial plaque images. By adding an attention mechanism, the generator can distinguish the importance of different feature channels, enhance the feature representation ability of the model, and improve the quality of the generated images. In terms of the generator, in this embodiment, the U-Net++ structure is adopted to replace the traditional U-Net structure, and its convolutional blocks are improved to make it more suitable for image generation tasks. At the same time, two attention mechanisms, ScSE and ECA–Net, are added, which can flexibly capture the dependencies between channels, enhance the feature representation ability of the model, and improve the image quality to better generate cranial artery plaque images with complex lesions.

[0059] (2) Discriminator

[0060] The discriminator adopts the Patch GAN structure, which maps the input medical image into an N×N matrix (patch), and each element in the matrix represents the probability that the corresponding local region of the image is a real sample.

[0061] The function of the discriminator is to independently discriminate the authenticity of each small block, which can focus more on learning the local texture and style features of the image, thereby improving the detail quality of the generated image. In terms of the discriminator, the Patch GAN structure is adopted. Compared with the traditional global GAN, it can pay more attention to the local texture and style features of the image and has advantages in generating high-resolution and detail-rich medical images. A discriminator is added to the original Pix2Pix network framework, which can distinguish the differences between the source domain and target domain images and enhance the constraints during image translation.

[0062] 2. Model Training

[0063] During the training process of the model, the 7T magnetic resonance plain scan images are used as the input of the model, and the 7T magnetic resonance enhanced images are used as the output of the model. The focal frequency loss is selected as the loss function during training. The images are transformed to the frequency domain by two-dimensional discrete Fourier transform, and the focal frequency loss is calculated according to the spectral weight matrix (dynamically determined by the non-uniform distribution of the current loss of each frequency during training). The elements of this matrix are defined as the α power of the difference between two frequency values. Each spectral coordinate value is mapped to an Euclidean vector in the two-dimensional space, taking into account both the amplitude and phase information of the spatial frequency. Finally, the complete focal frequency loss is scaled by the Euclidean distance of these vectors.

[0064] In this embodiment, the focal frequency loss is used as the loss function to encourage the model to approach the real enhanced image in terms of local details and global context. An attention-guided synthesis strategy is proposed so that the generator will pay more attention to the lesion part, adjust the local and global information of the image, focus on the difficult-to-process frequency components, generate images with rich detail variations, reduce the gap between the generated image and the real image in the frequency domain, and improve the image reconstruction and synthesis quality.

[0065] In some embodiments, when training and optimizing the model, the mini-batch stochastic gradient descent method is used to update the model weights. The Adam optimizer (with momentum parameters β1 = 0.5 and β2 = 0.999, and batch size of 16) is used for both the generator and the discriminator. The polynomial decay strategy is adopted to adjust the learning rate (learning rate is 0.0002). Spectral Normalization (SN) is used in the normalization layer to prevent gradient vanishing or explosion by restricting the spectral norm (i.e., the largest singular value) of the weight matrix. By adopting the above method, the stability of model training can be improved during the training process, the model parameters can be optimized, and the model can converge to better performance.

[0066] This embodiment provides an attention mechanism-guided generative adversarial model (ATT-GAN) for generating 7.0T MRI intracranial artery plaque enhanced images. Based on the pix2pix GAN framework, this model generates images through the adversarial training of the generator and the discriminator. The generator attempts to generate synthetic images (synthetic-T1) similar to the real enhanced images (CE-T1), and the discriminator differentiates between real images and generated images. The two continuously play against each other, making the images generated by the generator closer and closer to the real images. By virtue of some morphological and gray-scale differences caused by the susceptibility characteristics of different plaques themselves, as well as the completely matching data before and after enhancement in the same sequence, enhanced 7T MRI intracranial artery plaque images can be generated based on the non-contrast-enhanced 7T magnetic resonance intracranial artery plaque images. The generated enhanced images are basically equivalent to the image effects after using contrast agents, as Figure 2 shown Figure 2 The first row of images in

[0067] In some embodiments, a method for constructing an intracranial artery plaque magnetic resonance image enhancement processing model is also provided, which is used to construct the intracranial artery plaque magnetic resonance image enhancement processing model described in the first aspect. The construction process includes:

[0068] Taking the magnetic resonance plain scan image as the input and the corresponding magnetic resonance enhanced image as the output, train the attention-guided Pix2Pix GAN neural network;

[0069] During training, the mini-batch stochastic gradient descent method is used to update the model weights, and the Adam optimizer is used for both the generator and the discriminator; Spectral Normalization is used in the normalization layer to prevent gradient vanishing or explosion by restricting the spectral norm of the weight matrix.

