A processing method and device for coronary magnetic resonance imaging reconstruction model

Abstract

The embodiment of the present invention relates to a method and device for processing a coronary magnetic resonance imaging reconstruction model, the method comprising: constructing a coronary magnetic resonance imaging reconstruction model; performing model training on the coronary magnetic resonance imaging reconstruction model; after the model training is completed, using a magnetic resonance scanning device to perform a full-heart three-dimensional coronary imaging scan on a first subject according to a configured scanning sequence type, and performing K-space data sampling on the obtained scanning sequence data based on a preset undersampling method during the scanning process to obtain corresponding first undersampling K-space data; and inputting the first undersampling K-space data into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image; and performing coronary stenosis feature recognition based on the first imaging image to generate a corresponding first recognition report. The present invention can shorten the duration of coronary magnetic resonance imaging and improve imaging clarity.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and device for processing a coronary magnetic resonance imaging reconstruction model. Background Art

[0002] Coronary Atherosclerosis Heart Disease (CHD), also known as CHD for short, is a heart disease in which atherosclerotic lesions in the coronary arteries cause narrowing of the vascular lumen. The current conventional technical means of examining coronary heart disease is based on contrast-enhanced CT (Computer Tomography Angio, CTA) technology to perform imaging analysis of the patient's coronary vessels and identify the patient's coronary stenosis characteristics (coronary stenosis trend, plaque properties related to coronary stenosis, etc.) through CTA images. However, this CTA technology requires the injection of a large dose of contrast agent into the subject during the examination, which is an invasive examination method. It will still cause a certain degree of adverse effects on subjects with poor physical fitness and / or older age.

[0003] Magnetic resonance imaging (MRI) technology can provide patients with a non-invasive means of magnetic resonance angiography (MRA). The imaging technology based on respiratory navigator echo (NE) triggered magnetic resonance angiography can also allow patients to complete the test in a free breathing state without uncomfortable breath holding. Doctors can identify the characteristics of coronary artery stenosis in patients by analyzing coronary magnetic resonance imaging images. However, we have found in practice that coronary magnetic resonance imaging examinations also have some shortcomings, such as long imaging time and unstable imaging clarity. Among them, the long imaging time is related to the imaging principle of magnetic resonance imaging technology; the unstable imaging clarity is related to the sampling of the respiratory navigation trajectory; if the patient's breathing fluctuates greatly, it is easy to produce disturbances in the final imaging, and the disturbances here can be regarded as a kind of noise.

[0004] The imaging principle of magnetic resonance imaging technology is to generate a specified scanning sequence (such as SSFP sequence, FGRE sequence, etc.) through scanning, and extract and convert the sequence data based on the known data conversion rules between various scanning sequences and K space (Kspace), and fill the K space based on the converted data to obtain the corresponding full-sampled K space data, and perform Fourier inverse transform on the full-sampled K space data to obtain the corresponding magnetic resonance imaging image. It can be seen that the magnetic resonance imaging time can be roughly composed of the sequence scanning time, the sequence-K space filling time and the K space-image conversion time. If the above three types of time can be shortened, the overall time of magnetic resonance imaging can be shortened. At present, the first time period, namely the sequence scanning time, can be effectively shortened by multi-coil parallel scanning, but there is no good optimization mechanism for the latter two time periods (sequence-K space filling time, K space-image conversion time).

[0005] The compressed sensing algorithm is a technology for finding sparse solutions to underdetermined linear systems. The algorithm has been widely used in the field of communication signal processing. It can be used to undersample high-throughput communication signals and effectively reconstruct the original signals based on the undersampling results, thereby improving communication efficiency. The deep learning (DL) model of artificial intelligence technology can be used to simulate the algorithm deduction process, and can also be used to effectively reduce the noise of images to improve image clarity. The processing accuracy of the model can also be continuously improved through continuous training of the deep learning model. The technical problem to be solved by the present invention is how to improve the two types of deficiencies (long imaging time and unstable imaging clarity) currently existing in coronary magnetic resonance imaging based on the compressed sensing algorithm and the deep learning model. Summary of the invention

[0006] The purpose of the present invention is to provide a processing method, device, electronic device and computer-readable storage medium for a coronary magnetic resonance imaging reconstruction model in response to the defects of the prior art. The present invention pre-constructs a deep learning model, namely a compressed sensing algorithm model, based on a convolutional neural network structure to simulate the deduction process of a compressed sensing algorithm. The model can reconstruct a magnetic resonance imaging image based on an input undersampled K-space data and output the corresponding magnetic resonance imaging image; and pre-constructs a deep learning model, namely a deep learning denoising model, for denoising a magnetic resonance imaging image based on a U-Net model or a GAN model structure. The model can denoise an input magnetic resonance imaging image and output a high-definition imaging image; and pre-constructs an end-to-end coronary magnetic resonance imaging reconstruction model composed of a compressed sensing algorithm model + a deep learning denoising model. The coronary magnetic resonance imaging reconstruction model can be based on An input undersampled K-space data obtains a high-definition magnetic resonance imaging image; and the overall model training of the coronary magnetic resonance imaging reconstruction model is completed by training the compressed sensing algorithm model and the deep learning noise reduction model separately; and after the overall model training is completed, the coronary magnetic resonance imaging reconstruction model is introduced into the daily coronary magnetic resonance angiography scanning process, and in the scanning process, the filling time of the sequence-K space is shortened based on the undersampling method, the conversion time of the K space-image is shortened based on the coronary magnetic resonance imaging reconstruction model, and the clarity of the output coronary magnetic resonance imaging is improved, and the coronary stenosis feature is identified based on the coronary magnetic resonance imaging output by the coronary magnetic resonance imaging reconstruction model. The present invention can not only shorten the duration of coronary magnetic resonance imaging and improve the imaging clarity, but also improve the accuracy of coronary stenosis feature identification based on the high-definition coronary magnetic resonance imaging image.

[0007] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present invention provides a method for processing a coronary magnetic resonance imaging reconstruction model, the method comprising:

[0008] Construct a coronary magnetic resonance imaging reconstruction model;

[0009] Performing model training on the coronary magnetic resonance imaging reconstruction model;

[0010] After the model training is completed, a full-heart three-dimensional coronary artery imaging scan is performed on the first subject using a magnetic resonance scanning device according to the configured scanning sequence type, and during the scanning process, K-space data sampling is performed on the obtained scanning sequence data based on a preset undersampling method to obtain corresponding first under-sampled K-space data; and the first under-sampled K-space data is input into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image; and coronary stenosis features are identified based on the first imaging image to generate a corresponding first identification report.

[0011] Preferably, the coronary magnetic resonance imaging reconstruction model includes a compressed sensing algorithm model and a deep learning denoising model; the input end of the compressed sensing algorithm model is connected to the input end of the coronary magnetic resonance imaging reconstruction model, and the output end is connected to the input end of the deep learning denoising model; the output end of the deep learning denoising model is connected to the output end of the coronary magnetic resonance imaging reconstruction model;

[0012] The compressed sensing algorithm model is implemented based on a convolutional neural network; the compressed sensing algorithm model is used to perform magnetic resonance imaging image reconstruction processing according to the input undersampled K-space data and output a corresponding initial reconstruction image, specifically, the undersampled K-space data is brought into the compressed sensing algorithm objective function T, and the optimal solution of the initial reconstruction image is calculated in a direction that makes the compressed sensing algorithm objective function T reach a minimum value;

[0013] Among them, the objective function T of the compressed sensing algorithm is:

[0014]

[0015] P is the K-space undersampling matrix corresponding to the undersampling method, F is the Fourier transform matrix, X is the initial reconstruction image, M is the model mapping function of the compressed sensing algorithm model, Y is the undersampling K-space data, θ is the model parameter set of the compressed sensing algorithm model, λ is a preset regularization factor, ‖‖1, ‖‖2 are the corresponding L1 and L2 regularization terms respectively;

[0016] The deep learning denoising model is implemented based on a model structure of a class of deep learning models, and the deep learning model includes at least a U-Net model and a GAN model; the deep learning denoising model is used to perform magnetic resonance imaging denoising processing on the initial reconstructed image and output the corresponding final reconstructed image.

[0017] Preferably, the model training of the coronary magnetic resonance imaging reconstruction model specifically includes:

[0018] Acquire multiple high-definition whole-heart three-dimensional coronary magnetic resonance imaging images and their corresponding scanning sequence data to form corresponding first and second data sets; and train the compressed sensing algorithm model based on the first data set; and train the deep learning noise reduction model based on the second data set; the first data set includes multiple first data records; the first data record includes first training K-space data and a first label image; the second data set includes multiple second data records; the second data record includes a first training image and a second label image.

[0019] Furthermore, the acquisition of a plurality of high-definition whole-heart three-dimensional coronary magnetic resonance imaging images and their corresponding scanning sequence data to form corresponding first and second data sets specifically includes:

[0020] Using each of the acquired high-definition whole-heart three-dimensional coronary magnetic resonance imaging images as the corresponding first original image;

[0021] and performing K-space data sampling on the scanning sequence data corresponding to each of the first original images in the undersampling manner to obtain the corresponding first training K-space data; and using each of the first original images as a corresponding first label image; and forming a corresponding first data record from each of the first training K-space data and the corresponding first label image; and forming a corresponding first data set from all the obtained first data records;

[0022] And filter and denoise each of the first original images to obtain the corresponding second label image; and add noise to each of the first original images based on a plurality of preset disturbance noise adding schemes to obtain a plurality of corresponding first noise-added images; and use each of the first noise-added images as a corresponding first training image; and form a corresponding second data record from each of the first training images and the corresponding second label image; and form a corresponding second data set from all the obtained second data records.

[0023] Furthermore, the training of the compressed sensing algorithm model based on the first data set specifically includes:

[0024] Step 51, extracting the first first data record of the first data set as the corresponding current data record;

[0025] Step 52, extracting the corresponding first training K-space data and the first label map from the current data record;

[0026] Step 53, inputting the first training K-space data into the compressed sensing algorithm model to calculate and obtain a corresponding first training reconstruction image;

[0027] Step 54, bringing the first training K-space data, the first training reconstruction image and the first label image into a preset first training objective function O1 to calculate and obtain a corresponding first function value;

[0028] in,

[0029]

[0030] Y tr1 is the first training K-space data,

[0031] X tr1 For the first training reconstruction graph,

[0032] X tag1 is the first label map,

[0033] P is the K-space undersampling matrix,

[0034] F is the Fourier transform matrix,

[0035] ζ and λ are two preset regularization factors.

[0036] ‖‖1 and ‖‖2 are the corresponding L1 and L2 regularization terms respectively;

[0037] Step 55, identifying whether the first function value satisfies a preset first convergence function value range; if not, performing a round of optimization on the model parameter set of the compressed sensing algorithm model in the direction of minimizing the first training objective function L1 based on a preset model optimization algorithm, and returning to step 53 to continue training at the end of this round of optimization; if satisfied, identifying whether the current data record is the last of the first data set, if not, extracting the next first data record of the first data set as the new current data record and returning to step 52 to continue training, if yes, stopping training; the model optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm.