[0070] In some embodiments, an electronic device is also provided, including: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method for constructing the magnetic resonance image enhancement processing model of intracranial artery plaques described above.

[0071] In some embodiments, a computer program product is also provided, including computer programs / instructions, which when executed by a processor, implement the steps of the method for constructing the magnetic resonance image enhancement processing model of intracranial artery plaques described above.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.

Claims

1. A magnetic resonance image enhancement processing model for intracranial arterial plaques, characterized in that: include: Data preprocessing module, generator module, attention mechanism module, discriminator module and loss function module; The data preprocessing module is used to preprocess the plain scan magnetic resonance images of intracranial arterial plaques to provide input data for the model; The generator module is used to convert the input plain scan magnetic resonance image into an enhanced magnetic resonance image, wherein the plain scan magnetic resonance image is a 7T magnetic resonance image, and the enhanced magnetic resonance image is a magnetic resonance image after using a contrast agent; The attention mechanism module is used to enhance the model's attention to different feature channels during the conversion of the generator between the magnetic resonance plain scan image and the enhanced image; The discriminator module is used to discriminate the input MRI plain scan image and the converted MRI enhanced image to determine the authenticity of the image; The loss function module is used to calculate the loss between the generated image and the real image to guide model optimization.

2. The intracranial arterial plaque magnetic resonance image enhancement processing model according to claim 1, characterized in that: The architecture of the intracranial arterial plaque magnetic resonance image enhancement processing model is an attention-guided Pix2Pix GAN.

3. The intracranial arterial plaque magnetic resonance image enhancement processing model according to claim 1, characterized in that: The preprocessing includes collecting, segmenting, format conversion, registration, resampling, slice selection, cropping, standardization and quality screening of magnetic resonance images.

4. The intracranial arterial plaque magnetic resonance image enhancement processing model according to claim 1, characterized in that: The generator module is based on the U-Net++ structure, uses the Attention Block to extract the features of the input image and focus on important areas or features, and generates an MRI enhanced image using the input MRI plain scan image.

5. The intracranial arterial plaque magnetic resonance image enhancement processing model according to claim 1, characterized in that: The attention mechanism module adopts ScSE combined with ECA-Net attention mechanism; When using the ScSE attention mechanism, the convolution block slides on the input image or feature map to perform a weighted sum operation to extract image features; features from different sources are fused through a concatenation operation; a pooling layer is added after the convolution layer to reduce the data space dimension, reduce the computational load, and increase the robustness of the features; transposed convolution is used for upsampling to generate a high-resolution output image; and a pre-trained ResNet-50 is used as the basic network architecture for feature extraction; When using the ECA-Net attention mechanism, the global context information is fused through the squeezing operation, and then the size of the adaptive convolution kernel is calculated. The channel weight is calculated using one-dimensional convolution, and finally the weight is mapped between 0 and 1 using the Sigmoid activation function. The reshaped weight value is then multiplied with the original feature map to obtain feature maps under different weights.

6. The intracranial arterial plaque magnetic resonance image enhancement processing model according to claim 1, characterized in that: The discriminator module adopts the Patch GAN structure to distinguish the differences between source domain and target domain images and enhance the constraints in the image translation process.

7. The intracranial arterial plaque magnetic resonance image enhancement processing model according to claim 1, characterized in that: The loss function of the loss function module selects focal frequency loss.

8. A method for constructing a magnetic resonance image enhancement processing model for intracranial arterial plaques, characterized in that: Used to construct the intracranial arterial plaque magnetic resonance image enhancement processing model according to any one of claims 1 to 7, the construction process includes: The attention-guided Pix2Pix GAN neural network is trained with MRI plain scan images as input and MRI enhanced images corresponding to the MRI plain scan images as output. During training, the mini-batch stochastic gradient descent method is used to update the model weights. The Adam optimizer is used for both the generator and the discriminator. Spectral Normalization is used in the normalization layer to prevent the gradient from disappearing or exploding by limiting the spectral norm of the weight matrix.

9. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for constructing an intracranial arterial plaque magnetic resonance image enhancement processing model as described in claim 8.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the method for constructing an intracranial arterial plaque magnetic resonance image enhancement processing model described in claim 8 are implemented.

Images