[0038] Furthermore, the training of the deep learning denoising model based on the second data set specifically includes:

[0039] Step 61, extracting the first second data record of the second data set as the corresponding current data record;

[0040] Step 62, extracting the corresponding first training graph and second label graph from the current data record;

[0041] Step 63, inputting the first training image into the deep learning denoising model to calculate and obtain a corresponding second training reconstruction image;

[0042] Step 64, bringing the second training reconstruction image and the second label image into a preset first loss function L1 to calculate a corresponding first loss value; when the model structure of the deep learning denoising model is implemented based on a U-Net model, the corresponding first loss function L1 uses an L2 loss function by default; when the model structure of the deep learning denoising model is implemented based on a GAN model, the corresponding first loss function L1 uses a cross entropy loss function by default;

[0043] Step 65, whether the first loss value satisfies a preset first loss value range is identified; if not, a round of optimization is performed on the model parameter set of the deep learning denoising model in the direction of minimizing the first loss function L1 based on a preset model optimization algorithm, and at the end of this round of optimization, the method returns to step 63 to continue training; if satisfied, whether the current data record is the last second data record of the second data set is identified, if not, the next second data record of the second data set is extracted as the new current data record and the method returns to step 62 to continue training, if so, the training is stopped; the model optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm.

[0044] Preferably, the step of inputting the first under-sampled K-space data into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image specifically includes:

[0045] The first under-sampled K-space data is input into the coronary magnetic resonance imaging reconstruction model, and the compressed sensing algorithm model performs magnetic resonance imaging image reconstruction processing based on the input first under-sampled K-space data and outputs the corresponding current initial reconstruction image; and the deep learning denoising model performs magnetic resonance imaging image denoising processing on the current initial reconstruction image and outputs the corresponding current final reconstruction image; and the obtained current final reconstruction image is output as the corresponding first imaging image.

[0046] Preferably, the field strength levels of the magnetic resonance scanning device include 1.5T and 3.0T; when the field strength level is 1.5T, the corresponding scanning sequence type is configured by default as a steady-state free precession SSFP sequence; when the field strength level is 3.0T, the corresponding scanning sequence type is configured by default as a fast gradient echo FGRE sequence.

[0047] A second aspect of the embodiments of the present invention provides a device for implementing the processing method of the coronary magnetic resonance imaging reconstruction model described in the first aspect, the device comprising: a model building module, a model training module and a model application module;

[0048] The model building module is used to build a coronary magnetic resonance imaging reconstruction model;

[0049] The model training module is used to perform model training on the coronary magnetic resonance imaging reconstruction model;

[0050] The model application module is used to perform a whole-heart three-dimensional coronary artery imaging scan on a first subject using a magnetic resonance scanning device according to a configured scanning sequence type after model training is completed, and perform K-space data sampling on the obtained scanning sequence data based on a preset undersampling method during the scanning process to obtain corresponding first under-sampled K-space data; and input the first under-sampled K-space data into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image; and perform coronary artery stenosis feature recognition based on the first imaging image to generate a corresponding first recognition report.

[0051] A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0052] The processor is used to be coupled to the memory, read and execute instructions in the memory, so as to implement the method steps described in the first aspect above;

[0053] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0054] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions. When the computer instructions are executed by a computer, the computer executes the instructions of the method described in the first aspect above.

[0055] The embodiments of the present invention provide a processing method, device, electronic device and computer-readable storage medium for a coronary magnetic resonance imaging reconstruction model. As can be seen from the above content, the embodiments of the present invention pre-construct a deep learning model for simulating the deduction process of the compressed sensing algorithm based on the convolutional neural network structure, namely the compressed sensing algorithm model, which can reconstruct the magnetic resonance imaging image according to an input undersampled K-space data and output the corresponding magnetic resonance imaging image; and pre-construct a deep learning model for denoising the magnetic resonance imaging image based on the U-Net model or GAN model structure, namely the deep learning denoising model, which can denoise an input magnetic resonance imaging image and output a high-definition imaging image; and pre-construct an end-to-end coronary magnetic resonance imaging reconstruction model composed of the compressed sensing algorithm model + the deep learning denoising model, through which the coronary magnetic resonance imaging reconstruction The model can obtain a high-definition magnetic resonance imaging image based on an input undersampled K-space data; and complete the overall model training of the coronary magnetic resonance imaging reconstruction model by separately training the compressed sensing algorithm model and the deep learning noise reduction model; and after the overall model training is completed, the coronary magnetic resonance imaging reconstruction model is introduced into the daily coronary magnetic resonance angiography scanning process, and in the scanning process, the sequence-K space filling time is shortened based on the undersampling method, the K-space-image conversion time is shortened based on the coronary magnetic resonance imaging reconstruction model, and the clarity of the output coronary magnetic resonance imaging is improved, and the coronary stenosis feature recognition is performed based on the coronary magnetic resonance imaging output by the coronary magnetic resonance imaging reconstruction model. Through the embodiment of the present invention, not only can the coronary magnetic resonance imaging time be shortened and the imaging clarity be improved, but also the accuracy of coronary stenosis feature recognition can be improved based on the high-definition coronary magnetic resonance imaging image. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of a processing method for a coronary magnetic resonance imaging reconstruction model provided in Embodiment 1 of the present invention;

[0057] Figure 2 This is a module structure diagram of the coronary magnetic resonance imaging reconstruction model provided in Example 1 of the present invention;

[0058] Figure 3 A module structure diagram of a processing device for a coronary magnetic resonance imaging reconstruction model provided in Embodiment 2 of the present invention;

[0059] Figure 4 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Embodiment 1 of the present invention provides a method for processing a coronary magnetic resonance imaging reconstruction model, such as Figure 1 As shown in the schematic diagram of a processing method for a coronary magnetic resonance imaging reconstruction model provided in the first embodiment of the present invention, the method mainly comprises the following steps:

[0062] Step 1: Construct a coronary magnetic resonance imaging reconstruction model.

[0063] Here, if Figure 2 As shown in the module structure diagram of the coronary magnetic resonance imaging reconstruction model provided in the first embodiment of the present invention, the coronary magnetic resonance imaging reconstruction model of the embodiment of the present invention includes a compressed sensing algorithm model and a deep learning noise reduction model. The input end of the compressed sensing algorithm model is connected to the input end of the coronary magnetic resonance imaging reconstruction model, and the output end is connected to the input end of the deep learning noise reduction model; the output end of the deep learning noise reduction model is connected to the output end of the coronary magnetic resonance imaging reconstruction model.

[0064] (1) Compressed sensing algorithm model:

[0065] The compressed sensing algorithm model of the embodiment of the present invention is a deep learning model for simulating the deduction process of the compressed sensing algorithm, which is implemented based on a convolutional neural network. The compressed sensing algorithm model performs magnetic resonance imaging image reconstruction processing based on the input undersampled K-space data and outputs the corresponding initial reconstruction image. The reasoning principle of the model is as follows: the undersampled K-space data is brought into the preset compressed sensing algorithm objective function T, and the optimal solution of the initial reconstruction image is calculated in the direction that makes the compressed sensing algorithm objective function T reach the minimum value. The minimum value here defaults to 0; wherein,

[0066] The objective function T of the compressed sensing algorithm is:

[0067]

[0068] P is the K-space undersampling matrix corresponding to the undersampling method, F is the Fourier transform matrix, X is the optimal magnetic resonance imaging image calculated by the model based on the compressed sensing algorithm theory, that is, the initial reconstruction image; M is the model mapping function of the compressed sensing algorithm model, Y is the undersampled K-space data, θ is the model parameter set of the compressed sensing algorithm model; λ is the preset regularization factor, ‖‖1, ‖‖2 are the corresponding L1 and L2 regularization terms respectively;

[0069] Here, it is known from the well-known magnetic resonance imaging principle that the corresponding full K-space data can be obtained by Fourier transforming the magnetic resonance imaging image, and vice versa, the corresponding magnetic resonance imaging image can be obtained by inverse Fourier transforming the full K-space data. In addition, it is known from the well-known compressed sensing algorithm theory that the corresponding under-sampled K-space data can be obtained by under-sampling the full K-space data. That is to say, the result obtained by continuously transforming the inferred magnetic resonance imaging image, i.e., the initial reconstructed image X, through the K-space under-sampling matrix P and the Fourier transform matrix F, i.e., the calculation result of PFX is a measurement data of the under-sampled K-space inferred by the compressed sensing algorithm; the expression of the compressed sensing algorithm objective function T is It reflects the difference between the model extrapolation results and the actual results of the under-sampled K space. The expression λ‖X‖1 reflects the noise level of the initial reconstructed image. The closer these two expressions are to 0, that is, the closer the objective function T of the compressed sensing algorithm is to 0, the higher the reconstruction quality of the initial reconstructed image.

[0070] (2) Deep learning denoising model:

[0071] The deep learning denoising model of the embodiment of the present invention is a deep learning model for denoising magnetic resonance imaging images and outputting high-definition imaging images, that is, the deep learning denoising model is used to perform denoising processing on the input initial reconstruction image and output the corresponding final reconstruction image. The model structure of the deep learning denoising model can be implemented based on the model structure of multiple types of deep learning models, including at least the model structure implementation of the U-Net model and the model structure implementation of the GAN model; and the model mechanism of the U-Net model and the GAN model can be understood through public technical documents such as "U-Net: Convolutional Networks for Biomedical Image Segmentation" and "A Style-Based Generator Architecture for Generative Adversarial Networks", which will not be further repeated here.

[0072] Step 2, training the coronary magnetic resonance imaging reconstruction model;

[0073] Specifically, it includes: step 21, collecting a plurality of high-definition whole-heart three-dimensional coronary magnetic resonance imaging images and their corresponding scanning sequence data to form corresponding first and second data sets;

[0074] Wherein, the first data set includes a plurality of first data records; the first data record includes first training K-space data and a first label map; the second data set includes a plurality of second data records; the second data record includes a first training map and a second label map;

[0075] Specifically, it includes: step 211, taking each acquired high-definition whole-heart three-dimensional coronary magnetic resonance imaging image as a corresponding first original image;

[0076] Here, when performing high-definition whole-heart three-dimensional coronary magnetic resonance imaging image acquisition, the embodiment of the present invention collects a large number of whole-heart three-dimensional coronary magnetic resonance imaging images in advance, and evaluates the clarity level of each collected whole-heart three-dimensional coronary magnetic resonance imaging image by manual evaluation or machine evaluation, and pre-sets a high-level threshold, and collects the whole-heart three-dimensional coronary magnetic resonance imaging images whose assessed clarity level exceeds the high-level threshold as high-definition whole-heart three-dimensional coronary magnetic resonance imaging images; and then uses each collected high-definition whole-heart three-dimensional coronary magnetic resonance imaging image as a corresponding original acquisition data, i.e., a first original image;

[0077] Step 212, performing K-space data sampling on the scan sequence data corresponding to each first original image in a preset undersampling manner to obtain corresponding first training K-space data;

[0078] Here, each first original image has a corresponding imaging scanning sequence, and the scanning sequence data is all the data of the sequence; under normal circumstances, the scanning sequence data can be fully sampled to obtain the fully sampled K space data corresponding to the first original image, but the full sampling time is generally relatively long; the undersampling method preset in the embodiment of the present invention is an undersampling method that matches the compressed sensing algorithm, and the undersampling method generally includes a one-dimensional random variable density sampling method, a two-dimensional random variable density sampling method, a spiral sampling method, and a radial sampling method. By sampling the scanning sequence data in any undersampling method, a corresponding undersampled K space data can be obtained, and the data space size of the undersampled K space data is consistent with that of the fully sampled K space data, but there is a large amount of unsampled 0-value data in the undersampled K space data, that is, sampling the scanning sequence data in the undersampling method can achieve the purpose of doubling the filling time of the sequence-K space; the first training K space data obtained in the current step is the undersampled K space data corresponding to the first original image;

[0079] Step 213, each first original image is used as a corresponding first label image; each first training K-space data and the corresponding first label image form a corresponding first data record; and all the obtained first data records form a corresponding first data set;

[0080] The first data set obtained here will be used to train the compressed sensing algorithm model in the subsequent steps;

[0081] Step 214, filtering and denoising each first original image to obtain a corresponding second label image;

[0082] Here, the purpose of filtering and denoising each first original image in the embodiment of the present invention is to further improve the clarity of the magnetic resonance imaging image; it should be noted that the filtering and denoising scheme selected in the embodiment of the present invention here can be a filtering and denoising scheme for improving the clarity of the entire image, or a filtering and denoising scheme for improving the clarity of a local image area; for the filtering and denoising scheme for improving the clarity of the entire image, a conventional image filtering and denoising scheme can be used; and the filtering and denoising scheme for improving the clarity of a local image area requires special design, because the magnetic resonance imaging image processed by the embodiment of the present invention is a full-heart three-dimensional coronary magnetic resonance imaging for coronary angiography. Resonance imaging image, and the characteristic of the magnetic resonance imaging image is that each pixel point corresponds not only to a three-dimensional image coordinate, but also to a magnetic resonance signal intensity value, and there will be typical differences in the signal intensity of the blood pool in the blood vessel cavity, the inner and outer walls of the blood vessel cavity, the plaque tissue on the inner wall of the blood vessel, and the adjacent tissue on the outer wall of the blood vessel. Therefore, when designing a filtering noise reduction scheme to improve the clarity of the local image area, it is necessary to design it with the goal of improving the signal intensity of the blood pool in the cavity of the coronary blood vessel, the inner and outer walls, and the plaque tissue, and improving the signal intensity difference between the blood pool in the cavity of the coronary blood vessel, the inner and outer walls, the plaque tissue, and the adjacent tissue outside the blood vessel;

[0083] Step 215, performing noise adding processing on each first original image based on a plurality of preset perturbation noise adding schemes to obtain a plurality of corresponding first noise added images;

[0084] Here, the multiple disturbance noise adding schemes of the embodiment of the present invention may be composed of multiple random Gaussian noise schemes and multiple empirical disturbance noise schemes; wherein each random Gaussian noise scheme corresponds to a random Gaussian noise feature; and the multiple empirical disturbance noise schemes summarize the multiple types of disturbance mechanisms commonly used in coronary magnetic resonance imaging in advance, and simulate a corresponding disturbance noise map based on each type of summarized disturbance mechanism, and each disturbance noise map is used as a corresponding empirical disturbance noise scheme;

[0085] The current step 215 specifically includes:

[0086] Step 2151, adding noise to each first original image based on multiple random Gaussian noise schemes to obtain multiple corresponding first noise-added images; specifically: adding Gaussian noise to the current first original image based on Gaussian noise features corresponding to each random Gaussian noise scheme to obtain a corresponding first noise-added image;

[0087] Step 2152, adding noise to each first original image based on multiple empirical perturbation noise schemes to obtain multiple corresponding first noise-added images; specifically, superimposing the perturbation noise images corresponding to each empirical perturbation noise scheme with the current first original image to obtain a corresponding first noise-added image;

[0088] It can be seen that, through the multiple perturbation noise adding schemes of the current step 215, multiple first noise added images can be derived from one first original image;

[0089] Step 216, each first noise added image is used as a corresponding first training image; each first training image and the corresponding second label image form a corresponding second data record; and all the obtained second data records form a corresponding second data set;

[0090] The second dataset obtained here will be used to train the deep learning denoising model in the subsequent steps;

[0091] Step 22, training the compressed sensing algorithm model based on the first data set;

[0092] Specifically, it includes: step 221, extracting the first first data record of the first data set as the corresponding current data record;

[0093] Step 222, extracting corresponding first training K-space data and first label map from the current data record;

[0094] Step 223, inputting the first training K-space data into the compressed sensing algorithm model to calculate and obtain a corresponding first training reconstruction image;

[0095] Step 224, bringing the first training K-space data, the first training reconstruction image and the first label image into a preset first training objective function O1 to calculate and obtain a corresponding first function value;

[0096] in,

[0097]

[0098] Y tr1 is the first training K-space data,

[0099] X tr1 Reconstruct the first training map,

[0100] X tag1 is the first label image,

[0101] P is the K-space undersampling matrix,

[0102] F is the Fourier transform matrix,

[0103] ζ and λ are two preset regularization factors.

[0104] ‖‖1 and ‖‖2 are the corresponding L1 and L2 regularization terms respectively;

[0105] Step 225, identifying whether the first function value satisfies a preset first convergence function value range; if not, optimizing the model parameter set of the compressed sensing algorithm model in a direction of minimizing the first training objective function L1 based on a preset model optimization algorithm, and returning to step 223 to continue training at the end of this round of optimization; if satisfied, identifying whether the current data record is the last first data record of the first data set, if not, extracting the next first data record of the first data set as a new current data record and returning to step 222 to continue training, if yes, stopping training;

[0106] Among them, the model optimization algorithms include SDG optimization algorithm and Adam optimization algorithm;

[0107] Here, the first convergence function value range is a preselected function value range;

[0108] Step 23, training the deep learning denoising model based on the second data set;

[0109] Step 231, extracting the first second data record of the second data set as the corresponding current data record;

[0110] Step 232, extracting the corresponding first training graph and second label graph from the current data record;

[0111] Step 233, inputting the first training image into the deep learning denoising model to calculate and obtain a corresponding second training reconstruction image;

[0112] Step 234, bringing the second training reconstruction image and the second label image into a preset first loss function L1 to calculate and obtain a corresponding first loss value;

[0113] Here, there is a corresponding relationship between the first loss function L1 and the model structure of the deep learning denoising model; if the model structure of the deep learning denoising model is implemented based on the U-Net model, the corresponding first loss function L1 uses the L2 loss function by default; if the model structure of the deep learning denoising model is implemented based on the GAN model, the corresponding first loss function L1 uses the cross entropy loss function by default;

[0114] Step 235, identifying whether the first loss value satisfies a preset first loss value range; if not, optimizing the model parameter set of the deep learning denoising model in a direction of minimizing the first loss function L1 based on a preset model optimization algorithm, and returning to step 233 to continue training at the end of this round of optimization; if satisfied, identifying whether the current data record is the last second data record of the second data set, if not, extracting the next second data record of the second data set as a new current data record and returning to step 232 to continue training, if yes, stopping training;

[0115] Among them, the model optimization algorithms include SDG optimization algorithm and Adam optimization algorithm.

[0116] Here, the first loss value range is a preset loss value range.

[0117] Step 3, after the model training is completed, a whole-heart three-dimensional coronary artery imaging scan is performed on the first subject using a magnetic resonance scanning device according to the configured scanning sequence type, and during the scanning process, K-space data sampling is performed on the obtained scanning sequence data based on a preset undersampling method to obtain corresponding first undersampling K-space data; the first undersampling K-space data is input into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image; and coronary stenosis feature recognition is performed based on the first imaging image to generate a corresponding first recognition report;

[0118] Specifically, the method comprises: step 31, after the model training is completed, performing a full-heart three-dimensional coronary artery imaging scan on the first subject using a magnetic resonance scanning device according to a configured scanning sequence type, and performing K-space data sampling on the obtained scanning sequence data based on a preset undersampling method during the scanning process to obtain corresponding first undersampling K-space data;

[0119] Here, the field strength level of the magnetic resonance scanning device includes at least 1.5T and 3.0T; when the field strength level is 1.5T, the corresponding scanning sequence type is configured as a steady-state free precession SSFP sequence by default; when the field strength level is 3.0T, the corresponding scanning sequence type is configured as a fast gradient echo FGRE sequence by default;

[0120] Step 32, inputting the first under-sampled K-space data into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image;

[0121] Specifically, the first under-sampled K-space data is input into the coronary magnetic resonance imaging reconstruction model, and the compressed sensing algorithm model performs magnetic resonance imaging image reconstruction processing according to the input first under-sampled K-space data and outputs the corresponding current initial reconstruction image; and the deep learning denoising model performs magnetic resonance imaging image denoising processing on the current initial reconstruction image and outputs the corresponding current final reconstruction image; and the obtained current final reconstruction image is output as the corresponding first imaging image;

[0122] Step 33: performing coronary artery stenosis feature recognition based on the first imaging image to generate a corresponding first recognition report.

[0123] Here, in an embodiment of the present invention, when identifying the characteristics of coronary stenosis based on the first imaging image, the coronary stenosis trend and the plaque properties related to the coronary stenosis of the subject can be identified based on the first imaging image in an artificial manner, and the identification result can be used as a corresponding first identification report and output; it can also be based on another type of deep learning model to identify the stenotic blood vessel segment area of ​​the entire coronary artery area based on the first imaging image, and based on another type of deep learning model, further identify the plaque properties of each identified stenotic blood vessel segment, and the identification results of the two deep learning models form a corresponding first identification report and output it.

[0124] Figure 3 This is a module structure diagram of a processing device for a coronary magnetic resonance imaging reconstruction model provided in Embodiment 2 of the present invention. The device is a terminal device or server that implements the aforementioned method embodiment, or can be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a model building module 201, a model training module 202 and a model application module 203.

[0125] The model building module 201 is used to build a coronary magnetic resonance imaging reconstruction model.

[0126] The model training module 202 is used to perform model training on the coronary magnetic resonance imaging reconstruction model.

[0127] The model application module 203 is used to perform a whole-heart three-dimensional coronary artery imaging scan on the first subject using a magnetic resonance scanning device according to the configured scanning sequence type after the model training is completed, and perform K-space data sampling on the obtained scanning sequence data based on a preset undersampling method during the scanning process to obtain corresponding first under-sampled K-space data; and input the first under-sampled K-space data into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image; and perform coronary artery stenosis feature recognition based on the first imaging image to generate a corresponding first recognition report.

[0128] An embodiment of the present invention provides a processing device for a coronary magnetic resonance imaging reconstruction model, which can execute the method steps in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0129] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the model building module can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above-mentioned module is determined. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0130] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0131] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the above method embodiments are generated. The above computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) methods. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0132] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device may be a terminal device or a server for implementing the method of the aforementioned embodiment, or may be a terminal device or a server for implementing the method of the aforementioned embodiment connected to the aforementioned terminal device or server. Figure 4 As shown, the electronic device may include: a processor 301 (such as a CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiver 303. Various instructions may be stored in the memory 302 to complete various processing functions and implement the processing steps described in the aforementioned embodiment method. Preferably, the electronic device involved in the embodiment of the present invention also includes: a power supply 304, a system bus 305 and a communication port 306. The system bus 305 is used to realize the communication connection between components. The above-mentioned communication port 306 is used for connecting and communicating between the electronic device and other peripherals.

[0133] exist Figure 4The system bus 305 mentioned in the figure can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM), and may also include non-volatile memory (Non-Volatile Memory), such as at least one disk storage.

[0134] The above-mentioned processor can be a general-purpose processor, including a central processing unit CPU, a network processor (Network Processor, NP), a graphics processing unit (Graphics Processing Unit, GPU), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0135] It should be noted that an embodiment of the present invention further provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the method and processing process provided in the above embodiments.

[0136] An embodiment of the present invention further provides a chip for executing instructions, wherein the chip is used to execute the processing steps described in the aforementioned method embodiment.

[0137] The embodiments of the present invention provide a processing method, device, electronic device and computer-readable storage medium for a coronary magnetic resonance imaging reconstruction model. As can be seen from the above content, the embodiments of the present invention pre-construct a deep learning model for simulating the deduction process of the compressed sensing algorithm based on the convolutional neural network structure, namely the compressed sensing algorithm model, which can reconstruct the magnetic resonance imaging image according to an input undersampled K-space data and output the corresponding magnetic resonance imaging image; and pre-construct a deep learning model for denoising the magnetic resonance imaging image based on the U-Net model or GAN model structure, namely the deep learning denoising model, which can denoise an input magnetic resonance imaging image and output a high-definition imaging image; and pre-construct an end-to-end coronary magnetic resonance imaging reconstruction model composed of the compressed sensing algorithm model + the deep learning denoising model, through which the coronary magnetic resonance imaging reconstruction The model can obtain a high-definition magnetic resonance imaging image based on an input undersampled K-space data; and complete the overall model training of the coronary magnetic resonance imaging reconstruction model by separately training the compressed sensing algorithm model and the deep learning noise reduction model; and after the overall model training is completed, the coronary magnetic resonance imaging reconstruction model is introduced into the daily coronary magnetic resonance angiography scanning process, and in the scanning process, the sequence-K space filling time is shortened based on the undersampling method, the K-space-image conversion time is shortened based on the coronary magnetic resonance imaging reconstruction model, and the clarity of the output coronary magnetic resonance imaging is improved, and the coronary stenosis feature recognition is performed based on the coronary magnetic resonance imaging output by the coronary magnetic resonance imaging reconstruction model. Through the embodiment of the present invention, not only can the coronary magnetic resonance imaging time be shortened and the imaging clarity be improved, but also the accuracy of coronary stenosis feature recognition can be improved based on the high-definition coronary magnetic resonance imaging image.

[0138] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0139] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0140] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for processing a coronary magnetic resonance imaging reconstruction model, characterized in that: The method comprises: Construct a coronary magnetic resonance imaging reconstruction model; Performing model training on the coronary magnetic resonance imaging reconstruction model; After the model training is completed, a whole-heart three-dimensional coronary artery imaging scan is performed on the first subject using a magnetic resonance scanning device according to the configured scanning sequence type, and during the scanning process, K-space data sampling is performed on the obtained scanning sequence data based on a preset undersampling method to obtain corresponding first undersampling K-space data; and the first undersampling K-space data is input into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image; and coronary stenosis feature recognition is performed based on the first imaging image to generate a corresponding first recognition report; Wherein, the coronary magnetic resonance imaging reconstruction model includes a compressed sensing algorithm model and a deep learning noise reduction model; The compressed sensing algorithm objective function T used by the compressed sensing algorithm model is: ； P is the K-space undersampling matrix corresponding to the undersampling method, F is the Fourier transform matrix, X is the initial reconstruction image, M is the model mapping function of the compressed sensing algorithm model, Y is the undersampling K-space data, θ is the model parameter set of the compressed sensing algorithm model, λ is the preset regularization factor, They are the corresponding L1 and L2 regularization terms respectively; The model training of the coronary magnetic resonance imaging reconstruction model specifically includes: collecting a plurality of high-definition whole-heart three-dimensional coronary magnetic resonance imaging images and their corresponding scanning sequence data to form corresponding first and second data sets; and training the compressed sensing algorithm model based on the first data set; and training the deep learning noise reduction model based on the second data set; The first training objective function O1 used when training the deep learning denoising model based on the second data set is: , Y tr1 is the first training K-space data, X tr1 For the first training reconstruction graph, X tag1 is the first label image, P is the K-space undersampling matrix, F is the Fourier transform matrix, ζ and λ are two preset regularization factors.

2. The method for processing the coronary magnetic resonance imaging reconstruction model according to claim 1, characterized in that: The input end of the compressed sensing algorithm model is connected to the input end of the coronary magnetic resonance imaging reconstruction model, and the output end is connected to the input end of the deep learning denoising model; the output end of the deep learning denoising model is connected to the output end of the coronary magnetic resonance imaging reconstruction model; The compressed sensing algorithm model is implemented based on a convolutional neural network; the compressed sensing algorithm model is used to perform magnetic resonance imaging image reconstruction processing according to the input undersampled K-space data and output the corresponding initial reconstruction image, specifically, the undersampled K-space data is brought into the compressed sensing algorithm objective function T, and the optimal solution of the initial reconstruction image is calculated in a direction that makes the compressed sensing algorithm objective function T reach a minimum value; The deep learning denoising model is implemented based on a model structure of a class of deep learning models, and the deep learning model includes at least a U-Net model and a GAN model; the deep learning denoising model is used to perform magnetic resonance imaging denoising processing on the initial reconstructed image and output the corresponding final reconstructed image.

3. The method for processing the coronary magnetic resonance imaging reconstruction model according to claim 2, characterized in that: The model training of the coronary magnetic resonance imaging reconstruction model specifically includes: The first data set includes a plurality of first data records; the first data record includes the first training K-space data and the first label map; the second data set includes a plurality of second data records; the second data record includes a first training map and a second label map.

4. The method for processing the coronary magnetic resonance imaging reconstruction model according to claim 3, characterized in that: The acquisition of a plurality of high-definition whole-heart three-dimensional coronary magnetic resonance imaging images and their corresponding scanning sequence data to form corresponding first and second data sets specifically includes: Using each of the acquired high-definition whole-heart three-dimensional coronary magnetic resonance imaging images as a corresponding first original image; and performing K-space data sampling on the scanning sequence data corresponding to each of the first original images in the undersampling manner to obtain the corresponding first training K-space data; and using each of the first original images as a corresponding first label image; and forming a corresponding first data record from each of the first training K-space data and the corresponding first label image; and forming a corresponding first data set from all the obtained first data records; And filter and denoise each of the first original images to obtain the corresponding second label image; and add noise to each of the first original images based on a plurality of preset disturbance noise adding schemes to obtain a plurality of corresponding first noise-added images; and use each of the first noise-added images as a corresponding first training image; and form a corresponding second data record from each of the first training images and the corresponding second label image; and form a corresponding second data set from all the obtained second data records.

5. The method for processing the coronary magnetic resonance imaging reconstruction model according to claim 3, characterized in that: The training of the compressed sensing algorithm model based on the first data set specifically includes: Step 51, extracting the first first data record of the first data set as the corresponding current data record; Step 52, extracting the corresponding first training K-space data and the first label map from the current data record; Step 53, inputting the first training K-space data into the compressed sensing algorithm model to calculate and obtain the corresponding first training reconstruction image; Step 54, bringing the first training K-space data, the first training reconstruction image and the first label image into the preset first training objective function O1 to calculate and obtain a corresponding first function value; Step 55, identifying whether the first function value satisfies a preset first convergence function value range; if not, performing a round of optimization on the model parameter set of the compressed sensing algorithm model in the direction of minimizing the first training objective function L1 based on a preset model optimization algorithm, and returning to step 53 to continue training at the end of this round of optimization; if satisfied, identifying whether the current data record is the last of the first data set, if not, extracting the next first data record of the first data set as the new current data record and returning to step 52 to continue training, if yes, stopping training; the model optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm.

6. The method for processing the coronary magnetic resonance imaging reconstruction model according to claim 3, characterized in that: The training of the deep learning denoising model based on the second data set specifically includes: Step 61, extracting the first second data record of the second data set as the corresponding current data record; Step 62, extracting the corresponding first training graph and second label graph from the current data record; Step 63, inputting the first training image into the deep learning denoising model to calculate and obtain a corresponding second training reconstruction image; Step 64, bringing the second training reconstruction image and the second label image into a preset first loss function L1 to calculate a corresponding first loss value; when the model structure of the deep learning denoising model is implemented based on a U-Net model, the corresponding first loss function L1 uses an L2 loss function by default; when the model structure of the deep learning denoising model is implemented based on a GAN model, the corresponding first loss function L1 uses a cross entropy loss function by default; Step 65, whether the first loss value satisfies a preset first loss value range is identified; if not, a round of optimization is performed on the model parameter set of the deep learning denoising model in the direction of minimizing the first loss function L1 based on a preset model optimization algorithm, and at the end of this round of optimization, the method returns to step 63 to continue training; if satisfied, whether the current data record is the last second data record of the second data set is identified, if not, the next second data record of the second data set is extracted as the new current data record and the method returns to step 62 to continue training, if so, the training is stopped; the model optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm.

7. The method for processing the coronary magnetic resonance imaging reconstruction model according to claim 2, characterized in that: The step of inputting the first under-sampled K-space data into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image specifically includes: The first under-sampled K-space data is input into the coronary magnetic resonance imaging reconstruction model, and the compressed sensing algorithm model performs magnetic resonance imaging image reconstruction processing based on the input first under-sampled K-space data and outputs the corresponding current initial reconstruction image; and the deep learning denoising model performs magnetic resonance imaging image denoising processing on the current initial reconstruction image and outputs the corresponding current final reconstruction image; and the obtained current final reconstruction image is output as the corresponding first imaging image.

8. The method for processing a coronary magnetic resonance imaging reconstruction model according to claim 2, characterized in that: The field strength levels of the magnetic resonance scanning device include 1.5T and 3.0T; when the field strength level is 1.5T, the corresponding scanning sequence type is configured by default as a steady-state free precession SSFP sequence; when the field strength level is 3.0T, the corresponding scanning sequence type is configured by default as a fast gradient echo FGRE sequence.

9. A device for executing the processing method of coronary magnetic resonance imaging reconstruction model according to any one of claims 1 to 8, characterized in that: The device comprises: a model building module, a model training module and a model application module; The model building module is used to build a coronary magnetic resonance imaging reconstruction model; The model training module is used to perform model training on the coronary magnetic resonance imaging reconstruction model; The model application module is used to perform a whole-heart three-dimensional coronary artery imaging scan on a first subject using a magnetic resonance scanning device according to a configured scanning sequence type after model training is completed, and perform K-space data sampling on the obtained scanning sequence data based on a preset undersampling method during the scanning process to obtain corresponding first under-sampled K-space data; and input the first under-sampled K-space data into the coronary magnetic resonance imaging reconstruction model to perform magnetic resonance imaging image reconstruction and noise reduction processing to obtain a corresponding first imaging image; and perform coronary artery stenosis feature recognition based on the first imaging image to generate a corresponding first recognition report.

10. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 8; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a computer, the computer is enabled to execute the method according to any one of claims 1 to 8.

Images